Datasets:

smangrul

/

hug_stack

like 3

# Multi Subject Dreambooth for Inpainting Models Please note that this project is not actively maintained. However, you can open an issue and tag @gzguevara. [DreamBooth](https://arxiv.org/abs/2208.12242) is a method to personalize text2image models like stable diffusion given just a few(3~5) images of a subject. This project consists of **two parts**. Training Stable Diffusion for inpainting requieres prompt-image-mask pairs. The Unet of inpainiting models have 5 additional input channels (4 for the encoded masked-image and 1 for the mask itself). **The first part**, the `multi_inpaint_dataset.ipynb` notebook, demonstrates how make a 🤗 dataset of prompt-image-mask pairs. You can, however, skip the first part and move straight to the second part with the example datasets in this project. ([cat toy dataset masked](https://huggingface.co/datasets/gzguevara/cat_toy_masked), [mr. potato head dataset masked](https://huggingface.co/datasets/gzguevara/mr_potato_head_masked)) **The second part**, the `train_multi_subject_inpainting.py` training script, demonstrates how to implement a training procedure for one or more subjects and adapt it for stable diffusion for inpainting. ## 1. Data Collection: Make Prompt-Image-Mask Pairs Earlier training scripts have provided approaches like random masking for the training images. This project provides a notebook for more precise mask setting. The notebook can be found here: [](https://colab.research.google.com/drive/1JNEASI_B7pLW1srxhgln6nM0HoGAQT32?usp=sharing) The `multi_inpaint_dataset.ipynb` notebook, takes training & validation images, on which the user draws masks and provides prompts to make a prompt-image-mask pairs. This ensures that during training, the loss is computed on the area masking the object of interest, rather than on random areas. Moreover, the `multi_inpaint_dataset.ipynb` notebook allows you to build a validation dataset with corresponding masks for monitoring the training process. Example below:  You can build multiple datasets for every subject and upload them to the 🤗 hub. Later, when launching the training script you can indicate the paths of the datasets, on which you would like to finetune Stable Diffusion for inpaining. ## 2. Train Multi Subject Dreambooth for Inpainting ### 2.1. Setting The Training Configuration Before launching the training script, make sure to select the inpainting the target model, the output directory and the 🤗 datasets. ```bash export MODEL_NAME="runwayml/stable-diffusion-inpainting" export OUTPUT_DIR="path-to-save-model" export DATASET_1="gzguevara/mr_potato_head_masked" export DATASET_2="gzguevara/cat_toy_masked" ... # Further paths to 🤗 datasets ``` ### 2.2. Launching The Training Script ```bash accelerate launch train_multi_subject_dreambooth_inpaint.py \ --pretrained_model_name_or_path=$MODEL_NAME \ --instance_data_dir $DATASET_1 $DATASET_2 \ --output_dir=$OUTPUT_DIR \ --resolution=512 \ --train_batch_size=1 \ --gradient_accumulation_steps=2 \ --learning_rate=3e-6 \ --max_train_steps=500 \ --report_to_wandb ``` ### 2.3. Fine-tune text encoder with the UNet. The script also allows to fine-tune the `text_encoder` along with the `unet`. It's been observed experimentally that fine-tuning `text_encoder` gives much better results especially on faces. Pass the `--train_text_encoder` argument to the script to enable training `text_encoder`. ___Note: Training text encoder requires more memory, with this option the training won't fit on 16GB GPU. It needs at least 24GB VRAM.___ ```bash accelerate launch train_multi_subject_dreambooth_inpaint.py \ --pretrained_model_name_or_path=$MODEL_NAME \ --instance_data_dir $DATASET_1 $DATASET_2 \ --output_dir=$OUTPUT_DIR \ --resolution=512 \ --train_batch_size=1 \ --gradient_accumulation_steps=2 \ --learning_rate=2e-6 \ --max_train_steps=500 \ --report_to_wandb \ --train_text_encoder ``` ## 3. Results A [](https://wandb.ai/gzguevara/uncategorized/reports/Multi-Subject-Dreambooth-for-Inpainting--Vmlldzo2MzY5NDQ4?accessToken=y0nya2d7baguhbryxaikbfr1203amvn1jsmyl07vk122mrs7tnph037u1nqgse8t) is provided showing the training progress by every 50 steps. Note, the reported weights & baises run was performed on a A100 GPU with the following stetting: ```bash accelerate launch train_multi_subject_dreambooth_inpaint.py \ --pretrained_model_name_or_path=$MODEL_NAME \ --instance_data_dir $DATASET_1 $DATASET_2 \ --output_dir=$OUTPUT_DIR \ --resolution=512 \ --train_batch_size=10 \ --gradient_accumulation_steps=1 \ --learning_rate=1e-6 \ --max_train_steps=500 \ --report_to_wandb \ --train_text_encoder ``` Here you can see the target objects on my desk and next to my plant: 

diffusers/examples/research_projects/multi_subject_dreambooth_inpainting/README.md/0

## Training examples Creating a training image set is [described in a different document](https://huggingface.co/docs/datasets/image_process#image-datasets). ### Installing the dependencies Before running the scripts, make sure to install the library's training dependencies: **Important** To make sure you can successfully run the latest versions of the example scripts, we highly recommend **installing from source** and keeping the install up to date as we update the example scripts frequently and install some example-specific requirements. To do this, execute the following steps in a new virtual environment: ```bash git clone https://github.com/huggingface/diffusers cd diffusers pip install . ``` Then cd in the example folder and run ```bash pip install -r requirements.txt ``` And initialize an [🤗Accelerate](https://github.com/huggingface/accelerate/) environment with: ```bash accelerate config ``` #### Use ONNXRuntime to accelerate training In order to leverage onnxruntime to accelerate training, please use train_unconditional_ort.py The command to train a DDPM UNet model on the Oxford Flowers dataset with onnxruntime: ```bash accelerate launch train_unconditional.py \ --dataset_name="huggan/flowers-102-categories" \ --resolution=64 --center_crop --random_flip \ --output_dir="ddpm-ema-flowers-64" \ --use_ema \ --train_batch_size=16 \ --num_epochs=1 \ --gradient_accumulation_steps=1 \ --learning_rate=1e-4 \ --lr_warmup_steps=500 \ --mixed_precision=fp16 ``` Please contact Prathik Rao (prathikr), Sunghoon Choi (hanbitmyths), Ashwini Khade (askhade), or Peng Wang (pengwa) on github with any questions.

diffusers/examples/research_projects/onnxruntime/unconditional_image_generation/README.md/0

# coding=utf-8 # Copyright 2023 HuggingFace Inc. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import logging import os import sys import tempfile sys.path.append("..") from test_examples_utils import ExamplesTestsAccelerate, run_command # noqa: E402 logging.basicConfig(level=logging.DEBUG) logger = logging.getLogger() stream_handler = logging.StreamHandler(sys.stdout) logger.addHandler(stream_handler) class T2IAdapter(ExamplesTestsAccelerate): def test_t2i_adapter_sdxl(self): with tempfile.TemporaryDirectory() as tmpdir: test_args = f""" examples/t2i_adapter/train_t2i_adapter_sdxl.py --pretrained_model_name_or_path=hf-internal-testing/tiny-stable-diffusion-xl-pipe --adapter_model_name_or_path=hf-internal-testing/tiny-adapter --dataset_name=hf-internal-testing/fill10 --output_dir={tmpdir} --resolution=64 --train_batch_size=1 --gradient_accumulation_steps=1 --max_train_steps=9 --checkpointing_steps=2 """.split() run_command(self._launch_args + test_args) self.assertTrue(os.path.isfile(os.path.join(tmpdir, "diffusion_pytorch_model.safetensors")))

diffusers/examples/t2i_adapter/test_t2i_adapter.py/0

## Textual Inversion fine-tuning example for SDXL ``` export MODEL_NAME="stabilityai/stable-diffusion-xl-base-1.0" export DATA_DIR="./cat" accelerate launch textual_inversion_sdxl.py \ --pretrained_model_name_or_path=$MODEL_NAME \ --train_data_dir=$DATA_DIR \ --learnable_property="object" \ --placeholder_token="<cat-toy>" \ --initializer_token="toy" \ --mixed_precision="bf16" \ --resolution=768 \ --train_batch_size=1 \ --gradient_accumulation_steps=4 \ --max_train_steps=500 \ --learning_rate=5.0e-04 \ --scale_lr \ --lr_scheduler="constant" \ --lr_warmup_steps=0 \ --save_as_full_pipeline \ --output_dir="./textual_inversion_cat_sdxl" ``` For now, only training of the first text encoder is supported.

diffusers/examples/textual_inversion/README_sdxl.md/0

# Script for converting a HF Diffusers saved pipeline to a Stable Diffusion checkpoint. # *Only* converts the UNet, VAE, and Text Encoder. # Does not convert optimizer state or any other thing. import argparse import os.path as osp import re import torch from safetensors.torch import load_file, save_file # =================# # UNet Conversion # # =================# unet_conversion_map = [ # (stable-diffusion, HF Diffusers) ("time_embed.0.weight", "time_embedding.linear_1.weight"), ("time_embed.0.bias", "time_embedding.linear_1.bias"), ("time_embed.2.weight", "time_embedding.linear_2.weight"), ("time_embed.2.bias", "time_embedding.linear_2.bias"), ("input_blocks.0.0.weight", "conv_in.weight"), ("input_blocks.0.0.bias", "conv_in.bias"), ("out.0.weight", "conv_norm_out.weight"), ("out.0.bias", "conv_norm_out.bias"), ("out.2.weight", "conv_out.weight"), ("out.2.bias", "conv_out.bias"), # the following are for sdxl ("label_emb.0.0.weight", "add_embedding.linear_1.weight"), ("label_emb.0.0.bias", "add_embedding.linear_1.bias"), ("label_emb.0.2.weight", "add_embedding.linear_2.weight"), ("label_emb.0.2.bias", "add_embedding.linear_2.bias"), ] unet_conversion_map_resnet = [ # (stable-diffusion, HF Diffusers) ("in_layers.0", "norm1"), ("in_layers.2", "conv1"), ("out_layers.0", "norm2"), ("out_layers.3", "conv2"), ("emb_layers.1", "time_emb_proj"), ("skip_connection", "conv_shortcut"), ] unet_conversion_map_layer = [] # hardcoded number of downblocks and resnets/attentions... # would need smarter logic for other networks. for i in range(3): # loop over downblocks/upblocks for j in range(2): # loop over resnets/attentions for downblocks hf_down_res_prefix = f"down_blocks.{i}.resnets.{j}." sd_down_res_prefix = f"input_blocks.{3*i + j + 1}.0." unet_conversion_map_layer.append((sd_down_res_prefix, hf_down_res_prefix)) if i > 0: hf_down_atn_prefix = f"down_blocks.{i}.attentions.{j}." sd_down_atn_prefix = f"input_blocks.{3*i + j + 1}.1." unet_conversion_map_layer.append((sd_down_atn_prefix, hf_down_atn_prefix)) for j in range(4): # loop over resnets/attentions for upblocks hf_up_res_prefix = f"up_blocks.{i}.resnets.{j}." sd_up_res_prefix = f"output_blocks.{3*i + j}.0." unet_conversion_map_layer.append((sd_up_res_prefix, hf_up_res_prefix)) if i < 2: # no attention layers in up_blocks.0 hf_up_atn_prefix = f"up_blocks.{i}.attentions.{j}." sd_up_atn_prefix = f"output_blocks.{3 * i + j}.1." unet_conversion_map_layer.append((sd_up_atn_prefix, hf_up_atn_prefix)) if i < 3: # no downsample in down_blocks.3 hf_downsample_prefix = f"down_blocks.{i}.downsamplers.0.conv." sd_downsample_prefix = f"input_blocks.{3*(i+1)}.0.op." unet_conversion_map_layer.append((sd_downsample_prefix, hf_downsample_prefix)) # no upsample in up_blocks.3 hf_upsample_prefix = f"up_blocks.{i}.upsamplers.0." sd_upsample_prefix = f"output_blocks.{3*i + 2}.{1 if i == 0 else 2}." unet_conversion_map_layer.append((sd_upsample_prefix, hf_upsample_prefix)) unet_conversion_map_layer.append(("output_blocks.2.2.conv.", "output_blocks.2.1.conv.")) hf_mid_atn_prefix = "mid_block.attentions.0." sd_mid_atn_prefix = "middle_block.1." unet_conversion_map_layer.append((sd_mid_atn_prefix, hf_mid_atn_prefix)) for j in range(2): hf_mid_res_prefix = f"mid_block.resnets.{j}." sd_mid_res_prefix = f"middle_block.{2*j}." unet_conversion_map_layer.append((sd_mid_res_prefix, hf_mid_res_prefix)) def convert_unet_state_dict(unet_state_dict): # buyer beware: this is a *brittle* function, # and correct output requires that all of these pieces interact in # the exact order in which I have arranged them. mapping = {k: k for k in unet_state_dict.keys()} for sd_name, hf_name in unet_conversion_map: mapping[hf_name] = sd_name for k, v in mapping.items(): if "resnets" in k: for sd_part, hf_part in unet_conversion_map_resnet: v = v.replace(hf_part, sd_part) mapping[k] = v for k, v in mapping.items(): for sd_part, hf_part in unet_conversion_map_layer: v = v.replace(hf_part, sd_part) mapping[k] = v new_state_dict = {sd_name: unet_state_dict[hf_name] for hf_name, sd_name in mapping.items()} return new_state_dict # ================# # VAE Conversion # # ================# vae_conversion_map = [ # (stable-diffusion, HF Diffusers) ("nin_shortcut", "conv_shortcut"), ("norm_out", "conv_norm_out"), ("mid.attn_1.", "mid_block.attentions.0."), ] for i in range(4): # down_blocks have two resnets for j in range(2): hf_down_prefix = f"encoder.down_blocks.{i}.resnets.{j}." sd_down_prefix = f"encoder.down.{i}.block.{j}." vae_conversion_map.append((sd_down_prefix, hf_down_prefix)) if i < 3: hf_downsample_prefix = f"down_blocks.{i}.downsamplers.0." sd_downsample_prefix = f"down.{i}.downsample." vae_conversion_map.append((sd_downsample_prefix, hf_downsample_prefix)) hf_upsample_prefix = f"up_blocks.{i}.upsamplers.0." sd_upsample_prefix = f"up.{3-i}.upsample." vae_conversion_map.append((sd_upsample_prefix, hf_upsample_prefix)) # up_blocks have three resnets # also, up blocks in hf are numbered in reverse from sd for j in range(3): hf_up_prefix = f"decoder.up_blocks.{i}.resnets.{j}." sd_up_prefix = f"decoder.up.{3-i}.block.{j}." vae_conversion_map.append((sd_up_prefix, hf_up_prefix)) # this part accounts for mid blocks in both the encoder and the decoder for i in range(2): hf_mid_res_prefix = f"mid_block.resnets.{i}." sd_mid_res_prefix = f"mid.block_{i+1}." vae_conversion_map.append((sd_mid_res_prefix, hf_mid_res_prefix)) vae_conversion_map_attn = [ # (stable-diffusion, HF Diffusers) ("norm.", "group_norm."), # the following are for SDXL ("q.", "to_q."), ("k.", "to_k."), ("v.", "to_v."), ("proj_out.", "to_out.0."), ] def reshape_weight_for_sd(w): # convert HF linear weights to SD conv2d weights return w.reshape(*w.shape, 1, 1) def convert_vae_state_dict(vae_state_dict): mapping = {k: k for k in vae_state_dict.keys()} for k, v in mapping.items(): for sd_part, hf_part in vae_conversion_map: v = v.replace(hf_part, sd_part) mapping[k] = v for k, v in mapping.items(): if "attentions" in k: for sd_part, hf_part in vae_conversion_map_attn: v = v.replace(hf_part, sd_part) mapping[k] = v new_state_dict = {v: vae_state_dict[k] for k, v in mapping.items()} weights_to_convert = ["q", "k", "v", "proj_out"] for k, v in new_state_dict.items(): for weight_name in weights_to_convert: if f"mid.attn_1.{weight_name}.weight" in k: print(f"Reshaping {k} for SD format") new_state_dict[k] = reshape_weight_for_sd(v) return new_state_dict # =========================# # Text Encoder Conversion # # =========================# textenc_conversion_lst = [ # (stable-diffusion, HF Diffusers) ("transformer.resblocks.", "text_model.encoder.layers."), ("ln_1", "layer_norm1"), ("ln_2", "layer_norm2"), (".c_fc.", ".fc1."), (".c_proj.", ".fc2."), (".attn", ".self_attn"), ("ln_final.", "text_model.final_layer_norm."), ("token_embedding.weight", "text_model.embeddings.token_embedding.weight"), ("positional_embedding", "text_model.embeddings.position_embedding.weight"), ] protected = {re.escape(x[1]): x[0] for x in textenc_conversion_lst} textenc_pattern = re.compile("|".join(protected.keys())) # Ordering is from https://github.com/pytorch/pytorch/blob/master/test/cpp/api/modules.cpp code2idx = {"q": 0, "k": 1, "v": 2} def convert_openclip_text_enc_state_dict(text_enc_dict): new_state_dict = {} capture_qkv_weight = {} capture_qkv_bias = {} for k, v in text_enc_dict.items(): if ( k.endswith(".self_attn.q_proj.weight") or k.endswith(".self_attn.k_proj.weight") or k.endswith(".self_attn.v_proj.weight") ): k_pre = k[: -len(".q_proj.weight")] k_code = k[-len("q_proj.weight")] if k_pre not in capture_qkv_weight: capture_qkv_weight[k_pre] = [None, None, None] capture_qkv_weight[k_pre][code2idx[k_code]] = v continue if ( k.endswith(".self_attn.q_proj.bias") or k.endswith(".self_attn.k_proj.bias") or k.endswith(".self_attn.v_proj.bias") ): k_pre = k[: -len(".q_proj.bias")] k_code = k[-len("q_proj.bias")] if k_pre not in capture_qkv_bias: capture_qkv_bias[k_pre] = [None, None, None] capture_qkv_bias[k_pre][code2idx[k_code]] = v continue relabelled_key = textenc_pattern.sub(lambda m: protected[re.escape(m.group(0))], k) new_state_dict[relabelled_key] = v for k_pre, tensors in capture_qkv_weight.items(): if None in tensors: raise Exception("CORRUPTED MODEL: one of the q-k-v values for the text encoder was missing") relabelled_key = textenc_pattern.sub(lambda m: protected[re.escape(m.group(0))], k_pre) new_state_dict[relabelled_key + ".in_proj_weight"] = torch.cat(tensors) for k_pre, tensors in capture_qkv_bias.items(): if None in tensors: raise Exception("CORRUPTED MODEL: one of the q-k-v values for the text encoder was missing") relabelled_key = textenc_pattern.sub(lambda m: protected[re.escape(m.group(0))], k_pre) new_state_dict[relabelled_key + ".in_proj_bias"] = torch.cat(tensors) return new_state_dict def convert_openai_text_enc_state_dict(text_enc_dict): return text_enc_dict if __name__ == "__main__": parser = argparse.ArgumentParser() parser.add_argument("--model_path", default=None, type=str, required=True, help="Path to the model to convert.") parser.add_argument("--checkpoint_path", default=None, type=str, required=True, help="Path to the output model.") parser.add_argument("--half", action="store_true", help="Save weights in half precision.") parser.add_argument( "--use_safetensors", action="store_true", help="Save weights use safetensors, default is ckpt." ) args = parser.parse_args() assert args.model_path is not None, "Must provide a model path!" assert args.checkpoint_path is not None, "Must provide a checkpoint path!" # Path for safetensors unet_path = osp.join(args.model_path, "unet", "diffusion_pytorch_model.safetensors") vae_path = osp.join(args.model_path, "vae", "diffusion_pytorch_model.safetensors") text_enc_path = osp.join(args.model_path, "text_encoder", "model.safetensors") text_enc_2_path = osp.join(args.model_path, "text_encoder_2", "model.safetensors") # Load models from safetensors if it exists, if it doesn't pytorch if osp.exists(unet_path): unet_state_dict = load_file(unet_path, device="cpu") else: unet_path = osp.join(args.model_path, "unet", "diffusion_pytorch_model.bin") unet_state_dict = torch.load(unet_path, map_location="cpu") if osp.exists(vae_path): vae_state_dict = load_file(vae_path, device="cpu") else: vae_path = osp.join(args.model_path, "vae", "diffusion_pytorch_model.bin") vae_state_dict = torch.load(vae_path, map_location="cpu") if osp.exists(text_enc_path): text_enc_dict = load_file(text_enc_path, device="cpu") else: text_enc_path = osp.join(args.model_path, "text_encoder", "pytorch_model.bin") text_enc_dict = torch.load(text_enc_path, map_location="cpu") if osp.exists(text_enc_2_path): text_enc_2_dict = load_file(text_enc_2_path, device="cpu") else: text_enc_2_path = osp.join(args.model_path, "text_encoder_2", "pytorch_model.bin") text_enc_2_dict = torch.load(text_enc_2_path, map_location="cpu") # Convert the UNet model unet_state_dict = convert_unet_state_dict(unet_state_dict) unet_state_dict = {"model.diffusion_model." + k: v for k, v in unet_state_dict.items()} # Convert the VAE model vae_state_dict = convert_vae_state_dict(vae_state_dict) vae_state_dict = {"first_stage_model." + k: v for k, v in vae_state_dict.items()} text_enc_dict = convert_openai_text_enc_state_dict(text_enc_dict) text_enc_dict = {"conditioner.embedders.0.transformer." + k: v for k, v in text_enc_dict.items()} text_enc_2_dict = convert_openclip_text_enc_state_dict(text_enc_2_dict) text_enc_2_dict = {"conditioner.embedders.1.model." + k: v for k, v in text_enc_2_dict.items()} # Put together new checkpoint state_dict = {**unet_state_dict, **vae_state_dict, **text_enc_dict, **text_enc_2_dict} if args.half: state_dict = {k: v.half() for k, v in state_dict.items()} if args.use_safetensors: save_file(state_dict, args.checkpoint_path) else: state_dict = {"state_dict": state_dict} torch.save(state_dict, args.checkpoint_path)

diffusers/scripts/convert_diffusers_to_original_sdxl.py/0

# coding=utf-8 # Copyright 2023 The HuggingFace Inc. team. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """ Conversion script for the AudioLDM2 checkpoints.""" import argparse import re from typing import List, Union import torch import yaml from transformers import ( AutoFeatureExtractor, AutoTokenizer, ClapConfig, ClapModel, GPT2Config, GPT2Model, SpeechT5HifiGan, SpeechT5HifiGanConfig, T5Config, T5EncoderModel, ) from diffusers import ( AudioLDM2Pipeline, AudioLDM2ProjectionModel, AudioLDM2UNet2DConditionModel, AutoencoderKL, DDIMScheduler, DPMSolverMultistepScheduler, EulerAncestralDiscreteScheduler, EulerDiscreteScheduler, HeunDiscreteScheduler, LMSDiscreteScheduler, PNDMScheduler, ) from diffusers.utils import is_safetensors_available from diffusers.utils.import_utils import BACKENDS_MAPPING # Copied from diffusers.pipelines.stable_diffusion.convert_from_ckpt.shave_segments def shave_segments(path, n_shave_prefix_segments=1): """ Removes segments. Positive values shave the first segments, negative shave the last segments. """ if n_shave_prefix_segments >= 0: return ".".join(path.split(".")[n_shave_prefix_segments:]) else: return ".".join(path.split(".")[:n_shave_prefix_segments]) # Copied from diffusers.pipelines.stable_diffusion.convert_from_ckpt.renew_resnet_paths def renew_resnet_paths(old_list, n_shave_prefix_segments=0): """ Updates paths inside resnets to the new naming scheme (local renaming) """ mapping = [] for old_item in old_list: new_item = old_item.replace("in_layers.0", "norm1") new_item = new_item.replace("in_layers.2", "conv1") new_item = new_item.replace("out_layers.0", "norm2") new_item = new_item.replace("out_layers.3", "conv2") new_item = new_item.replace("emb_layers.1", "time_emb_proj") new_item = new_item.replace("skip_connection", "conv_shortcut") new_item = shave_segments(new_item, n_shave_prefix_segments=n_shave_prefix_segments) mapping.append({"old": old_item, "new": new_item}) return mapping # Copied from diffusers.pipelines.stable_diffusion.convert_from_ckpt.renew_vae_resnet_paths def renew_vae_resnet_paths(old_list, n_shave_prefix_segments=0): """ Updates paths inside resnets to the new naming scheme (local renaming) """ mapping = [] for old_item in old_list: new_item = old_item new_item = new_item.replace("nin_shortcut", "conv_shortcut") new_item = shave_segments(new_item, n_shave_prefix_segments=n_shave_prefix_segments) mapping.append({"old": old_item, "new": new_item}) return mapping # Copied from diffusers.pipelines.stable_diffusion.convert_from_ckpt.renew_attention_paths def renew_attention_paths(old_list): """ Updates paths inside attentions to the new naming scheme (local renaming) """ mapping = [] for old_item in old_list: new_item = old_item # new_item = new_item.replace('norm.weight', 'group_norm.weight') # new_item = new_item.replace('norm.bias', 'group_norm.bias') # new_item = new_item.replace('proj_out.weight', 'proj_attn.weight') # new_item = new_item.replace('proj_out.bias', 'proj_attn.bias') # new_item = shave_segments(new_item, n_shave_prefix_segments=n_shave_prefix_segments) mapping.append({"old": old_item, "new": new_item}) return mapping def renew_vae_attention_paths(old_list, n_shave_prefix_segments=0): """ Updates paths inside attentions to the new naming scheme (local renaming) """ mapping = [] for old_item in old_list: new_item = old_item new_item = new_item.replace("norm.weight", "group_norm.weight") new_item = new_item.replace("norm.bias", "group_norm.bias") new_item = new_item.replace("q.weight", "to_q.weight") new_item = new_item.replace("q.bias", "to_q.bias") new_item = new_item.replace("k.weight", "to_k.weight") new_item = new_item.replace("k.bias", "to_k.bias") new_item = new_item.replace("v.weight", "to_v.weight") new_item = new_item.replace("v.bias", "to_v.bias") new_item = new_item.replace("proj_out.weight", "to_out.0.weight") new_item = new_item.replace("proj_out.bias", "to_out.0.bias") new_item = shave_segments(new_item, n_shave_prefix_segments=n_shave_prefix_segments) mapping.append({"old": old_item, "new": new_item}) return mapping def assign_to_checkpoint( paths, checkpoint, old_checkpoint, attention_paths_to_split=None, additional_replacements=None, config=None ): """ This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and takes into account additional replacements that may arise. Assigns the weights to the new checkpoint. """ assert isinstance(paths, list), "Paths should be a list of dicts containing 'old' and 'new' keys." # Splits the attention layers into three variables. if attention_paths_to_split is not None: for path, path_map in attention_paths_to_split.items(): old_tensor = old_checkpoint[path] channels = old_tensor.shape[0] // 3 target_shape = (-1, channels) if len(old_tensor.shape) == 3 else (-1) num_heads = old_tensor.shape[0] // config["num_head_channels"] // 3 old_tensor = old_tensor.reshape((num_heads, 3 * channels // num_heads) + old_tensor.shape[1:]) query, key, value = old_tensor.split(channels // num_heads, dim=1) checkpoint[path_map["query"]] = query.reshape(target_shape) checkpoint[path_map["key"]] = key.reshape(target_shape) checkpoint[path_map["value"]] = value.reshape(target_shape) for path in paths: new_path = path["new"] # These have already been assigned if attention_paths_to_split is not None and new_path in attention_paths_to_split: continue if additional_replacements is not None: for replacement in additional_replacements: new_path = new_path.replace(replacement["old"], replacement["new"]) # proj_attn.weight has to be converted from conv 1D to linear if "proj_attn.weight" in new_path: checkpoint[new_path] = old_checkpoint[path["old"]][:, :, 0] else: checkpoint[new_path] = old_checkpoint[path["old"]] def conv_attn_to_linear(checkpoint): keys = list(checkpoint.keys()) attn_keys = ["to_q.weight", "to_k.weight", "to_v.weight"] proj_key = "to_out.0.weight" for key in keys: if ".".join(key.split(".")[-2:]) in attn_keys or ".".join(key.split(".")[-3:]) == proj_key: if checkpoint[key].ndim > 2: checkpoint[key] = checkpoint[key].squeeze() def create_unet_diffusers_config(original_config, image_size: int): """ Creates a UNet config for diffusers based on the config of the original AudioLDM2 model. """ unet_params = original_config["model"]["params"]["unet_config"]["params"] vae_params = original_config["model"]["params"]["first_stage_config"]["params"]["ddconfig"] block_out_channels = [unet_params["model_channels"] * mult for mult in unet_params["channel_mult"]] down_block_types = [] resolution = 1 for i in range(len(block_out_channels)): block_type = "CrossAttnDownBlock2D" if resolution in unet_params["attention_resolutions"] else "DownBlock2D" down_block_types.append(block_type) if i != len(block_out_channels) - 1: resolution *= 2 up_block_types = [] for i in range(len(block_out_channels)): block_type = "CrossAttnUpBlock2D" if resolution in unet_params["attention_resolutions"] else "UpBlock2D" up_block_types.append(block_type) resolution //= 2 vae_scale_factor = 2 ** (len(vae_params["ch_mult"]) - 1) cross_attention_dim = list(unet_params["context_dim"]) if "context_dim" in unet_params else block_out_channels if len(cross_attention_dim) > 1: # require two or more cross-attention layers per-block, each of different dimension cross_attention_dim = [cross_attention_dim for _ in range(len(block_out_channels))] config = { "sample_size": image_size // vae_scale_factor, "in_channels": unet_params["in_channels"], "out_channels": unet_params["out_channels"], "down_block_types": tuple(down_block_types), "up_block_types": tuple(up_block_types), "block_out_channels": tuple(block_out_channels), "layers_per_block": unet_params["num_res_blocks"], "transformer_layers_per_block": unet_params["transformer_depth"], "cross_attention_dim": tuple(cross_attention_dim), } return config # Adapted from diffusers.pipelines.stable_diffusion.convert_from_ckpt.create_vae_diffusers_config def create_vae_diffusers_config(original_config, checkpoint, image_size: int): """ Creates a VAE config for diffusers based on the config of the original AudioLDM2 model. Compared to the original Stable Diffusion conversion, this function passes a *learnt* VAE scaling factor to the diffusers VAE. """ vae_params = original_config["model"]["params"]["first_stage_config"]["params"]["ddconfig"] _ = original_config["model"]["params"]["first_stage_config"]["params"]["embed_dim"] block_out_channels = [vae_params["ch"] * mult for mult in vae_params["ch_mult"]] down_block_types = ["DownEncoderBlock2D"] * len(block_out_channels) up_block_types = ["UpDecoderBlock2D"] * len(block_out_channels) scaling_factor = checkpoint["scale_factor"] if "scale_by_std" in original_config["model"]["params"] else 0.18215 config = { "sample_size": image_size, "in_channels": vae_params["in_channels"], "out_channels": vae_params["out_ch"], "down_block_types": tuple(down_block_types), "up_block_types": tuple(up_block_types), "block_out_channels": tuple(block_out_channels), "latent_channels": vae_params["z_channels"], "layers_per_block": vae_params["num_res_blocks"], "scaling_factor": float(scaling_factor), } return config # Copied from diffusers.pipelines.stable_diffusion.convert_from_ckpt.create_diffusers_schedular def create_diffusers_schedular(original_config): schedular = DDIMScheduler( num_train_timesteps=original_config["model"]["params"]["timesteps"], beta_start=original_config["model"]["params"]["linear_start"], beta_end=original_config["model"]["params"]["linear_end"], beta_schedule="scaled_linear", ) return schedular def convert_ldm_unet_checkpoint(checkpoint, config, path=None, extract_ema=False): """ Takes a state dict and a config, and returns a converted UNet checkpoint. """ # extract state_dict for UNet unet_state_dict = {} keys = list(checkpoint.keys()) unet_key = "model.diffusion_model." # at least a 100 parameters have to start with `model_ema` in order for the checkpoint to be EMA if sum(k.startswith("model_ema") for k in keys) > 100 and extract_ema: print(f"Checkpoint {path} has both EMA and non-EMA weights.") print( "In this conversion only the EMA weights are extracted. If you want to instead extract the non-EMA" " weights (useful to continue fine-tuning), please make sure to remove the `--extract_ema` flag." ) for key in keys: if key.startswith("model.diffusion_model"): flat_ema_key = "model_ema." + "".join(key.split(".")[1:]) unet_state_dict[key.replace(unet_key, "")] = checkpoint.pop(flat_ema_key) else: if sum(k.startswith("model_ema") for k in keys) > 100: print( "In this conversion only the non-EMA weights are extracted. If you want to instead extract the EMA" " weights (usually better for inference), please make sure to add the `--extract_ema` flag." ) # strip the unet prefix from the weight names for key in keys: if key.startswith(unet_key): unet_state_dict[key.replace(unet_key, "")] = checkpoint.pop(key) new_checkpoint = {} new_checkpoint["time_embedding.linear_1.weight"] = unet_state_dict["time_embed.0.weight"] new_checkpoint["time_embedding.linear_1.bias"] = unet_state_dict["time_embed.0.bias"] new_checkpoint["time_embedding.linear_2.weight"] = unet_state_dict["time_embed.2.weight"] new_checkpoint["time_embedding.linear_2.bias"] = unet_state_dict["time_embed.2.bias"] new_checkpoint["conv_in.weight"] = unet_state_dict["input_blocks.0.0.weight"] new_checkpoint["conv_in.bias"] = unet_state_dict["input_blocks.0.0.bias"] new_checkpoint["conv_norm_out.weight"] = unet_state_dict["out.0.weight"] new_checkpoint["conv_norm_out.bias"] = unet_state_dict["out.0.bias"] new_checkpoint["conv_out.weight"] = unet_state_dict["out.2.weight"] new_checkpoint["conv_out.bias"] = unet_state_dict["out.2.bias"] # Retrieves the keys for the input blocks only num_input_blocks = len({".".join(layer.split(".")[:2]) for layer in unet_state_dict if "input_blocks" in layer}) input_blocks = { layer_id: [key for key in unet_state_dict if f"input_blocks.{layer_id}." in key] for layer_id in range(num_input_blocks) } # Retrieves the keys for the middle blocks only num_middle_blocks = len({".".join(layer.split(".")[:2]) for layer in unet_state_dict if "middle_block" in layer}) middle_blocks = { layer_id: [key for key in unet_state_dict if f"middle_block.{layer_id}." in key] for layer_id in range(num_middle_blocks) } # Retrieves the keys for the output blocks only num_output_blocks = len({".".join(layer.split(".")[:2]) for layer in unet_state_dict if "output_blocks" in layer}) output_blocks = { layer_id: [key for key in unet_state_dict if f"output_blocks.{layer_id}." in key] for layer_id in range(num_output_blocks) } # Check how many Transformer blocks we have per layer if isinstance(config.get("cross_attention_dim"), (list, tuple)): if isinstance(config["cross_attention_dim"][0], (list, tuple)): # in this case we have multiple cross-attention layers per-block num_attention_layers = len(config.get("cross_attention_dim")[0]) else: num_attention_layers = 1 if config.get("extra_self_attn_layer"): num_attention_layers += 1 for i in range(1, num_input_blocks): block_id = (i - 1) // (config["layers_per_block"] + 1) layer_in_block_id = (i - 1) % (config["layers_per_block"] + 1) resnets = [ key for key in input_blocks[i] if f"input_blocks.{i}.0" in key and f"input_blocks.{i}.0.op" not in key ] attentions = [key for key in input_blocks[i] if f"input_blocks.{i}.0" not in key] if f"input_blocks.{i}.0.op.weight" in unet_state_dict: new_checkpoint[f"down_blocks.{block_id}.downsamplers.0.conv.weight"] = unet_state_dict.pop( f"input_blocks.{i}.0.op.weight" ) new_checkpoint[f"down_blocks.{block_id}.downsamplers.0.conv.bias"] = unet_state_dict.pop( f"input_blocks.{i}.0.op.bias" ) paths = renew_resnet_paths(resnets) meta_path = {"old": f"input_blocks.{i}.0", "new": f"down_blocks.{block_id}.resnets.{layer_in_block_id}"} assign_to_checkpoint( paths, new_checkpoint, unet_state_dict, additional_replacements=[meta_path], config=config ) if len(attentions): paths = renew_attention_paths(attentions) meta_path = [ { "old": f"input_blocks.{i}.{1 + layer_id}", "new": f"down_blocks.{block_id}.attentions.{layer_in_block_id * num_attention_layers + layer_id}", } for layer_id in range(num_attention_layers) ] assign_to_checkpoint( paths, new_checkpoint, unet_state_dict, additional_replacements=meta_path, config=config ) resnet_0 = middle_blocks[0] resnet_1 = middle_blocks[num_middle_blocks - 1] resnet_0_paths = renew_resnet_paths(resnet_0) meta_path = {"old": "middle_block.0", "new": "mid_block.resnets.0"} assign_to_checkpoint( resnet_0_paths, new_checkpoint, unet_state_dict, additional_replacements=[meta_path], config=config ) resnet_1_paths = renew_resnet_paths(resnet_1) meta_path = {"old": f"middle_block.{len(middle_blocks) - 1}", "new": "mid_block.resnets.1"} assign_to_checkpoint( resnet_1_paths, new_checkpoint, unet_state_dict, additional_replacements=[meta_path], config=config ) for i in range(1, num_middle_blocks - 1): attentions = middle_blocks[i] attentions_paths = renew_attention_paths(attentions) meta_path = {"old": f"middle_block.{i}", "new": f"mid_block.attentions.{i - 1}"} assign_to_checkpoint( attentions_paths, new_checkpoint, unet_state_dict, additional_replacements=[meta_path], config=config ) for i in range(num_output_blocks): block_id = i // (config["layers_per_block"] + 1) layer_in_block_id = i % (config["layers_per_block"] + 1) output_block_layers = [shave_segments(name, 2) for name in output_blocks[i]] output_block_list = {} for layer in output_block_layers: layer_id, layer_name = layer.split(".")[0], shave_segments(layer, 1) if layer_id in output_block_list: output_block_list[layer_id].append(layer_name) else: output_block_list[layer_id] = [layer_name] if len(output_block_list) > 1: resnets = [key for key in output_blocks[i] if f"output_blocks.{i}.0" in key] attentions = [key for key in output_blocks[i] if f"output_blocks.{i}.0" not in key] paths = renew_resnet_paths(resnets) meta_path = {"old": f"output_blocks.{i}.0", "new": f"up_blocks.{block_id}.resnets.{layer_in_block_id}"} assign_to_checkpoint( paths, new_checkpoint, unet_state_dict, additional_replacements=[meta_path], config=config ) output_block_list = {k: sorted(v) for k, v in output_block_list.items()} if ["conv.bias", "conv.weight"] in output_block_list.values(): index = list(output_block_list.values()).index(["conv.bias", "conv.weight"]) new_checkpoint[f"up_blocks.{block_id}.upsamplers.0.conv.weight"] = unet_state_dict[ f"output_blocks.{i}.{index}.conv.weight" ] new_checkpoint[f"up_blocks.{block_id}.upsamplers.0.conv.bias"] = unet_state_dict[ f"output_blocks.{i}.{index}.conv.bias" ] attentions.remove(f"output_blocks.{i}.{index}.conv.bias") attentions.remove(f"output_blocks.{i}.{index}.conv.weight") # Clear attentions as they have been attributed above. if len(attentions) == 2: attentions = [] if len(attentions): paths = renew_attention_paths(attentions) meta_path = [ { "old": f"output_blocks.{i}.{1 + layer_id}", "new": f"up_blocks.{block_id}.attentions.{layer_in_block_id * num_attention_layers + layer_id}", } for layer_id in range(num_attention_layers) ] assign_to_checkpoint( paths, new_checkpoint, unet_state_dict, additional_replacements=meta_path, config=config ) else: resnet_0_paths = renew_resnet_paths(output_block_layers, n_shave_prefix_segments=1) for path in resnet_0_paths: old_path = ".".join(["output_blocks", str(i), path["old"]]) new_path = ".".join(["up_blocks", str(block_id), "resnets", str(layer_in_block_id), path["new"]]) new_checkpoint[new_path] = unet_state_dict[old_path] return new_checkpoint def convert_ldm_vae_checkpoint(checkpoint, config): # extract state dict for VAE vae_state_dict = {} vae_key = "first_stage_model." keys = list(checkpoint.keys()) for key in keys: if key.startswith(vae_key): vae_state_dict[key.replace(vae_key, "")] = checkpoint.get(key) new_checkpoint = {} new_checkpoint["encoder.conv_in.weight"] = vae_state_dict["encoder.conv_in.weight"] new_checkpoint["encoder.conv_in.bias"] = vae_state_dict["encoder.conv_in.bias"] new_checkpoint["encoder.conv_out.weight"] = vae_state_dict["encoder.conv_out.weight"] new_checkpoint["encoder.conv_out.bias"] = vae_state_dict["encoder.conv_out.bias"] new_checkpoint["encoder.conv_norm_out.weight"] = vae_state_dict["encoder.norm_out.weight"] new_checkpoint["encoder.conv_norm_out.bias"] = vae_state_dict["encoder.norm_out.bias"] new_checkpoint["decoder.conv_in.weight"] = vae_state_dict["decoder.conv_in.weight"] new_checkpoint["decoder.conv_in.bias"] = vae_state_dict["decoder.conv_in.bias"] new_checkpoint["decoder.conv_out.weight"] = vae_state_dict["decoder.conv_out.weight"] new_checkpoint["decoder.conv_out.bias"] = vae_state_dict["decoder.conv_out.bias"] new_checkpoint["decoder.conv_norm_out.weight"] = vae_state_dict["decoder.norm_out.weight"] new_checkpoint["decoder.conv_norm_out.bias"] = vae_state_dict["decoder.norm_out.bias"] new_checkpoint["quant_conv.weight"] = vae_state_dict["quant_conv.weight"] new_checkpoint["quant_conv.bias"] = vae_state_dict["quant_conv.bias"] new_checkpoint["post_quant_conv.weight"] = vae_state_dict["post_quant_conv.weight"] new_checkpoint["post_quant_conv.bias"] = vae_state_dict["post_quant_conv.bias"] # Retrieves the keys for the encoder down blocks only num_down_blocks = len({".".join(layer.split(".")[:3]) for layer in vae_state_dict if "encoder.down" in layer}) down_blocks = { layer_id: [key for key in vae_state_dict if f"down.{layer_id}" in key] for layer_id in range(num_down_blocks) } # Retrieves the keys for the decoder up blocks only num_up_blocks = len({".".join(layer.split(".")[:3]) for layer in vae_state_dict if "decoder.up" in layer}) up_blocks = { layer_id: [key for key in vae_state_dict if f"up.{layer_id}" in key] for layer_id in range(num_up_blocks) } for i in range(num_down_blocks): resnets = [key for key in down_blocks[i] if f"down.{i}" in key and f"down.{i}.downsample" not in key] if f"encoder.down.{i}.downsample.conv.weight" in vae_state_dict: new_checkpoint[f"encoder.down_blocks.{i}.downsamplers.0.conv.weight"] = vae_state_dict.pop( f"encoder.down.{i}.downsample.conv.weight" ) new_checkpoint[f"encoder.down_blocks.{i}.downsamplers.0.conv.bias"] = vae_state_dict.pop( f"encoder.down.{i}.downsample.conv.bias" ) paths = renew_vae_resnet_paths(resnets) meta_path = {"old": f"down.{i}.block", "new": f"down_blocks.{i}.resnets"} assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config) mid_resnets = [key for key in vae_state_dict if "encoder.mid.block" in key] num_mid_res_blocks = 2 for i in range(1, num_mid_res_blocks + 1): resnets = [key for key in mid_resnets if f"encoder.mid.block_{i}" in key] paths = renew_vae_resnet_paths(resnets) meta_path = {"old": f"mid.block_{i}", "new": f"mid_block.resnets.{i - 1}"} assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config) mid_attentions = [key for key in vae_state_dict if "encoder.mid.attn" in key] paths = renew_vae_attention_paths(mid_attentions) meta_path = {"old": "mid.attn_1", "new": "mid_block.attentions.0"} assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config) conv_attn_to_linear(new_checkpoint) for i in range(num_up_blocks): block_id = num_up_blocks - 1 - i resnets = [ key for key in up_blocks[block_id] if f"up.{block_id}" in key and f"up.{block_id}.upsample" not in key ] if f"decoder.up.{block_id}.upsample.conv.weight" in vae_state_dict: new_checkpoint[f"decoder.up_blocks.{i}.upsamplers.0.conv.weight"] = vae_state_dict[ f"decoder.up.{block_id}.upsample.conv.weight" ] new_checkpoint[f"decoder.up_blocks.{i}.upsamplers.0.conv.bias"] = vae_state_dict[ f"decoder.up.{block_id}.upsample.conv.bias" ] paths = renew_vae_resnet_paths(resnets) meta_path = {"old": f"up.{block_id}.block", "new": f"up_blocks.{i}.resnets"} assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config) mid_resnets = [key for key in vae_state_dict if "decoder.mid.block" in key] num_mid_res_blocks = 2 for i in range(1, num_mid_res_blocks + 1): resnets = [key for key in mid_resnets if f"decoder.mid.block_{i}" in key] paths = renew_vae_resnet_paths(resnets) meta_path = {"old": f"mid.block_{i}", "new": f"mid_block.resnets.{i - 1}"} assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config) mid_attentions = [key for key in vae_state_dict if "decoder.mid.attn" in key] paths = renew_vae_attention_paths(mid_attentions) meta_path = {"old": "mid.attn_1", "new": "mid_block.attentions.0"} assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config) conv_attn_to_linear(new_checkpoint) return new_checkpoint CLAP_KEYS_TO_MODIFY_MAPPING = { "text_branch": "text_model", "audio_branch": "audio_model.audio_encoder", "attn": "attention.self", "self.proj": "output.dense", "attention.self_mask": "attn_mask", "mlp.fc1": "intermediate.dense", "mlp.fc2": "output.dense", "norm1": "layernorm_before", "norm2": "layernorm_after", "bn0": "batch_norm", } CLAP_KEYS_TO_IGNORE = [ "text_transform", "audio_transform", "stft", "logmel_extractor", "tscam_conv", "head", "attn_mask", ] CLAP_EXPECTED_MISSING_KEYS = ["text_model.embeddings.token_type_ids"] def convert_open_clap_checkpoint(checkpoint): """ Takes a state dict and returns a converted CLAP checkpoint. """ # extract state dict for CLAP text embedding model, discarding the audio component model_state_dict = {} model_key = "clap.model." keys = list(checkpoint.keys()) for key in keys: if key.startswith(model_key): model_state_dict[key.replace(model_key, "")] = checkpoint.get(key) new_checkpoint = {} sequential_layers_pattern = r".*sequential.(\d+).*" text_projection_pattern = r".*_projection.(\d+).*" for key, value in model_state_dict.items(): # check if key should be ignored in mapping - if so map it to a key name that we'll filter out at the end for key_to_ignore in CLAP_KEYS_TO_IGNORE: if key_to_ignore in key: key = "spectrogram" # check if any key needs to be modified for key_to_modify, new_key in CLAP_KEYS_TO_MODIFY_MAPPING.items(): if key_to_modify in key: key = key.replace(key_to_modify, new_key) if re.match(sequential_layers_pattern, key): # replace sequential layers with list sequential_layer = re.match(sequential_layers_pattern, key).group(1) key = key.replace(f"sequential.{sequential_layer}.", f"layers.{int(sequential_layer)//3}.linear.") elif re.match(text_projection_pattern, key): projecton_layer = int(re.match(text_projection_pattern, key).group(1)) # Because in CLAP they use `nn.Sequential`... transformers_projection_layer = 1 if projecton_layer == 0 else 2 key = key.replace(f"_projection.{projecton_layer}.", f"_projection.linear{transformers_projection_layer}.") if "audio" and "qkv" in key: # split qkv into query key and value mixed_qkv = value qkv_dim = mixed_qkv.size(0) // 3 query_layer = mixed_qkv[:qkv_dim] key_layer = mixed_qkv[qkv_dim : qkv_dim * 2] value_layer = mixed_qkv[qkv_dim * 2 :] new_checkpoint[key.replace("qkv", "query")] = query_layer new_checkpoint[key.replace("qkv", "key")] = key_layer new_checkpoint[key.replace("qkv", "value")] = value_layer elif key != "spectrogram": new_checkpoint[key] = value return new_checkpoint def create_transformers_vocoder_config(original_config): """ Creates a config for transformers SpeechT5HifiGan based on the config of the vocoder model. """ vocoder_params = original_config["model"]["params"]["vocoder_config"]["params"] config = { "model_in_dim": vocoder_params["num_mels"], "sampling_rate": vocoder_params["sampling_rate"], "upsample_initial_channel": vocoder_params["upsample_initial_channel"], "upsample_rates": list(vocoder_params["upsample_rates"]), "upsample_kernel_sizes": list(vocoder_params["upsample_kernel_sizes"]), "resblock_kernel_sizes": list(vocoder_params["resblock_kernel_sizes"]), "resblock_dilation_sizes": [ list(resblock_dilation) for resblock_dilation in vocoder_params["resblock_dilation_sizes"] ], "normalize_before": False, } return config def extract_sub_model(checkpoint, key_prefix): """ Takes a state dict and returns the state dict for a particular sub-model. """ sub_model_state_dict = {} keys = list(checkpoint.keys()) for key in keys: if key.startswith(key_prefix): sub_model_state_dict[key.replace(key_prefix, "")] = checkpoint.get(key) return sub_model_state_dict def convert_hifigan_checkpoint(checkpoint, config): """ Takes a state dict and config, and returns a converted HiFiGAN vocoder checkpoint. """ # extract state dict for vocoder vocoder_state_dict = extract_sub_model(checkpoint, key_prefix="first_stage_model.vocoder.") # fix upsampler keys, everything else is correct already for i in range(len(config.upsample_rates)): vocoder_state_dict[f"upsampler.{i}.weight"] = vocoder_state_dict.pop(f"ups.{i}.weight") vocoder_state_dict[f"upsampler.{i}.bias"] = vocoder_state_dict.pop(f"ups.{i}.bias") if not config.normalize_before: # if we don't set normalize_before then these variables are unused, so we set them to their initialised values vocoder_state_dict["mean"] = torch.zeros(config.model_in_dim) vocoder_state_dict["scale"] = torch.ones(config.model_in_dim) return vocoder_state_dict def convert_projection_checkpoint(checkpoint): projection_state_dict = {} conditioner_state_dict = extract_sub_model(checkpoint, key_prefix="cond_stage_models.0.") projection_state_dict["sos_embed"] = conditioner_state_dict["start_of_sequence_tokens.weight"][0] projection_state_dict["sos_embed_1"] = conditioner_state_dict["start_of_sequence_tokens.weight"][1] projection_state_dict["eos_embed"] = conditioner_state_dict["end_of_sequence_tokens.weight"][0] projection_state_dict["eos_embed_1"] = conditioner_state_dict["end_of_sequence_tokens.weight"][1] projection_state_dict["projection.weight"] = conditioner_state_dict["input_sequence_embed_linear.0.weight"] projection_state_dict["projection.bias"] = conditioner_state_dict["input_sequence_embed_linear.0.bias"] projection_state_dict["projection_1.weight"] = conditioner_state_dict["input_sequence_embed_linear.1.weight"] projection_state_dict["projection_1.bias"] = conditioner_state_dict["input_sequence_embed_linear.1.bias"] return projection_state_dict # Adapted from https://github.com/haoheliu/AudioLDM2/blob/81ad2c6ce015c1310387695e2dae975a7d2ed6fd/audioldm2/utils.py#L143 DEFAULT_CONFIG = { "model": { "params": { "linear_start": 0.0015, "linear_end": 0.0195, "timesteps": 1000, "channels": 8, "scale_by_std": True, "unet_config": { "target": "audioldm2.latent_diffusion.openaimodel.UNetModel", "params": { "context_dim": [None, 768, 1024], "in_channels": 8, "out_channels": 8, "model_channels": 128, "attention_resolutions": [8, 4, 2], "num_res_blocks": 2, "channel_mult": [1, 2, 3, 5], "num_head_channels": 32, "transformer_depth": 1, }, }, "first_stage_config": { "target": "audioldm2.variational_autoencoder.autoencoder.AutoencoderKL", "params": { "embed_dim": 8, "ddconfig": { "z_channels": 8, "resolution": 256, "in_channels": 1, "out_ch": 1, "ch": 128, "ch_mult": [1, 2, 4], "num_res_blocks": 2, }, }, }, "cond_stage_config": { "crossattn_audiomae_generated": { "target": "audioldm2.latent_diffusion.modules.encoders.modules.SequenceGenAudioMAECond", "params": { "sequence_gen_length": 8, "sequence_input_embed_dim": [512, 1024], }, } }, "vocoder_config": { "target": "audioldm2.first_stage_model.vocoder", "params": { "upsample_rates": [5, 4, 2, 2, 2], "upsample_kernel_sizes": [16, 16, 8, 4, 4], "upsample_initial_channel": 1024, "resblock_kernel_sizes": [3, 7, 11], "resblock_dilation_sizes": [[1, 3, 5], [1, 3, 5], [1, 3, 5]], "num_mels": 64, "sampling_rate": 16000, }, }, }, }, } def load_pipeline_from_original_AudioLDM2_ckpt( checkpoint_path: str, original_config_file: str = None, image_size: int = 1024, prediction_type: str = None, extract_ema: bool = False, scheduler_type: str = "ddim", cross_attention_dim: Union[List, List[List]] = None, transformer_layers_per_block: int = None, device: str = None, from_safetensors: bool = False, ) -> AudioLDM2Pipeline: """ Load an AudioLDM2 pipeline object from a `.ckpt`/`.safetensors` file and (ideally) a `.yaml` config file. Although many of the arguments can be automatically inferred, some of these rely on brittle checks against the global step count, which will likely fail for models that have undergone further fine-tuning. Therefore, it is recommended that you override the default values and/or supply an `original_config_file` wherever possible. Args: checkpoint_path (`str`): Path to `.ckpt` file. original_config_file (`str`): Path to `.yaml` config file corresponding to the original architecture. If `None`, will be automatically set to the AudioLDM2 base config. image_size (`int`, *optional*, defaults to 1024): The image size that the model was trained on. prediction_type (`str`, *optional*): The prediction type that the model was trained on. If `None`, will be automatically inferred by looking for a key in the config. For the default config, the prediction type is `'epsilon'`. scheduler_type (`str`, *optional*, defaults to 'ddim'): Type of scheduler to use. Should be one of `["pndm", "lms", "heun", "euler", "euler-ancestral", "dpm", "ddim"]`. cross_attention_dim (`list`, *optional*, defaults to `None`): The dimension of the cross-attention layers. If `None`, the cross-attention dimension will be automatically inferred. Set to `[768, 1024]` for the base model, or `[768, 1024, None]` for the large model. transformer_layers_per_block (`int`, *optional*, defaults to `None`): The number of transformer layers in each transformer block. If `None`, number of layers will be " "automatically inferred. Set to `1` for the base model, or `2` for the large model. extract_ema (`bool`, *optional*, defaults to `False`): Only relevant for checkpoints that have both EMA and non-EMA weights. Whether to extract the EMA weights or not. Defaults to `False`. Pass `True` to extract the EMA weights. EMA weights usually yield higher quality images for inference. Non-EMA weights are usually better to continue fine-tuning. device (`str`, *optional*, defaults to `None`): The device to use. Pass `None` to determine automatically. from_safetensors (`str`, *optional*, defaults to `False`): If `checkpoint_path` is in `safetensors` format, load checkpoint with safetensors instead of PyTorch. return: An AudioLDM2Pipeline object representing the passed-in `.ckpt`/`.safetensors` file. """ if from_safetensors: if not is_safetensors_available(): raise ValueError(BACKENDS_MAPPING["safetensors"][1]) from safetensors import safe_open checkpoint = {} with safe_open(checkpoint_path, framework="pt", device="cpu") as f: for key in f.keys(): checkpoint[key] = f.get_tensor(key) else: if device is None: device = "cuda" if torch.cuda.is_available() else "cpu" checkpoint = torch.load(checkpoint_path, map_location=device) else: checkpoint = torch.load(checkpoint_path, map_location=device) if "state_dict" in checkpoint: checkpoint = checkpoint["state_dict"] if original_config_file is None: original_config = DEFAULT_CONFIG else: original_config = yaml.safe_load(original_config_file) if image_size is not None: original_config["model"]["params"]["unet_config"]["params"]["image_size"] = image_size if cross_attention_dim is not None: original_config["model"]["params"]["unet_config"]["params"]["context_dim"] = cross_attention_dim if transformer_layers_per_block is not None: original_config["model"]["params"]["unet_config"]["params"]["transformer_depth"] = transformer_layers_per_block if ( "parameterization" in original_config["model"]["params"] and original_config["model"]["params"]["parameterization"] == "v" ): if prediction_type is None: prediction_type = "v_prediction" else: if prediction_type is None: prediction_type = "epsilon" num_train_timesteps = original_config["model"]["params"]["timesteps"] beta_start = original_config["model"]["params"]["linear_start"] beta_end = original_config["model"]["params"]["linear_end"] scheduler = DDIMScheduler( beta_end=beta_end, beta_schedule="scaled_linear", beta_start=beta_start, num_train_timesteps=num_train_timesteps, steps_offset=1, clip_sample=False, set_alpha_to_one=False, prediction_type=prediction_type, ) # make sure scheduler works correctly with DDIM scheduler.register_to_config(clip_sample=False) if scheduler_type == "pndm": config = dict(scheduler.config) config["skip_prk_steps"] = True scheduler = PNDMScheduler.from_config(config) elif scheduler_type == "lms": scheduler = LMSDiscreteScheduler.from_config(scheduler.config) elif scheduler_type == "heun": scheduler = HeunDiscreteScheduler.from_config(scheduler.config) elif scheduler_type == "euler": scheduler = EulerDiscreteScheduler.from_config(scheduler.config) elif scheduler_type == "euler-ancestral": scheduler = EulerAncestralDiscreteScheduler.from_config(scheduler.config) elif scheduler_type == "dpm": scheduler = DPMSolverMultistepScheduler.from_config(scheduler.config) elif scheduler_type == "ddim": scheduler = scheduler else: raise ValueError(f"Scheduler of type {scheduler_type} doesn't exist!") # Convert the UNet2DModel unet_config = create_unet_diffusers_config(original_config, image_size=image_size) unet = AudioLDM2UNet2DConditionModel(**unet_config) converted_unet_checkpoint = convert_ldm_unet_checkpoint( checkpoint, unet_config, path=checkpoint_path, extract_ema=extract_ema ) unet.load_state_dict(converted_unet_checkpoint) # Convert the VAE model vae_config = create_vae_diffusers_config(original_config, checkpoint=checkpoint, image_size=image_size) converted_vae_checkpoint = convert_ldm_vae_checkpoint(checkpoint, vae_config) vae = AutoencoderKL(**vae_config) vae.load_state_dict(converted_vae_checkpoint) # Convert the joint audio-text encoding model clap_config = ClapConfig.from_pretrained("laion/clap-htsat-unfused") clap_config.audio_config.update( { "patch_embeds_hidden_size": 128, "hidden_size": 1024, "depths": [2, 2, 12, 2], } ) # AudioLDM2 uses the same tokenizer and feature extractor as the original CLAP model clap_tokenizer = AutoTokenizer.from_pretrained("laion/clap-htsat-unfused") clap_feature_extractor = AutoFeatureExtractor.from_pretrained("laion/clap-htsat-unfused") converted_clap_model = convert_open_clap_checkpoint(checkpoint) clap_model = ClapModel(clap_config) missing_keys, unexpected_keys = clap_model.load_state_dict(converted_clap_model, strict=False) # we expect not to have token_type_ids in our original state dict so let's ignore them missing_keys = list(set(missing_keys) - set(CLAP_EXPECTED_MISSING_KEYS)) if len(unexpected_keys) > 0: raise ValueError(f"Unexpected keys when loading CLAP model: {unexpected_keys}") if len(missing_keys) > 0: raise ValueError(f"Missing keys when loading CLAP model: {missing_keys}") # Convert the vocoder model vocoder_config = create_transformers_vocoder_config(original_config) vocoder_config = SpeechT5HifiGanConfig(**vocoder_config) converted_vocoder_checkpoint = convert_hifigan_checkpoint(checkpoint, vocoder_config) vocoder = SpeechT5HifiGan(vocoder_config) vocoder.load_state_dict(converted_vocoder_checkpoint) # Convert the Flan-T5 encoder model: AudioLDM2 uses the same configuration and tokenizer as the original Flan-T5 large model t5_config = T5Config.from_pretrained("google/flan-t5-large") converted_t5_checkpoint = extract_sub_model(checkpoint, key_prefix="cond_stage_models.1.model.") t5_tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-large") # hard-coded in the original implementation (i.e. not retrievable from the config) t5_tokenizer.model_max_length = 128 t5_model = T5EncoderModel(t5_config) t5_model.load_state_dict(converted_t5_checkpoint) # Convert the GPT2 encoder model: AudioLDM2 uses the same configuration as the original GPT2 base model gpt2_config = GPT2Config.from_pretrained("gpt2") gpt2_model = GPT2Model(gpt2_config) gpt2_model.config.max_new_tokens = original_config["model"]["params"]["cond_stage_config"][ "crossattn_audiomae_generated" ]["params"]["sequence_gen_length"] converted_gpt2_checkpoint = extract_sub_model(checkpoint, key_prefix="cond_stage_models.0.model.") gpt2_model.load_state_dict(converted_gpt2_checkpoint) # Convert the extra embedding / projection layers projection_model = AudioLDM2ProjectionModel(clap_config.projection_dim, t5_config.d_model, gpt2_config.n_embd) converted_projection_checkpoint = convert_projection_checkpoint(checkpoint) projection_model.load_state_dict(converted_projection_checkpoint) # Instantiate the diffusers pipeline pipe = AudioLDM2Pipeline( vae=vae, text_encoder=clap_model, text_encoder_2=t5_model, projection_model=projection_model, language_model=gpt2_model, tokenizer=clap_tokenizer, tokenizer_2=t5_tokenizer, feature_extractor=clap_feature_extractor, unet=unet, scheduler=scheduler, vocoder=vocoder, ) return pipe if __name__ == "__main__": parser = argparse.ArgumentParser() parser.add_argument( "--checkpoint_path", default=None, type=str, required=True, help="Path to the checkpoint to convert." ) parser.add_argument( "--original_config_file", default=None, type=str, help="The YAML config file corresponding to the original architecture.", ) parser.add_argument( "--cross_attention_dim", default=None, type=int, nargs="+", help="The dimension of the cross-attention layers. If `None`, the cross-attention dimension will be " "automatically inferred. Set to `768+1024` for the base model, or `768+1024+640` for the large model", ) parser.add_argument( "--transformer_layers_per_block", default=None, type=int, help="The number of transformer layers in each transformer block. If `None`, number of layers will be " "automatically inferred. Set to `1` for the base model, or `2` for the large model.", ) parser.add_argument( "--scheduler_type", default="ddim", type=str, help="Type of scheduler to use. Should be one of ['pndm', 'lms', 'ddim', 'euler', 'euler-ancestral', 'dpm']", ) parser.add_argument( "--image_size", default=1048, type=int, help="The image size that the model was trained on.", ) parser.add_argument( "--prediction_type", default=None, type=str, help=("The prediction type that the model was trained on."), ) parser.add_argument( "--extract_ema", action="store_true", help=( "Only relevant for checkpoints that have both EMA and non-EMA weights. Whether to extract the EMA weights" " or not. Defaults to `False`. Add `--extract_ema` to extract the EMA weights. EMA weights usually yield" " higher quality images for inference. Non-EMA weights are usually better to continue fine-tuning." ), ) parser.add_argument( "--from_safetensors", action="store_true", help="If `--checkpoint_path` is in `safetensors` format, load checkpoint with safetensors instead of PyTorch.", ) parser.add_argument( "--to_safetensors", action="store_true", help="Whether to store pipeline in safetensors format or not.", ) parser.add_argument("--dump_path", default=None, type=str, required=True, help="Path to the output model.") parser.add_argument("--device", type=str, help="Device to use (e.g. cpu, cuda:0, cuda:1, etc.)") args = parser.parse_args() pipe = load_pipeline_from_original_AudioLDM2_ckpt( checkpoint_path=args.checkpoint_path, original_config_file=args.original_config_file, image_size=args.image_size, prediction_type=args.prediction_type, extract_ema=args.extract_ema, scheduler_type=args.scheduler_type, cross_attention_dim=args.cross_attention_dim, transformer_layers_per_block=args.transformer_layers_per_block, from_safetensors=args.from_safetensors, device=args.device, ) pipe.save_pretrained(args.dump_path, safe_serialization=args.to_safetensors)

diffusers/scripts/convert_original_audioldm2_to_diffusers.py/0

import argparse import io import requests import torch import yaml from diffusers import AutoencoderKL from diffusers.pipelines.stable_diffusion.convert_from_ckpt import ( assign_to_checkpoint, conv_attn_to_linear, create_vae_diffusers_config, renew_vae_attention_paths, renew_vae_resnet_paths, ) def custom_convert_ldm_vae_checkpoint(checkpoint, config): vae_state_dict = checkpoint new_checkpoint = {} new_checkpoint["encoder.conv_in.weight"] = vae_state_dict["encoder.conv_in.weight"] new_checkpoint["encoder.conv_in.bias"] = vae_state_dict["encoder.conv_in.bias"] new_checkpoint["encoder.conv_out.weight"] = vae_state_dict["encoder.conv_out.weight"] new_checkpoint["encoder.conv_out.bias"] = vae_state_dict["encoder.conv_out.bias"] new_checkpoint["encoder.conv_norm_out.weight"] = vae_state_dict["encoder.norm_out.weight"] new_checkpoint["encoder.conv_norm_out.bias"] = vae_state_dict["encoder.norm_out.bias"] new_checkpoint["decoder.conv_in.weight"] = vae_state_dict["decoder.conv_in.weight"] new_checkpoint["decoder.conv_in.bias"] = vae_state_dict["decoder.conv_in.bias"] new_checkpoint["decoder.conv_out.weight"] = vae_state_dict["decoder.conv_out.weight"] new_checkpoint["decoder.conv_out.bias"] = vae_state_dict["decoder.conv_out.bias"] new_checkpoint["decoder.conv_norm_out.weight"] = vae_state_dict["decoder.norm_out.weight"] new_checkpoint["decoder.conv_norm_out.bias"] = vae_state_dict["decoder.norm_out.bias"] new_checkpoint["quant_conv.weight"] = vae_state_dict["quant_conv.weight"] new_checkpoint["quant_conv.bias"] = vae_state_dict["quant_conv.bias"] new_checkpoint["post_quant_conv.weight"] = vae_state_dict["post_quant_conv.weight"] new_checkpoint["post_quant_conv.bias"] = vae_state_dict["post_quant_conv.bias"] # Retrieves the keys for the encoder down blocks only num_down_blocks = len({".".join(layer.split(".")[:3]) for layer in vae_state_dict if "encoder.down" in layer}) down_blocks = { layer_id: [key for key in vae_state_dict if f"down.{layer_id}" in key] for layer_id in range(num_down_blocks) } # Retrieves the keys for the decoder up blocks only num_up_blocks = len({".".join(layer.split(".")[:3]) for layer in vae_state_dict if "decoder.up" in layer}) up_blocks = { layer_id: [key for key in vae_state_dict if f"up.{layer_id}" in key] for layer_id in range(num_up_blocks) } for i in range(num_down_blocks): resnets = [key for key in down_blocks[i] if f"down.{i}" in key and f"down.{i}.downsample" not in key] if f"encoder.down.{i}.downsample.conv.weight" in vae_state_dict: new_checkpoint[f"encoder.down_blocks.{i}.downsamplers.0.conv.weight"] = vae_state_dict.pop( f"encoder.down.{i}.downsample.conv.weight" ) new_checkpoint[f"encoder.down_blocks.{i}.downsamplers.0.conv.bias"] = vae_state_dict.pop( f"encoder.down.{i}.downsample.conv.bias" ) paths = renew_vae_resnet_paths(resnets) meta_path = {"old": f"down.{i}.block", "new": f"down_blocks.{i}.resnets"} assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config) mid_resnets = [key for key in vae_state_dict if "encoder.mid.block" in key] num_mid_res_blocks = 2 for i in range(1, num_mid_res_blocks + 1): resnets = [key for key in mid_resnets if f"encoder.mid.block_{i}" in key] paths = renew_vae_resnet_paths(resnets) meta_path = {"old": f"mid.block_{i}", "new": f"mid_block.resnets.{i - 1}"} assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config) mid_attentions = [key for key in vae_state_dict if "encoder.mid.attn" in key] paths = renew_vae_attention_paths(mid_attentions) meta_path = {"old": "mid.attn_1", "new": "mid_block.attentions.0"} assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config) conv_attn_to_linear(new_checkpoint) for i in range(num_up_blocks): block_id = num_up_blocks - 1 - i resnets = [ key for key in up_blocks[block_id] if f"up.{block_id}" in key and f"up.{block_id}.upsample" not in key ] if f"decoder.up.{block_id}.upsample.conv.weight" in vae_state_dict: new_checkpoint[f"decoder.up_blocks.{i}.upsamplers.0.conv.weight"] = vae_state_dict[ f"decoder.up.{block_id}.upsample.conv.weight" ] new_checkpoint[f"decoder.up_blocks.{i}.upsamplers.0.conv.bias"] = vae_state_dict[ f"decoder.up.{block_id}.upsample.conv.bias" ] paths = renew_vae_resnet_paths(resnets) meta_path = {"old": f"up.{block_id}.block", "new": f"up_blocks.{i}.resnets"} assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config) mid_resnets = [key for key in vae_state_dict if "decoder.mid.block" in key] num_mid_res_blocks = 2 for i in range(1, num_mid_res_blocks + 1): resnets = [key for key in mid_resnets if f"decoder.mid.block_{i}" in key] paths = renew_vae_resnet_paths(resnets) meta_path = {"old": f"mid.block_{i}", "new": f"mid_block.resnets.{i - 1}"} assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config) mid_attentions = [key for key in vae_state_dict if "decoder.mid.attn" in key] paths = renew_vae_attention_paths(mid_attentions) meta_path = {"old": "mid.attn_1", "new": "mid_block.attentions.0"} assign_to_checkpoint(paths, new_checkpoint, vae_state_dict, additional_replacements=[meta_path], config=config) conv_attn_to_linear(new_checkpoint) return new_checkpoint def vae_pt_to_vae_diffuser( checkpoint_path: str, output_path: str, ): # Only support V1 r = requests.get( " https://raw.githubusercontent.com/CompVis/stable-diffusion/main/configs/stable-diffusion/v1-inference.yaml" ) io_obj = io.BytesIO(r.content) original_config = yaml.safe_load(io_obj) image_size = 512 device = "cuda" if torch.cuda.is_available() else "cpu" if checkpoint_path.endswith("safetensors"): from safetensors import safe_open checkpoint = {} with safe_open(checkpoint_path, framework="pt", device="cpu") as f: for key in f.keys(): checkpoint[key] = f.get_tensor(key) else: checkpoint = torch.load(checkpoint_path, map_location=device)["state_dict"] # Convert the VAE model. vae_config = create_vae_diffusers_config(original_config, image_size=image_size) converted_vae_checkpoint = custom_convert_ldm_vae_checkpoint(checkpoint, vae_config) vae = AutoencoderKL(**vae_config) vae.load_state_dict(converted_vae_checkpoint) vae.save_pretrained(output_path) if __name__ == "__main__": parser = argparse.ArgumentParser() parser.add_argument("--vae_pt_path", default=None, type=str, required=True, help="Path to the VAE.pt to convert.") parser.add_argument("--dump_path", default=None, type=str, required=True, help="Path to the VAE.pt to convert.") args = parser.parse_args() vae_pt_to_vae_diffuser(args.vae_pt_path, args.dump_path)

diffusers/scripts/convert_vae_pt_to_diffusers.py/0

from .rl import ValueGuidedRLPipeline

diffusers/src/diffusers/experimental/__init__.py/0

# Models For more detail on the models, please refer to the [docs](https://huggingface.co/docs/diffusers/api/models/overview).

diffusers/src/diffusers/models/README.md/0

# Copyright 2023 The HuggingFace Team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from typing import Optional, Tuple import torch import torch.nn as nn import torch.nn.functional as F from ..utils import USE_PEFT_BACKEND from .lora import LoRACompatibleConv from .normalization import RMSNorm from .upsampling import upfirdn2d_native class Downsample1D(nn.Module): """A 1D downsampling layer with an optional convolution. Parameters: channels (`int`): number of channels in the inputs and outputs. use_conv (`bool`, default `False`): option to use a convolution. out_channels (`int`, optional): number of output channels. Defaults to `channels`. padding (`int`, default `1`): padding for the convolution. name (`str`, default `conv`): name of the downsampling 1D layer. """ def __init__( self, channels: int, use_conv: bool = False, out_channels: Optional[int] = None, padding: int = 1, name: str = "conv", ): super().__init__() self.channels = channels self.out_channels = out_channels or channels self.use_conv = use_conv self.padding = padding stride = 2 self.name = name if use_conv: self.conv = nn.Conv1d(self.channels, self.out_channels, 3, stride=stride, padding=padding) else: assert self.channels == self.out_channels self.conv = nn.AvgPool1d(kernel_size=stride, stride=stride) def forward(self, inputs: torch.Tensor) -> torch.Tensor: assert inputs.shape[1] == self.channels return self.conv(inputs) class Downsample2D(nn.Module): """A 2D downsampling layer with an optional convolution. Parameters: channels (`int`): number of channels in the inputs and outputs. use_conv (`bool`, default `False`): option to use a convolution. out_channels (`int`, optional): number of output channels. Defaults to `channels`. padding (`int`, default `1`): padding for the convolution. name (`str`, default `conv`): name of the downsampling 2D layer. """ def __init__( self, channels: int, use_conv: bool = False, out_channels: Optional[int] = None, padding: int = 1, name: str = "conv", kernel_size=3, norm_type=None, eps=None, elementwise_affine=None, bias=True, ): super().__init__() self.channels = channels self.out_channels = out_channels or channels self.use_conv = use_conv self.padding = padding stride = 2 self.name = name conv_cls = nn.Conv2d if USE_PEFT_BACKEND else LoRACompatibleConv if norm_type == "ln_norm": self.norm = nn.LayerNorm(channels, eps, elementwise_affine) elif norm_type == "rms_norm": self.norm = RMSNorm(channels, eps, elementwise_affine) elif norm_type is None: self.norm = None else: raise ValueError(f"unknown norm_type: {norm_type}") if use_conv: conv = conv_cls( self.channels, self.out_channels, kernel_size=kernel_size, stride=stride, padding=padding, bias=bias ) else: assert self.channels == self.out_channels conv = nn.AvgPool2d(kernel_size=stride, stride=stride) # TODO(Suraj, Patrick) - clean up after weight dicts are correctly renamed if name == "conv": self.Conv2d_0 = conv self.conv = conv elif name == "Conv2d_0": self.conv = conv else: self.conv = conv def forward(self, hidden_states: torch.FloatTensor, scale: float = 1.0) -> torch.FloatTensor: assert hidden_states.shape[1] == self.channels if self.norm is not None: hidden_states = self.norm(hidden_states.permute(0, 2, 3, 1)).permute(0, 3, 1, 2) if self.use_conv and self.padding == 0: pad = (0, 1, 0, 1) hidden_states = F.pad(hidden_states, pad, mode="constant", value=0) assert hidden_states.shape[1] == self.channels if not USE_PEFT_BACKEND: if isinstance(self.conv, LoRACompatibleConv): hidden_states = self.conv(hidden_states, scale) else: hidden_states = self.conv(hidden_states) else: hidden_states = self.conv(hidden_states) return hidden_states class FirDownsample2D(nn.Module): """A 2D FIR downsampling layer with an optional convolution. Parameters: channels (`int`): number of channels in the inputs and outputs. use_conv (`bool`, default `False`): option to use a convolution. out_channels (`int`, optional): number of output channels. Defaults to `channels`. fir_kernel (`tuple`, default `(1, 3, 3, 1)`): kernel for the FIR filter. """ def __init__( self, channels: Optional[int] = None, out_channels: Optional[int] = None, use_conv: bool = False, fir_kernel: Tuple[int, int, int, int] = (1, 3, 3, 1), ): super().__init__() out_channels = out_channels if out_channels else channels if use_conv: self.Conv2d_0 = nn.Conv2d(channels, out_channels, kernel_size=3, stride=1, padding=1) self.fir_kernel = fir_kernel self.use_conv = use_conv self.out_channels = out_channels def _downsample_2d( self, hidden_states: torch.FloatTensor, weight: Optional[torch.FloatTensor] = None, kernel: Optional[torch.FloatTensor] = None, factor: int = 2, gain: float = 1, ) -> torch.FloatTensor: """Fused `Conv2d()` followed by `downsample_2d()`. Padding is performed only once at the beginning, not between the operations. The fused op is considerably more efficient than performing the same calculation using standard TensorFlow ops. It supports gradients of arbitrary order. Args: hidden_states (`torch.FloatTensor`): Input tensor of the shape `[N, C, H, W]` or `[N, H, W, C]`. weight (`torch.FloatTensor`, *optional*): Weight tensor of the shape `[filterH, filterW, inChannels, outChannels]`. Grouped convolution can be performed by `inChannels = x.shape[0] // numGroups`. kernel (`torch.FloatTensor`, *optional*): FIR filter of the shape `[firH, firW]` or `[firN]` (separable). The default is `[1] * factor`, which corresponds to average pooling. factor (`int`, *optional*, default to `2`): Integer downsampling factor. gain (`float`, *optional*, default to `1.0`): Scaling factor for signal magnitude. Returns: output (`torch.FloatTensor`): Tensor of the shape `[N, C, H // factor, W // factor]` or `[N, H // factor, W // factor, C]`, and same datatype as `x`. """ assert isinstance(factor, int) and factor >= 1 if kernel is None: kernel = [1] * factor # setup kernel kernel = torch.tensor(kernel, dtype=torch.float32) if kernel.ndim == 1: kernel = torch.outer(kernel, kernel) kernel /= torch.sum(kernel) kernel = kernel * gain if self.use_conv: _, _, convH, convW = weight.shape pad_value = (kernel.shape[0] - factor) + (convW - 1) stride_value = [factor, factor] upfirdn_input = upfirdn2d_native( hidden_states, torch.tensor(kernel, device=hidden_states.device), pad=((pad_value + 1) // 2, pad_value // 2), ) output = F.conv2d(upfirdn_input, weight, stride=stride_value, padding=0) else: pad_value = kernel.shape[0] - factor output = upfirdn2d_native( hidden_states, torch.tensor(kernel, device=hidden_states.device), down=factor, pad=((pad_value + 1) // 2, pad_value // 2), ) return output def forward(self, hidden_states: torch.FloatTensor) -> torch.FloatTensor: if self.use_conv: downsample_input = self._downsample_2d(hidden_states, weight=self.Conv2d_0.weight, kernel=self.fir_kernel) hidden_states = downsample_input + self.Conv2d_0.bias.reshape(1, -1, 1, 1) else: hidden_states = self._downsample_2d(hidden_states, kernel=self.fir_kernel, factor=2) return hidden_states # downsample/upsample layer used in k-upscaler, might be able to use FirDownsample2D/DirUpsample2D instead class KDownsample2D(nn.Module): r"""A 2D K-downsampling layer. Parameters: pad_mode (`str`, *optional*, default to `"reflect"`): the padding mode to use. """ def __init__(self, pad_mode: str = "reflect"): super().__init__() self.pad_mode = pad_mode kernel_1d = torch.tensor([[1 / 8, 3 / 8, 3 / 8, 1 / 8]]) self.pad = kernel_1d.shape[1] // 2 - 1 self.register_buffer("kernel", kernel_1d.T @ kernel_1d, persistent=False) def forward(self, inputs: torch.Tensor) -> torch.Tensor: inputs = F.pad(inputs, (self.pad,) * 4, self.pad_mode) weight = inputs.new_zeros( [ inputs.shape[1], inputs.shape[1], self.kernel.shape[0], self.kernel.shape[1], ] ) indices = torch.arange(inputs.shape[1], device=inputs.device) kernel = self.kernel.to(weight)[None, :].expand(inputs.shape[1], -1, -1) weight[indices, indices] = kernel return F.conv2d(inputs, weight, stride=2) def downsample_2d( hidden_states: torch.FloatTensor, kernel: Optional[torch.FloatTensor] = None, factor: int = 2, gain: float = 1, ) -> torch.FloatTensor: r"""Downsample2D a batch of 2D images with the given filter. Accepts a batch of 2D images of the shape `[N, C, H, W]` or `[N, H, W, C]` and downsamples each image with the given filter. The filter is normalized so that if the input pixels are constant, they will be scaled by the specified `gain`. Pixels outside the image are assumed to be zero, and the filter is padded with zeros so that its shape is a multiple of the downsampling factor. Args: hidden_states (`torch.FloatTensor`) Input tensor of the shape `[N, C, H, W]` or `[N, H, W, C]`. kernel (`torch.FloatTensor`, *optional*): FIR filter of the shape `[firH, firW]` or `[firN]` (separable). The default is `[1] * factor`, which corresponds to average pooling. factor (`int`, *optional*, default to `2`): Integer downsampling factor. gain (`float`, *optional*, default to `1.0`): Scaling factor for signal magnitude. Returns: output (`torch.FloatTensor`): Tensor of the shape `[N, C, H // factor, W // factor]` """ assert isinstance(factor, int) and factor >= 1 if kernel is None: kernel = [1] * factor kernel = torch.tensor(kernel, dtype=torch.float32) if kernel.ndim == 1: kernel = torch.outer(kernel, kernel) kernel /= torch.sum(kernel) kernel = kernel * gain pad_value = kernel.shape[0] - factor output = upfirdn2d_native( hidden_states, kernel.to(device=hidden_states.device), down=factor, pad=((pad_value + 1) // 2, pad_value // 2), ) return output

diffusers/src/diffusers/models/downsampling.py/0

# Copyright 2023 The HuggingFace Team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from typing import Any, Dict, Optional, Tuple, Union import numpy as np import torch import torch.nn.functional as F from torch import nn from ...utils import is_torch_version, logging from ...utils.torch_utils import apply_freeu from ..activations import get_activation from ..attention_processor import Attention, AttnAddedKVProcessor, AttnAddedKVProcessor2_0 from ..normalization import AdaGroupNorm from ..resnet import ( Downsample2D, FirDownsample2D, FirUpsample2D, KDownsample2D, KUpsample2D, ResnetBlock2D, ResnetBlockCondNorm2D, Upsample2D, ) from ..transformers.dual_transformer_2d import DualTransformer2DModel from ..transformers.transformer_2d import Transformer2DModel logger = logging.get_logger(__name__) # pylint: disable=invalid-name def get_down_block( down_block_type: str, num_layers: int, in_channels: int, out_channels: int, temb_channels: int, add_downsample: bool, resnet_eps: float, resnet_act_fn: str, transformer_layers_per_block: int = 1, num_attention_heads: Optional[int] = None, resnet_groups: Optional[int] = None, cross_attention_dim: Optional[int] = None, downsample_padding: Optional[int] = None, dual_cross_attention: bool = False, use_linear_projection: bool = False, only_cross_attention: bool = False, upcast_attention: bool = False, resnet_time_scale_shift: str = "default", attention_type: str = "default", resnet_skip_time_act: bool = False, resnet_out_scale_factor: float = 1.0, cross_attention_norm: Optional[str] = None, attention_head_dim: Optional[int] = None, downsample_type: Optional[str] = None, dropout: float = 0.0, ): # If attn head dim is not defined, we default it to the number of heads if attention_head_dim is None: logger.warn( f"It is recommended to provide `attention_head_dim` when calling `get_down_block`. Defaulting `attention_head_dim` to {num_attention_heads}." ) attention_head_dim = num_attention_heads down_block_type = down_block_type[7:] if down_block_type.startswith("UNetRes") else down_block_type if down_block_type == "DownBlock2D": return DownBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, dropout=dropout, add_downsample=add_downsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, downsample_padding=downsample_padding, resnet_time_scale_shift=resnet_time_scale_shift, ) elif down_block_type == "ResnetDownsampleBlock2D": return ResnetDownsampleBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, dropout=dropout, add_downsample=add_downsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, resnet_time_scale_shift=resnet_time_scale_shift, skip_time_act=resnet_skip_time_act, output_scale_factor=resnet_out_scale_factor, ) elif down_block_type == "AttnDownBlock2D": if add_downsample is False: downsample_type = None else: downsample_type = downsample_type or "conv" # default to 'conv' return AttnDownBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, dropout=dropout, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, downsample_padding=downsample_padding, attention_head_dim=attention_head_dim, resnet_time_scale_shift=resnet_time_scale_shift, downsample_type=downsample_type, ) elif down_block_type == "CrossAttnDownBlock2D": if cross_attention_dim is None: raise ValueError("cross_attention_dim must be specified for CrossAttnDownBlock2D") return CrossAttnDownBlock2D( num_layers=num_layers, transformer_layers_per_block=transformer_layers_per_block, in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, dropout=dropout, add_downsample=add_downsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, downsample_padding=downsample_padding, cross_attention_dim=cross_attention_dim, num_attention_heads=num_attention_heads, dual_cross_attention=dual_cross_attention, use_linear_projection=use_linear_projection, only_cross_attention=only_cross_attention, upcast_attention=upcast_attention, resnet_time_scale_shift=resnet_time_scale_shift, attention_type=attention_type, ) elif down_block_type == "SimpleCrossAttnDownBlock2D": if cross_attention_dim is None: raise ValueError("cross_attention_dim must be specified for SimpleCrossAttnDownBlock2D") return SimpleCrossAttnDownBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, dropout=dropout, add_downsample=add_downsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, cross_attention_dim=cross_attention_dim, attention_head_dim=attention_head_dim, resnet_time_scale_shift=resnet_time_scale_shift, skip_time_act=resnet_skip_time_act, output_scale_factor=resnet_out_scale_factor, only_cross_attention=only_cross_attention, cross_attention_norm=cross_attention_norm, ) elif down_block_type == "SkipDownBlock2D": return SkipDownBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, dropout=dropout, add_downsample=add_downsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, downsample_padding=downsample_padding, resnet_time_scale_shift=resnet_time_scale_shift, ) elif down_block_type == "AttnSkipDownBlock2D": return AttnSkipDownBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, dropout=dropout, add_downsample=add_downsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, attention_head_dim=attention_head_dim, resnet_time_scale_shift=resnet_time_scale_shift, ) elif down_block_type == "DownEncoderBlock2D": return DownEncoderBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, dropout=dropout, add_downsample=add_downsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, downsample_padding=downsample_padding, resnet_time_scale_shift=resnet_time_scale_shift, ) elif down_block_type == "AttnDownEncoderBlock2D": return AttnDownEncoderBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, dropout=dropout, add_downsample=add_downsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, downsample_padding=downsample_padding, attention_head_dim=attention_head_dim, resnet_time_scale_shift=resnet_time_scale_shift, ) elif down_block_type == "KDownBlock2D": return KDownBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, dropout=dropout, add_downsample=add_downsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, ) elif down_block_type == "KCrossAttnDownBlock2D": return KCrossAttnDownBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, dropout=dropout, add_downsample=add_downsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, cross_attention_dim=cross_attention_dim, attention_head_dim=attention_head_dim, add_self_attention=True if not add_downsample else False, ) raise ValueError(f"{down_block_type} does not exist.") def get_up_block( up_block_type: str, num_layers: int, in_channels: int, out_channels: int, prev_output_channel: int, temb_channels: int, add_upsample: bool, resnet_eps: float, resnet_act_fn: str, resolution_idx: Optional[int] = None, transformer_layers_per_block: int = 1, num_attention_heads: Optional[int] = None, resnet_groups: Optional[int] = None, cross_attention_dim: Optional[int] = None, dual_cross_attention: bool = False, use_linear_projection: bool = False, only_cross_attention: bool = False, upcast_attention: bool = False, resnet_time_scale_shift: str = "default", attention_type: str = "default", resnet_skip_time_act: bool = False, resnet_out_scale_factor: float = 1.0, cross_attention_norm: Optional[str] = None, attention_head_dim: Optional[int] = None, upsample_type: Optional[str] = None, dropout: float = 0.0, ) -> nn.Module: # If attn head dim is not defined, we default it to the number of heads if attention_head_dim is None: logger.warn( f"It is recommended to provide `attention_head_dim` when calling `get_up_block`. Defaulting `attention_head_dim` to {num_attention_heads}." ) attention_head_dim = num_attention_heads up_block_type = up_block_type[7:] if up_block_type.startswith("UNetRes") else up_block_type if up_block_type == "UpBlock2D": return UpBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, prev_output_channel=prev_output_channel, temb_channels=temb_channels, resolution_idx=resolution_idx, dropout=dropout, add_upsample=add_upsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, resnet_time_scale_shift=resnet_time_scale_shift, ) elif up_block_type == "ResnetUpsampleBlock2D": return ResnetUpsampleBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, prev_output_channel=prev_output_channel, temb_channels=temb_channels, resolution_idx=resolution_idx, dropout=dropout, add_upsample=add_upsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, resnet_time_scale_shift=resnet_time_scale_shift, skip_time_act=resnet_skip_time_act, output_scale_factor=resnet_out_scale_factor, ) elif up_block_type == "CrossAttnUpBlock2D": if cross_attention_dim is None: raise ValueError("cross_attention_dim must be specified for CrossAttnUpBlock2D") return CrossAttnUpBlock2D( num_layers=num_layers, transformer_layers_per_block=transformer_layers_per_block, in_channels=in_channels, out_channels=out_channels, prev_output_channel=prev_output_channel, temb_channels=temb_channels, resolution_idx=resolution_idx, dropout=dropout, add_upsample=add_upsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, cross_attention_dim=cross_attention_dim, num_attention_heads=num_attention_heads, dual_cross_attention=dual_cross_attention, use_linear_projection=use_linear_projection, only_cross_attention=only_cross_attention, upcast_attention=upcast_attention, resnet_time_scale_shift=resnet_time_scale_shift, attention_type=attention_type, ) elif up_block_type == "SimpleCrossAttnUpBlock2D": if cross_attention_dim is None: raise ValueError("cross_attention_dim must be specified for SimpleCrossAttnUpBlock2D") return SimpleCrossAttnUpBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, prev_output_channel=prev_output_channel, temb_channels=temb_channels, resolution_idx=resolution_idx, dropout=dropout, add_upsample=add_upsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, cross_attention_dim=cross_attention_dim, attention_head_dim=attention_head_dim, resnet_time_scale_shift=resnet_time_scale_shift, skip_time_act=resnet_skip_time_act, output_scale_factor=resnet_out_scale_factor, only_cross_attention=only_cross_attention, cross_attention_norm=cross_attention_norm, ) elif up_block_type == "AttnUpBlock2D": if add_upsample is False: upsample_type = None else: upsample_type = upsample_type or "conv" # default to 'conv' return AttnUpBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, prev_output_channel=prev_output_channel, temb_channels=temb_channels, resolution_idx=resolution_idx, dropout=dropout, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, attention_head_dim=attention_head_dim, resnet_time_scale_shift=resnet_time_scale_shift, upsample_type=upsample_type, ) elif up_block_type == "SkipUpBlock2D": return SkipUpBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, prev_output_channel=prev_output_channel, temb_channels=temb_channels, resolution_idx=resolution_idx, dropout=dropout, add_upsample=add_upsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_time_scale_shift=resnet_time_scale_shift, ) elif up_block_type == "AttnSkipUpBlock2D": return AttnSkipUpBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, prev_output_channel=prev_output_channel, temb_channels=temb_channels, resolution_idx=resolution_idx, dropout=dropout, add_upsample=add_upsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, attention_head_dim=attention_head_dim, resnet_time_scale_shift=resnet_time_scale_shift, ) elif up_block_type == "UpDecoderBlock2D": return UpDecoderBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, resolution_idx=resolution_idx, dropout=dropout, add_upsample=add_upsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, resnet_time_scale_shift=resnet_time_scale_shift, temb_channels=temb_channels, ) elif up_block_type == "AttnUpDecoderBlock2D": return AttnUpDecoderBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, resolution_idx=resolution_idx, dropout=dropout, add_upsample=add_upsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, resnet_groups=resnet_groups, attention_head_dim=attention_head_dim, resnet_time_scale_shift=resnet_time_scale_shift, temb_channels=temb_channels, ) elif up_block_type == "KUpBlock2D": return KUpBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, resolution_idx=resolution_idx, dropout=dropout, add_upsample=add_upsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, ) elif up_block_type == "KCrossAttnUpBlock2D": return KCrossAttnUpBlock2D( num_layers=num_layers, in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, resolution_idx=resolution_idx, dropout=dropout, add_upsample=add_upsample, resnet_eps=resnet_eps, resnet_act_fn=resnet_act_fn, cross_attention_dim=cross_attention_dim, attention_head_dim=attention_head_dim, ) raise ValueError(f"{up_block_type} does not exist.") class AutoencoderTinyBlock(nn.Module): """ Tiny Autoencoder block used in [`AutoencoderTiny`]. It is a mini residual module consisting of plain conv + ReLU blocks. Args: in_channels (`int`): The number of input channels. out_channels (`int`): The number of output channels. act_fn (`str`): ` The activation function to use. Supported values are `"swish"`, `"mish"`, `"gelu"`, and `"relu"`. Returns: `torch.FloatTensor`: A tensor with the same shape as the input tensor, but with the number of channels equal to `out_channels`. """ def __init__(self, in_channels: int, out_channels: int, act_fn: str): super().__init__() act_fn = get_activation(act_fn) self.conv = nn.Sequential( nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1), act_fn, nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1), act_fn, nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1), ) self.skip = ( nn.Conv2d(in_channels, out_channels, kernel_size=1, bias=False) if in_channels != out_channels else nn.Identity() ) self.fuse = nn.ReLU() def forward(self, x: torch.FloatTensor) -> torch.FloatTensor: return self.fuse(self.conv(x) + self.skip(x)) class UNetMidBlock2D(nn.Module): """ A 2D UNet mid-block [`UNetMidBlock2D`] with multiple residual blocks and optional attention blocks. Args: in_channels (`int`): The number of input channels. temb_channels (`int`): The number of temporal embedding channels. dropout (`float`, *optional*, defaults to 0.0): The dropout rate. num_layers (`int`, *optional*, defaults to 1): The number of residual blocks. resnet_eps (`float`, *optional*, 1e-6 ): The epsilon value for the resnet blocks. resnet_time_scale_shift (`str`, *optional*, defaults to `default`): The type of normalization to apply to the time embeddings. This can help to improve the performance of the model on tasks with long-range temporal dependencies. resnet_act_fn (`str`, *optional*, defaults to `swish`): The activation function for the resnet blocks. resnet_groups (`int`, *optional*, defaults to 32): The number of groups to use in the group normalization layers of the resnet blocks. attn_groups (`Optional[int]`, *optional*, defaults to None): The number of groups for the attention blocks. resnet_pre_norm (`bool`, *optional*, defaults to `True`): Whether to use pre-normalization for the resnet blocks. add_attention (`bool`, *optional*, defaults to `True`): Whether to add attention blocks. attention_head_dim (`int`, *optional*, defaults to 1): Dimension of a single attention head. The number of attention heads is determined based on this value and the number of input channels. output_scale_factor (`float`, *optional*, defaults to 1.0): The output scale factor. Returns: `torch.FloatTensor`: The output of the last residual block, which is a tensor of shape `(batch_size, in_channels, height, width)`. """ def __init__( self, in_channels: int, temb_channels: int, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", # default, spatial resnet_act_fn: str = "swish", resnet_groups: int = 32, attn_groups: Optional[int] = None, resnet_pre_norm: bool = True, add_attention: bool = True, attention_head_dim: int = 1, output_scale_factor: float = 1.0, ): super().__init__() resnet_groups = resnet_groups if resnet_groups is not None else min(in_channels // 4, 32) self.add_attention = add_attention if attn_groups is None: attn_groups = resnet_groups if resnet_time_scale_shift == "default" else None # there is always at least one resnet if resnet_time_scale_shift == "spatial": resnets = [ ResnetBlockCondNorm2D( in_channels=in_channels, out_channels=in_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm="spatial", non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, ) ] else: resnets = [ ResnetBlock2D( in_channels=in_channels, out_channels=in_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ] attentions = [] if attention_head_dim is None: logger.warn( f"It is not recommend to pass `attention_head_dim=None`. Defaulting `attention_head_dim` to `in_channels`: {in_channels}." ) attention_head_dim = in_channels for _ in range(num_layers): if self.add_attention: attentions.append( Attention( in_channels, heads=in_channels // attention_head_dim, dim_head=attention_head_dim, rescale_output_factor=output_scale_factor, eps=resnet_eps, norm_num_groups=attn_groups, spatial_norm_dim=temb_channels if resnet_time_scale_shift == "spatial" else None, residual_connection=True, bias=True, upcast_softmax=True, _from_deprecated_attn_block=True, ) ) else: attentions.append(None) if resnet_time_scale_shift == "spatial": resnets.append( ResnetBlockCondNorm2D( in_channels=in_channels, out_channels=in_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm="spatial", non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, ) ) else: resnets.append( ResnetBlock2D( in_channels=in_channels, out_channels=in_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) self.attentions = nn.ModuleList(attentions) self.resnets = nn.ModuleList(resnets) def forward(self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None) -> torch.FloatTensor: hidden_states = self.resnets[0](hidden_states, temb) for attn, resnet in zip(self.attentions, self.resnets[1:]): if attn is not None: hidden_states = attn(hidden_states, temb=temb) hidden_states = resnet(hidden_states, temb) return hidden_states class UNetMidBlock2DCrossAttn(nn.Module): def __init__( self, in_channels: int, temb_channels: int, dropout: float = 0.0, num_layers: int = 1, transformer_layers_per_block: Union[int, Tuple[int]] = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, num_attention_heads: int = 1, output_scale_factor: float = 1.0, cross_attention_dim: int = 1280, dual_cross_attention: bool = False, use_linear_projection: bool = False, upcast_attention: bool = False, attention_type: str = "default", ): super().__init__() self.has_cross_attention = True self.num_attention_heads = num_attention_heads resnet_groups = resnet_groups if resnet_groups is not None else min(in_channels // 4, 32) # support for variable transformer layers per block if isinstance(transformer_layers_per_block, int): transformer_layers_per_block = [transformer_layers_per_block] * num_layers # there is always at least one resnet resnets = [ ResnetBlock2D( in_channels=in_channels, out_channels=in_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ] attentions = [] for i in range(num_layers): if not dual_cross_attention: attentions.append( Transformer2DModel( num_attention_heads, in_channels // num_attention_heads, in_channels=in_channels, num_layers=transformer_layers_per_block[i], cross_attention_dim=cross_attention_dim, norm_num_groups=resnet_groups, use_linear_projection=use_linear_projection, upcast_attention=upcast_attention, attention_type=attention_type, ) ) else: attentions.append( DualTransformer2DModel( num_attention_heads, in_channels // num_attention_heads, in_channels=in_channels, num_layers=1, cross_attention_dim=cross_attention_dim, norm_num_groups=resnet_groups, ) ) resnets.append( ResnetBlock2D( in_channels=in_channels, out_channels=in_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) self.attentions = nn.ModuleList(attentions) self.resnets = nn.ModuleList(resnets) self.gradient_checkpointing = False def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, encoder_hidden_states: Optional[torch.FloatTensor] = None, attention_mask: Optional[torch.FloatTensor] = None, cross_attention_kwargs: Optional[Dict[str, Any]] = None, encoder_attention_mask: Optional[torch.FloatTensor] = None, ) -> torch.FloatTensor: lora_scale = cross_attention_kwargs.get("scale", 1.0) if cross_attention_kwargs is not None else 1.0 hidden_states = self.resnets[0](hidden_states, temb, scale=lora_scale) for attn, resnet in zip(self.attentions, self.resnets[1:]): if self.training and self.gradient_checkpointing: def create_custom_forward(module, return_dict=None): def custom_forward(*inputs): if return_dict is not None: return module(*inputs, return_dict=return_dict) else: return module(*inputs) return custom_forward ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {} hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, cross_attention_kwargs=cross_attention_kwargs, attention_mask=attention_mask, encoder_attention_mask=encoder_attention_mask, return_dict=False, )[0] hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb, **ckpt_kwargs, ) else: hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, cross_attention_kwargs=cross_attention_kwargs, attention_mask=attention_mask, encoder_attention_mask=encoder_attention_mask, return_dict=False, )[0] hidden_states = resnet(hidden_states, temb, scale=lora_scale) return hidden_states class UNetMidBlock2DSimpleCrossAttn(nn.Module): def __init__( self, in_channels: int, temb_channels: int, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, attention_head_dim: int = 1, output_scale_factor: float = 1.0, cross_attention_dim: int = 1280, skip_time_act: bool = False, only_cross_attention: bool = False, cross_attention_norm: Optional[str] = None, ): super().__init__() self.has_cross_attention = True self.attention_head_dim = attention_head_dim resnet_groups = resnet_groups if resnet_groups is not None else min(in_channels // 4, 32) self.num_heads = in_channels // self.attention_head_dim # there is always at least one resnet resnets = [ ResnetBlock2D( in_channels=in_channels, out_channels=in_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, skip_time_act=skip_time_act, ) ] attentions = [] for _ in range(num_layers): processor = ( AttnAddedKVProcessor2_0() if hasattr(F, "scaled_dot_product_attention") else AttnAddedKVProcessor() ) attentions.append( Attention( query_dim=in_channels, cross_attention_dim=in_channels, heads=self.num_heads, dim_head=self.attention_head_dim, added_kv_proj_dim=cross_attention_dim, norm_num_groups=resnet_groups, bias=True, upcast_softmax=True, only_cross_attention=only_cross_attention, cross_attention_norm=cross_attention_norm, processor=processor, ) ) resnets.append( ResnetBlock2D( in_channels=in_channels, out_channels=in_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, skip_time_act=skip_time_act, ) ) self.attentions = nn.ModuleList(attentions) self.resnets = nn.ModuleList(resnets) def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, encoder_hidden_states: Optional[torch.FloatTensor] = None, attention_mask: Optional[torch.FloatTensor] = None, cross_attention_kwargs: Optional[Dict[str, Any]] = None, encoder_attention_mask: Optional[torch.FloatTensor] = None, ) -> torch.FloatTensor: cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {} lora_scale = cross_attention_kwargs.get("scale", 1.0) if attention_mask is None: # if encoder_hidden_states is defined: we are doing cross-attn, so we should use cross-attn mask. mask = None if encoder_hidden_states is None else encoder_attention_mask else: # when attention_mask is defined: we don't even check for encoder_attention_mask. # this is to maintain compatibility with UnCLIP, which uses 'attention_mask' param for cross-attn masks. # TODO: UnCLIP should express cross-attn mask via encoder_attention_mask param instead of via attention_mask. # then we can simplify this whole if/else block to: # mask = attention_mask if encoder_hidden_states is None else encoder_attention_mask mask = attention_mask hidden_states = self.resnets[0](hidden_states, temb, scale=lora_scale) for attn, resnet in zip(self.attentions, self.resnets[1:]): # attn hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, attention_mask=mask, **cross_attention_kwargs, ) # resnet hidden_states = resnet(hidden_states, temb, scale=lora_scale) return hidden_states class AttnDownBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, temb_channels: int, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, attention_head_dim: int = 1, output_scale_factor: float = 1.0, downsample_padding: int = 1, downsample_type: str = "conv", ): super().__init__() resnets = [] attentions = [] self.downsample_type = downsample_type if attention_head_dim is None: logger.warn( f"It is not recommend to pass `attention_head_dim=None`. Defaulting `attention_head_dim` to `in_channels`: {out_channels}." ) attention_head_dim = out_channels for i in range(num_layers): in_channels = in_channels if i == 0 else out_channels resnets.append( ResnetBlock2D( in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) attentions.append( Attention( out_channels, heads=out_channels // attention_head_dim, dim_head=attention_head_dim, rescale_output_factor=output_scale_factor, eps=resnet_eps, norm_num_groups=resnet_groups, residual_connection=True, bias=True, upcast_softmax=True, _from_deprecated_attn_block=True, ) ) self.attentions = nn.ModuleList(attentions) self.resnets = nn.ModuleList(resnets) if downsample_type == "conv": self.downsamplers = nn.ModuleList( [ Downsample2D( out_channels, use_conv=True, out_channels=out_channels, padding=downsample_padding, name="op" ) ] ) elif downsample_type == "resnet": self.downsamplers = nn.ModuleList( [ ResnetBlock2D( in_channels=out_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, down=True, ) ] ) else: self.downsamplers = None def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, upsample_size: Optional[int] = None, cross_attention_kwargs: Optional[Dict[str, Any]] = None, ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]: cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {} lora_scale = cross_attention_kwargs.get("scale", 1.0) output_states = () for resnet, attn in zip(self.resnets, self.attentions): cross_attention_kwargs.update({"scale": lora_scale}) hidden_states = resnet(hidden_states, temb, scale=lora_scale) hidden_states = attn(hidden_states, **cross_attention_kwargs) output_states = output_states + (hidden_states,) if self.downsamplers is not None: for downsampler in self.downsamplers: if self.downsample_type == "resnet": hidden_states = downsampler(hidden_states, temb=temb, scale=lora_scale) else: hidden_states = downsampler(hidden_states, scale=lora_scale) output_states += (hidden_states,) return hidden_states, output_states class CrossAttnDownBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, temb_channels: int, dropout: float = 0.0, num_layers: int = 1, transformer_layers_per_block: Union[int, Tuple[int]] = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, num_attention_heads: int = 1, cross_attention_dim: int = 1280, output_scale_factor: float = 1.0, downsample_padding: int = 1, add_downsample: bool = True, dual_cross_attention: bool = False, use_linear_projection: bool = False, only_cross_attention: bool = False, upcast_attention: bool = False, attention_type: str = "default", ): super().__init__() resnets = [] attentions = [] self.has_cross_attention = True self.num_attention_heads = num_attention_heads if isinstance(transformer_layers_per_block, int): transformer_layers_per_block = [transformer_layers_per_block] * num_layers for i in range(num_layers): in_channels = in_channels if i == 0 else out_channels resnets.append( ResnetBlock2D( in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) if not dual_cross_attention: attentions.append( Transformer2DModel( num_attention_heads, out_channels // num_attention_heads, in_channels=out_channels, num_layers=transformer_layers_per_block[i], cross_attention_dim=cross_attention_dim, norm_num_groups=resnet_groups, use_linear_projection=use_linear_projection, only_cross_attention=only_cross_attention, upcast_attention=upcast_attention, attention_type=attention_type, ) ) else: attentions.append( DualTransformer2DModel( num_attention_heads, out_channels // num_attention_heads, in_channels=out_channels, num_layers=1, cross_attention_dim=cross_attention_dim, norm_num_groups=resnet_groups, ) ) self.attentions = nn.ModuleList(attentions) self.resnets = nn.ModuleList(resnets) if add_downsample: self.downsamplers = nn.ModuleList( [ Downsample2D( out_channels, use_conv=True, out_channels=out_channels, padding=downsample_padding, name="op" ) ] ) else: self.downsamplers = None self.gradient_checkpointing = False def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, encoder_hidden_states: Optional[torch.FloatTensor] = None, attention_mask: Optional[torch.FloatTensor] = None, cross_attention_kwargs: Optional[Dict[str, Any]] = None, encoder_attention_mask: Optional[torch.FloatTensor] = None, additional_residuals: Optional[torch.FloatTensor] = None, ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]: output_states = () lora_scale = cross_attention_kwargs.get("scale", 1.0) if cross_attention_kwargs is not None else 1.0 blocks = list(zip(self.resnets, self.attentions)) for i, (resnet, attn) in enumerate(blocks): if self.training and self.gradient_checkpointing: def create_custom_forward(module, return_dict=None): def custom_forward(*inputs): if return_dict is not None: return module(*inputs, return_dict=return_dict) else: return module(*inputs) return custom_forward ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {} hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb, **ckpt_kwargs, ) hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, cross_attention_kwargs=cross_attention_kwargs, attention_mask=attention_mask, encoder_attention_mask=encoder_attention_mask, return_dict=False, )[0] else: hidden_states = resnet(hidden_states, temb, scale=lora_scale) hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, cross_attention_kwargs=cross_attention_kwargs, attention_mask=attention_mask, encoder_attention_mask=encoder_attention_mask, return_dict=False, )[0] # apply additional residuals to the output of the last pair of resnet and attention blocks if i == len(blocks) - 1 and additional_residuals is not None: hidden_states = hidden_states + additional_residuals output_states = output_states + (hidden_states,) if self.downsamplers is not None: for downsampler in self.downsamplers: hidden_states = downsampler(hidden_states, scale=lora_scale) output_states = output_states + (hidden_states,) return hidden_states, output_states class DownBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, temb_channels: int, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, output_scale_factor: float = 1.0, add_downsample: bool = True, downsample_padding: int = 1, ): super().__init__() resnets = [] for i in range(num_layers): in_channels = in_channels if i == 0 else out_channels resnets.append( ResnetBlock2D( in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) self.resnets = nn.ModuleList(resnets) if add_downsample: self.downsamplers = nn.ModuleList( [ Downsample2D( out_channels, use_conv=True, out_channels=out_channels, padding=downsample_padding, name="op" ) ] ) else: self.downsamplers = None self.gradient_checkpointing = False def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, scale: float = 1.0 ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]: output_states = () for resnet in self.resnets: if self.training and self.gradient_checkpointing: def create_custom_forward(module): def custom_forward(*inputs): return module(*inputs) return custom_forward if is_torch_version(">=", "1.11.0"): hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb, use_reentrant=False ) else: hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb ) else: hidden_states = resnet(hidden_states, temb, scale=scale) output_states = output_states + (hidden_states,) if self.downsamplers is not None: for downsampler in self.downsamplers: hidden_states = downsampler(hidden_states, scale=scale) output_states = output_states + (hidden_states,) return hidden_states, output_states class DownEncoderBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, output_scale_factor: float = 1.0, add_downsample: bool = True, downsample_padding: int = 1, ): super().__init__() resnets = [] for i in range(num_layers): in_channels = in_channels if i == 0 else out_channels if resnet_time_scale_shift == "spatial": resnets.append( ResnetBlockCondNorm2D( in_channels=in_channels, out_channels=out_channels, temb_channels=None, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm="spatial", non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, ) ) else: resnets.append( ResnetBlock2D( in_channels=in_channels, out_channels=out_channels, temb_channels=None, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) self.resnets = nn.ModuleList(resnets) if add_downsample: self.downsamplers = nn.ModuleList( [ Downsample2D( out_channels, use_conv=True, out_channels=out_channels, padding=downsample_padding, name="op" ) ] ) else: self.downsamplers = None def forward(self, hidden_states: torch.FloatTensor, scale: float = 1.0) -> torch.FloatTensor: for resnet in self.resnets: hidden_states = resnet(hidden_states, temb=None, scale=scale) if self.downsamplers is not None: for downsampler in self.downsamplers: hidden_states = downsampler(hidden_states, scale) return hidden_states class AttnDownEncoderBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, attention_head_dim: int = 1, output_scale_factor: float = 1.0, add_downsample: bool = True, downsample_padding: int = 1, ): super().__init__() resnets = [] attentions = [] if attention_head_dim is None: logger.warn( f"It is not recommend to pass `attention_head_dim=None`. Defaulting `attention_head_dim` to `in_channels`: {out_channels}." ) attention_head_dim = out_channels for i in range(num_layers): in_channels = in_channels if i == 0 else out_channels if resnet_time_scale_shift == "spatial": resnets.append( ResnetBlockCondNorm2D( in_channels=in_channels, out_channels=out_channels, temb_channels=None, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm="spatial", non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, ) ) else: resnets.append( ResnetBlock2D( in_channels=in_channels, out_channels=out_channels, temb_channels=None, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) attentions.append( Attention( out_channels, heads=out_channels // attention_head_dim, dim_head=attention_head_dim, rescale_output_factor=output_scale_factor, eps=resnet_eps, norm_num_groups=resnet_groups, residual_connection=True, bias=True, upcast_softmax=True, _from_deprecated_attn_block=True, ) ) self.attentions = nn.ModuleList(attentions) self.resnets = nn.ModuleList(resnets) if add_downsample: self.downsamplers = nn.ModuleList( [ Downsample2D( out_channels, use_conv=True, out_channels=out_channels, padding=downsample_padding, name="op" ) ] ) else: self.downsamplers = None def forward(self, hidden_states: torch.FloatTensor, scale: float = 1.0) -> torch.FloatTensor: for resnet, attn in zip(self.resnets, self.attentions): hidden_states = resnet(hidden_states, temb=None, scale=scale) cross_attention_kwargs = {"scale": scale} hidden_states = attn(hidden_states, **cross_attention_kwargs) if self.downsamplers is not None: for downsampler in self.downsamplers: hidden_states = downsampler(hidden_states, scale) return hidden_states class AttnSkipDownBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, temb_channels: int, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_pre_norm: bool = True, attention_head_dim: int = 1, output_scale_factor: float = np.sqrt(2.0), add_downsample: bool = True, ): super().__init__() self.attentions = nn.ModuleList([]) self.resnets = nn.ModuleList([]) if attention_head_dim is None: logger.warn( f"It is not recommend to pass `attention_head_dim=None`. Defaulting `attention_head_dim` to `in_channels`: {out_channels}." ) attention_head_dim = out_channels for i in range(num_layers): in_channels = in_channels if i == 0 else out_channels self.resnets.append( ResnetBlock2D( in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=min(in_channels // 4, 32), groups_out=min(out_channels // 4, 32), dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) self.attentions.append( Attention( out_channels, heads=out_channels // attention_head_dim, dim_head=attention_head_dim, rescale_output_factor=output_scale_factor, eps=resnet_eps, norm_num_groups=32, residual_connection=True, bias=True, upcast_softmax=True, _from_deprecated_attn_block=True, ) ) if add_downsample: self.resnet_down = ResnetBlock2D( in_channels=out_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=min(out_channels // 4, 32), dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, use_in_shortcut=True, down=True, kernel="fir", ) self.downsamplers = nn.ModuleList([FirDownsample2D(out_channels, out_channels=out_channels)]) self.skip_conv = nn.Conv2d(3, out_channels, kernel_size=(1, 1), stride=(1, 1)) else: self.resnet_down = None self.downsamplers = None self.skip_conv = None def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, skip_sample: Optional[torch.FloatTensor] = None, scale: float = 1.0, ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...], torch.FloatTensor]: output_states = () for resnet, attn in zip(self.resnets, self.attentions): hidden_states = resnet(hidden_states, temb, scale=scale) cross_attention_kwargs = {"scale": scale} hidden_states = attn(hidden_states, **cross_attention_kwargs) output_states += (hidden_states,) if self.downsamplers is not None: hidden_states = self.resnet_down(hidden_states, temb, scale=scale) for downsampler in self.downsamplers: skip_sample = downsampler(skip_sample) hidden_states = self.skip_conv(skip_sample) + hidden_states output_states += (hidden_states,) return hidden_states, output_states, skip_sample class SkipDownBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, temb_channels: int, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_pre_norm: bool = True, output_scale_factor: float = np.sqrt(2.0), add_downsample: bool = True, downsample_padding: int = 1, ): super().__init__() self.resnets = nn.ModuleList([]) for i in range(num_layers): in_channels = in_channels if i == 0 else out_channels self.resnets.append( ResnetBlock2D( in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=min(in_channels // 4, 32), groups_out=min(out_channels // 4, 32), dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) if add_downsample: self.resnet_down = ResnetBlock2D( in_channels=out_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=min(out_channels // 4, 32), dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, use_in_shortcut=True, down=True, kernel="fir", ) self.downsamplers = nn.ModuleList([FirDownsample2D(out_channels, out_channels=out_channels)]) self.skip_conv = nn.Conv2d(3, out_channels, kernel_size=(1, 1), stride=(1, 1)) else: self.resnet_down = None self.downsamplers = None self.skip_conv = None def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, skip_sample: Optional[torch.FloatTensor] = None, scale: float = 1.0, ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...], torch.FloatTensor]: output_states = () for resnet in self.resnets: hidden_states = resnet(hidden_states, temb, scale) output_states += (hidden_states,) if self.downsamplers is not None: hidden_states = self.resnet_down(hidden_states, temb, scale) for downsampler in self.downsamplers: skip_sample = downsampler(skip_sample) hidden_states = self.skip_conv(skip_sample) + hidden_states output_states += (hidden_states,) return hidden_states, output_states, skip_sample class ResnetDownsampleBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, temb_channels: int, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, output_scale_factor: float = 1.0, add_downsample: bool = True, skip_time_act: bool = False, ): super().__init__() resnets = [] for i in range(num_layers): in_channels = in_channels if i == 0 else out_channels resnets.append( ResnetBlock2D( in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, skip_time_act=skip_time_act, ) ) self.resnets = nn.ModuleList(resnets) if add_downsample: self.downsamplers = nn.ModuleList( [ ResnetBlock2D( in_channels=out_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, skip_time_act=skip_time_act, down=True, ) ] ) else: self.downsamplers = None self.gradient_checkpointing = False def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, scale: float = 1.0 ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]: output_states = () for resnet in self.resnets: if self.training and self.gradient_checkpointing: def create_custom_forward(module): def custom_forward(*inputs): return module(*inputs) return custom_forward if is_torch_version(">=", "1.11.0"): hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb, use_reentrant=False ) else: hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb ) else: hidden_states = resnet(hidden_states, temb, scale) output_states = output_states + (hidden_states,) if self.downsamplers is not None: for downsampler in self.downsamplers: hidden_states = downsampler(hidden_states, temb, scale) output_states = output_states + (hidden_states,) return hidden_states, output_states class SimpleCrossAttnDownBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, temb_channels: int, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, attention_head_dim: int = 1, cross_attention_dim: int = 1280, output_scale_factor: float = 1.0, add_downsample: bool = True, skip_time_act: bool = False, only_cross_attention: bool = False, cross_attention_norm: Optional[str] = None, ): super().__init__() self.has_cross_attention = True resnets = [] attentions = [] self.attention_head_dim = attention_head_dim self.num_heads = out_channels // self.attention_head_dim for i in range(num_layers): in_channels = in_channels if i == 0 else out_channels resnets.append( ResnetBlock2D( in_channels=in_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, skip_time_act=skip_time_act, ) ) processor = ( AttnAddedKVProcessor2_0() if hasattr(F, "scaled_dot_product_attention") else AttnAddedKVProcessor() ) attentions.append( Attention( query_dim=out_channels, cross_attention_dim=out_channels, heads=self.num_heads, dim_head=attention_head_dim, added_kv_proj_dim=cross_attention_dim, norm_num_groups=resnet_groups, bias=True, upcast_softmax=True, only_cross_attention=only_cross_attention, cross_attention_norm=cross_attention_norm, processor=processor, ) ) self.attentions = nn.ModuleList(attentions) self.resnets = nn.ModuleList(resnets) if add_downsample: self.downsamplers = nn.ModuleList( [ ResnetBlock2D( in_channels=out_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, skip_time_act=skip_time_act, down=True, ) ] ) else: self.downsamplers = None self.gradient_checkpointing = False def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, encoder_hidden_states: Optional[torch.FloatTensor] = None, attention_mask: Optional[torch.FloatTensor] = None, cross_attention_kwargs: Optional[Dict[str, Any]] = None, encoder_attention_mask: Optional[torch.FloatTensor] = None, ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]: output_states = () cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {} lora_scale = cross_attention_kwargs.get("scale", 1.0) if attention_mask is None: # if encoder_hidden_states is defined: we are doing cross-attn, so we should use cross-attn mask. mask = None if encoder_hidden_states is None else encoder_attention_mask else: # when attention_mask is defined: we don't even check for encoder_attention_mask. # this is to maintain compatibility with UnCLIP, which uses 'attention_mask' param for cross-attn masks. # TODO: UnCLIP should express cross-attn mask via encoder_attention_mask param instead of via attention_mask. # then we can simplify this whole if/else block to: # mask = attention_mask if encoder_hidden_states is None else encoder_attention_mask mask = attention_mask for resnet, attn in zip(self.resnets, self.attentions): if self.training and self.gradient_checkpointing: def create_custom_forward(module, return_dict=None): def custom_forward(*inputs): if return_dict is not None: return module(*inputs, return_dict=return_dict) else: return module(*inputs) return custom_forward hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb) hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, attention_mask=mask, **cross_attention_kwargs, ) else: hidden_states = resnet(hidden_states, temb, scale=lora_scale) hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, attention_mask=mask, **cross_attention_kwargs, ) output_states = output_states + (hidden_states,) if self.downsamplers is not None: for downsampler in self.downsamplers: hidden_states = downsampler(hidden_states, temb, scale=lora_scale) output_states = output_states + (hidden_states,) return hidden_states, output_states class KDownBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, temb_channels: int, dropout: float = 0.0, num_layers: int = 4, resnet_eps: float = 1e-5, resnet_act_fn: str = "gelu", resnet_group_size: int = 32, add_downsample: bool = False, ): super().__init__() resnets = [] for i in range(num_layers): in_channels = in_channels if i == 0 else out_channels groups = in_channels // resnet_group_size groups_out = out_channels // resnet_group_size resnets.append( ResnetBlockCondNorm2D( in_channels=in_channels, out_channels=out_channels, dropout=dropout, temb_channels=temb_channels, groups=groups, groups_out=groups_out, eps=resnet_eps, non_linearity=resnet_act_fn, time_embedding_norm="ada_group", conv_shortcut_bias=False, ) ) self.resnets = nn.ModuleList(resnets) if add_downsample: # YiYi's comments- might be able to use FirDownsample2D, look into details later self.downsamplers = nn.ModuleList([KDownsample2D()]) else: self.downsamplers = None self.gradient_checkpointing = False def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, scale: float = 1.0 ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]: output_states = () for resnet in self.resnets: if self.training and self.gradient_checkpointing: def create_custom_forward(module): def custom_forward(*inputs): return module(*inputs) return custom_forward if is_torch_version(">=", "1.11.0"): hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb, use_reentrant=False ) else: hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb ) else: hidden_states = resnet(hidden_states, temb, scale) output_states += (hidden_states,) if self.downsamplers is not None: for downsampler in self.downsamplers: hidden_states = downsampler(hidden_states) return hidden_states, output_states class KCrossAttnDownBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, temb_channels: int, cross_attention_dim: int, dropout: float = 0.0, num_layers: int = 4, resnet_group_size: int = 32, add_downsample: bool = True, attention_head_dim: int = 64, add_self_attention: bool = False, resnet_eps: float = 1e-5, resnet_act_fn: str = "gelu", ): super().__init__() resnets = [] attentions = [] self.has_cross_attention = True for i in range(num_layers): in_channels = in_channels if i == 0 else out_channels groups = in_channels // resnet_group_size groups_out = out_channels // resnet_group_size resnets.append( ResnetBlockCondNorm2D( in_channels=in_channels, out_channels=out_channels, dropout=dropout, temb_channels=temb_channels, groups=groups, groups_out=groups_out, eps=resnet_eps, non_linearity=resnet_act_fn, time_embedding_norm="ada_group", conv_shortcut_bias=False, ) ) attentions.append( KAttentionBlock( out_channels, out_channels // attention_head_dim, attention_head_dim, cross_attention_dim=cross_attention_dim, temb_channels=temb_channels, attention_bias=True, add_self_attention=add_self_attention, cross_attention_norm="layer_norm", group_size=resnet_group_size, ) ) self.resnets = nn.ModuleList(resnets) self.attentions = nn.ModuleList(attentions) if add_downsample: self.downsamplers = nn.ModuleList([KDownsample2D()]) else: self.downsamplers = None self.gradient_checkpointing = False def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, encoder_hidden_states: Optional[torch.FloatTensor] = None, attention_mask: Optional[torch.FloatTensor] = None, cross_attention_kwargs: Optional[Dict[str, Any]] = None, encoder_attention_mask: Optional[torch.FloatTensor] = None, ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]: output_states = () lora_scale = cross_attention_kwargs.get("scale", 1.0) if cross_attention_kwargs is not None else 1.0 for resnet, attn in zip(self.resnets, self.attentions): if self.training and self.gradient_checkpointing: def create_custom_forward(module, return_dict=None): def custom_forward(*inputs): if return_dict is not None: return module(*inputs, return_dict=return_dict) else: return module(*inputs) return custom_forward ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {} hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb, **ckpt_kwargs, ) hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, emb=temb, attention_mask=attention_mask, cross_attention_kwargs=cross_attention_kwargs, encoder_attention_mask=encoder_attention_mask, ) else: hidden_states = resnet(hidden_states, temb, scale=lora_scale) hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, emb=temb, attention_mask=attention_mask, cross_attention_kwargs=cross_attention_kwargs, encoder_attention_mask=encoder_attention_mask, ) if self.downsamplers is None: output_states += (None,) else: output_states += (hidden_states,) if self.downsamplers is not None: for downsampler in self.downsamplers: hidden_states = downsampler(hidden_states) return hidden_states, output_states class AttnUpBlock2D(nn.Module): def __init__( self, in_channels: int, prev_output_channel: int, out_channels: int, temb_channels: int, resolution_idx: int = None, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, attention_head_dim: int = 1, output_scale_factor: float = 1.0, upsample_type: str = "conv", ): super().__init__() resnets = [] attentions = [] self.upsample_type = upsample_type if attention_head_dim is None: logger.warn( f"It is not recommend to pass `attention_head_dim=None`. Defaulting `attention_head_dim` to `in_channels`: {out_channels}." ) attention_head_dim = out_channels for i in range(num_layers): res_skip_channels = in_channels if (i == num_layers - 1) else out_channels resnet_in_channels = prev_output_channel if i == 0 else out_channels resnets.append( ResnetBlock2D( in_channels=resnet_in_channels + res_skip_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) attentions.append( Attention( out_channels, heads=out_channels // attention_head_dim, dim_head=attention_head_dim, rescale_output_factor=output_scale_factor, eps=resnet_eps, norm_num_groups=resnet_groups, residual_connection=True, bias=True, upcast_softmax=True, _from_deprecated_attn_block=True, ) ) self.attentions = nn.ModuleList(attentions) self.resnets = nn.ModuleList(resnets) if upsample_type == "conv": self.upsamplers = nn.ModuleList([Upsample2D(out_channels, use_conv=True, out_channels=out_channels)]) elif upsample_type == "resnet": self.upsamplers = nn.ModuleList( [ ResnetBlock2D( in_channels=out_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, up=True, ) ] ) else: self.upsamplers = None self.resolution_idx = resolution_idx def forward( self, hidden_states: torch.FloatTensor, res_hidden_states_tuple: Tuple[torch.FloatTensor, ...], temb: Optional[torch.FloatTensor] = None, upsample_size: Optional[int] = None, scale: float = 1.0, ) -> torch.FloatTensor: for resnet, attn in zip(self.resnets, self.attentions): # pop res hidden states res_hidden_states = res_hidden_states_tuple[-1] res_hidden_states_tuple = res_hidden_states_tuple[:-1] hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1) hidden_states = resnet(hidden_states, temb, scale=scale) cross_attention_kwargs = {"scale": scale} hidden_states = attn(hidden_states, **cross_attention_kwargs) if self.upsamplers is not None: for upsampler in self.upsamplers: if self.upsample_type == "resnet": hidden_states = upsampler(hidden_states, temb=temb, scale=scale) else: hidden_states = upsampler(hidden_states, scale=scale) return hidden_states class CrossAttnUpBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, prev_output_channel: int, temb_channels: int, resolution_idx: Optional[int] = None, dropout: float = 0.0, num_layers: int = 1, transformer_layers_per_block: Union[int, Tuple[int]] = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, num_attention_heads: int = 1, cross_attention_dim: int = 1280, output_scale_factor: float = 1.0, add_upsample: bool = True, dual_cross_attention: bool = False, use_linear_projection: bool = False, only_cross_attention: bool = False, upcast_attention: bool = False, attention_type: str = "default", ): super().__init__() resnets = [] attentions = [] self.has_cross_attention = True self.num_attention_heads = num_attention_heads if isinstance(transformer_layers_per_block, int): transformer_layers_per_block = [transformer_layers_per_block] * num_layers for i in range(num_layers): res_skip_channels = in_channels if (i == num_layers - 1) else out_channels resnet_in_channels = prev_output_channel if i == 0 else out_channels resnets.append( ResnetBlock2D( in_channels=resnet_in_channels + res_skip_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) if not dual_cross_attention: attentions.append( Transformer2DModel( num_attention_heads, out_channels // num_attention_heads, in_channels=out_channels, num_layers=transformer_layers_per_block[i], cross_attention_dim=cross_attention_dim, norm_num_groups=resnet_groups, use_linear_projection=use_linear_projection, only_cross_attention=only_cross_attention, upcast_attention=upcast_attention, attention_type=attention_type, ) ) else: attentions.append( DualTransformer2DModel( num_attention_heads, out_channels // num_attention_heads, in_channels=out_channels, num_layers=1, cross_attention_dim=cross_attention_dim, norm_num_groups=resnet_groups, ) ) self.attentions = nn.ModuleList(attentions) self.resnets = nn.ModuleList(resnets) if add_upsample: self.upsamplers = nn.ModuleList([Upsample2D(out_channels, use_conv=True, out_channels=out_channels)]) else: self.upsamplers = None self.gradient_checkpointing = False self.resolution_idx = resolution_idx def forward( self, hidden_states: torch.FloatTensor, res_hidden_states_tuple: Tuple[torch.FloatTensor, ...], temb: Optional[torch.FloatTensor] = None, encoder_hidden_states: Optional[torch.FloatTensor] = None, cross_attention_kwargs: Optional[Dict[str, Any]] = None, upsample_size: Optional[int] = None, attention_mask: Optional[torch.FloatTensor] = None, encoder_attention_mask: Optional[torch.FloatTensor] = None, ) -> torch.FloatTensor: lora_scale = cross_attention_kwargs.get("scale", 1.0) if cross_attention_kwargs is not None else 1.0 is_freeu_enabled = ( getattr(self, "s1", None) and getattr(self, "s2", None) and getattr(self, "b1", None) and getattr(self, "b2", None) ) for resnet, attn in zip(self.resnets, self.attentions): # pop res hidden states res_hidden_states = res_hidden_states_tuple[-1] res_hidden_states_tuple = res_hidden_states_tuple[:-1] # FreeU: Only operate on the first two stages if is_freeu_enabled: hidden_states, res_hidden_states = apply_freeu( self.resolution_idx, hidden_states, res_hidden_states, s1=self.s1, s2=self.s2, b1=self.b1, b2=self.b2, ) hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1) if self.training and self.gradient_checkpointing: def create_custom_forward(module, return_dict=None): def custom_forward(*inputs): if return_dict is not None: return module(*inputs, return_dict=return_dict) else: return module(*inputs) return custom_forward ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {} hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb, **ckpt_kwargs, ) hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, cross_attention_kwargs=cross_attention_kwargs, attention_mask=attention_mask, encoder_attention_mask=encoder_attention_mask, return_dict=False, )[0] else: hidden_states = resnet(hidden_states, temb, scale=lora_scale) hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, cross_attention_kwargs=cross_attention_kwargs, attention_mask=attention_mask, encoder_attention_mask=encoder_attention_mask, return_dict=False, )[0] if self.upsamplers is not None: for upsampler in self.upsamplers: hidden_states = upsampler(hidden_states, upsample_size, scale=lora_scale) return hidden_states class UpBlock2D(nn.Module): def __init__( self, in_channels: int, prev_output_channel: int, out_channels: int, temb_channels: int, resolution_idx: Optional[int] = None, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, output_scale_factor: float = 1.0, add_upsample: bool = True, ): super().__init__() resnets = [] for i in range(num_layers): res_skip_channels = in_channels if (i == num_layers - 1) else out_channels resnet_in_channels = prev_output_channel if i == 0 else out_channels resnets.append( ResnetBlock2D( in_channels=resnet_in_channels + res_skip_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) self.resnets = nn.ModuleList(resnets) if add_upsample: self.upsamplers = nn.ModuleList([Upsample2D(out_channels, use_conv=True, out_channels=out_channels)]) else: self.upsamplers = None self.gradient_checkpointing = False self.resolution_idx = resolution_idx def forward( self, hidden_states: torch.FloatTensor, res_hidden_states_tuple: Tuple[torch.FloatTensor, ...], temb: Optional[torch.FloatTensor] = None, upsample_size: Optional[int] = None, scale: float = 1.0, ) -> torch.FloatTensor: is_freeu_enabled = ( getattr(self, "s1", None) and getattr(self, "s2", None) and getattr(self, "b1", None) and getattr(self, "b2", None) ) for resnet in self.resnets: # pop res hidden states res_hidden_states = res_hidden_states_tuple[-1] res_hidden_states_tuple = res_hidden_states_tuple[:-1] # FreeU: Only operate on the first two stages if is_freeu_enabled: hidden_states, res_hidden_states = apply_freeu( self.resolution_idx, hidden_states, res_hidden_states, s1=self.s1, s2=self.s2, b1=self.b1, b2=self.b2, ) hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1) if self.training and self.gradient_checkpointing: def create_custom_forward(module): def custom_forward(*inputs): return module(*inputs) return custom_forward if is_torch_version(">=", "1.11.0"): hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb, use_reentrant=False ) else: hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb ) else: hidden_states = resnet(hidden_states, temb, scale=scale) if self.upsamplers is not None: for upsampler in self.upsamplers: hidden_states = upsampler(hidden_states, upsample_size, scale=scale) return hidden_states class UpDecoderBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, resolution_idx: Optional[int] = None, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", # default, spatial resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, output_scale_factor: float = 1.0, add_upsample: bool = True, temb_channels: Optional[int] = None, ): super().__init__() resnets = [] for i in range(num_layers): input_channels = in_channels if i == 0 else out_channels if resnet_time_scale_shift == "spatial": resnets.append( ResnetBlockCondNorm2D( in_channels=input_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm="spatial", non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, ) ) else: resnets.append( ResnetBlock2D( in_channels=input_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) self.resnets = nn.ModuleList(resnets) if add_upsample: self.upsamplers = nn.ModuleList([Upsample2D(out_channels, use_conv=True, out_channels=out_channels)]) else: self.upsamplers = None self.resolution_idx = resolution_idx def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, scale: float = 1.0 ) -> torch.FloatTensor: for resnet in self.resnets: hidden_states = resnet(hidden_states, temb=temb, scale=scale) if self.upsamplers is not None: for upsampler in self.upsamplers: hidden_states = upsampler(hidden_states) return hidden_states class AttnUpDecoderBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, resolution_idx: Optional[int] = None, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, attention_head_dim: int = 1, output_scale_factor: float = 1.0, add_upsample: bool = True, temb_channels: Optional[int] = None, ): super().__init__() resnets = [] attentions = [] if attention_head_dim is None: logger.warn( f"It is not recommend to pass `attention_head_dim=None`. Defaulting `attention_head_dim` to `out_channels`: {out_channels}." ) attention_head_dim = out_channels for i in range(num_layers): input_channels = in_channels if i == 0 else out_channels if resnet_time_scale_shift == "spatial": resnets.append( ResnetBlockCondNorm2D( in_channels=input_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm="spatial", non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, ) ) else: resnets.append( ResnetBlock2D( in_channels=input_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) attentions.append( Attention( out_channels, heads=out_channels // attention_head_dim, dim_head=attention_head_dim, rescale_output_factor=output_scale_factor, eps=resnet_eps, norm_num_groups=resnet_groups if resnet_time_scale_shift != "spatial" else None, spatial_norm_dim=temb_channels if resnet_time_scale_shift == "spatial" else None, residual_connection=True, bias=True, upcast_softmax=True, _from_deprecated_attn_block=True, ) ) self.attentions = nn.ModuleList(attentions) self.resnets = nn.ModuleList(resnets) if add_upsample: self.upsamplers = nn.ModuleList([Upsample2D(out_channels, use_conv=True, out_channels=out_channels)]) else: self.upsamplers = None self.resolution_idx = resolution_idx def forward( self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None, scale: float = 1.0 ) -> torch.FloatTensor: for resnet, attn in zip(self.resnets, self.attentions): hidden_states = resnet(hidden_states, temb=temb, scale=scale) cross_attention_kwargs = {"scale": scale} hidden_states = attn(hidden_states, temb=temb, **cross_attention_kwargs) if self.upsamplers is not None: for upsampler in self.upsamplers: hidden_states = upsampler(hidden_states, scale=scale) return hidden_states class AttnSkipUpBlock2D(nn.Module): def __init__( self, in_channels: int, prev_output_channel: int, out_channels: int, temb_channels: int, resolution_idx: Optional[int] = None, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_pre_norm: bool = True, attention_head_dim: int = 1, output_scale_factor: float = np.sqrt(2.0), add_upsample: bool = True, ): super().__init__() self.attentions = nn.ModuleList([]) self.resnets = nn.ModuleList([]) for i in range(num_layers): res_skip_channels = in_channels if (i == num_layers - 1) else out_channels resnet_in_channels = prev_output_channel if i == 0 else out_channels self.resnets.append( ResnetBlock2D( in_channels=resnet_in_channels + res_skip_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=min(resnet_in_channels + res_skip_channels // 4, 32), groups_out=min(out_channels // 4, 32), dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) if attention_head_dim is None: logger.warn( f"It is not recommend to pass `attention_head_dim=None`. Defaulting `attention_head_dim` to `out_channels`: {out_channels}." ) attention_head_dim = out_channels self.attentions.append( Attention( out_channels, heads=out_channels // attention_head_dim, dim_head=attention_head_dim, rescale_output_factor=output_scale_factor, eps=resnet_eps, norm_num_groups=32, residual_connection=True, bias=True, upcast_softmax=True, _from_deprecated_attn_block=True, ) ) self.upsampler = FirUpsample2D(in_channels, out_channels=out_channels) if add_upsample: self.resnet_up = ResnetBlock2D( in_channels=out_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=min(out_channels // 4, 32), groups_out=min(out_channels // 4, 32), dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, use_in_shortcut=True, up=True, kernel="fir", ) self.skip_conv = nn.Conv2d(out_channels, 3, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) self.skip_norm = torch.nn.GroupNorm( num_groups=min(out_channels // 4, 32), num_channels=out_channels, eps=resnet_eps, affine=True ) self.act = nn.SiLU() else: self.resnet_up = None self.skip_conv = None self.skip_norm = None self.act = None self.resolution_idx = resolution_idx def forward( self, hidden_states: torch.FloatTensor, res_hidden_states_tuple: Tuple[torch.FloatTensor, ...], temb: Optional[torch.FloatTensor] = None, skip_sample=None, scale: float = 1.0, ) -> Tuple[torch.FloatTensor, torch.FloatTensor]: for resnet in self.resnets: # pop res hidden states res_hidden_states = res_hidden_states_tuple[-1] res_hidden_states_tuple = res_hidden_states_tuple[:-1] hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1) hidden_states = resnet(hidden_states, temb, scale=scale) cross_attention_kwargs = {"scale": scale} hidden_states = self.attentions[0](hidden_states, **cross_attention_kwargs) if skip_sample is not None: skip_sample = self.upsampler(skip_sample) else: skip_sample = 0 if self.resnet_up is not None: skip_sample_states = self.skip_norm(hidden_states) skip_sample_states = self.act(skip_sample_states) skip_sample_states = self.skip_conv(skip_sample_states) skip_sample = skip_sample + skip_sample_states hidden_states = self.resnet_up(hidden_states, temb, scale=scale) return hidden_states, skip_sample class SkipUpBlock2D(nn.Module): def __init__( self, in_channels: int, prev_output_channel: int, out_channels: int, temb_channels: int, resolution_idx: Optional[int] = None, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_pre_norm: bool = True, output_scale_factor: float = np.sqrt(2.0), add_upsample: bool = True, upsample_padding: int = 1, ): super().__init__() self.resnets = nn.ModuleList([]) for i in range(num_layers): res_skip_channels = in_channels if (i == num_layers - 1) else out_channels resnet_in_channels = prev_output_channel if i == 0 else out_channels self.resnets.append( ResnetBlock2D( in_channels=resnet_in_channels + res_skip_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=min((resnet_in_channels + res_skip_channels) // 4, 32), groups_out=min(out_channels // 4, 32), dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, ) ) self.upsampler = FirUpsample2D(in_channels, out_channels=out_channels) if add_upsample: self.resnet_up = ResnetBlock2D( in_channels=out_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=min(out_channels // 4, 32), groups_out=min(out_channels // 4, 32), dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, use_in_shortcut=True, up=True, kernel="fir", ) self.skip_conv = nn.Conv2d(out_channels, 3, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) self.skip_norm = torch.nn.GroupNorm( num_groups=min(out_channels // 4, 32), num_channels=out_channels, eps=resnet_eps, affine=True ) self.act = nn.SiLU() else: self.resnet_up = None self.skip_conv = None self.skip_norm = None self.act = None self.resolution_idx = resolution_idx def forward( self, hidden_states: torch.FloatTensor, res_hidden_states_tuple: Tuple[torch.FloatTensor, ...], temb: Optional[torch.FloatTensor] = None, skip_sample=None, scale: float = 1.0, ) -> Tuple[torch.FloatTensor, torch.FloatTensor]: for resnet in self.resnets: # pop res hidden states res_hidden_states = res_hidden_states_tuple[-1] res_hidden_states_tuple = res_hidden_states_tuple[:-1] hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1) hidden_states = resnet(hidden_states, temb, scale=scale) if skip_sample is not None: skip_sample = self.upsampler(skip_sample) else: skip_sample = 0 if self.resnet_up is not None: skip_sample_states = self.skip_norm(hidden_states) skip_sample_states = self.act(skip_sample_states) skip_sample_states = self.skip_conv(skip_sample_states) skip_sample = skip_sample + skip_sample_states hidden_states = self.resnet_up(hidden_states, temb, scale=scale) return hidden_states, skip_sample class ResnetUpsampleBlock2D(nn.Module): def __init__( self, in_channels: int, prev_output_channel: int, out_channels: int, temb_channels: int, resolution_idx: Optional[int] = None, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, output_scale_factor: float = 1.0, add_upsample: bool = True, skip_time_act: bool = False, ): super().__init__() resnets = [] for i in range(num_layers): res_skip_channels = in_channels if (i == num_layers - 1) else out_channels resnet_in_channels = prev_output_channel if i == 0 else out_channels resnets.append( ResnetBlock2D( in_channels=resnet_in_channels + res_skip_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, skip_time_act=skip_time_act, ) ) self.resnets = nn.ModuleList(resnets) if add_upsample: self.upsamplers = nn.ModuleList( [ ResnetBlock2D( in_channels=out_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, skip_time_act=skip_time_act, up=True, ) ] ) else: self.upsamplers = None self.gradient_checkpointing = False self.resolution_idx = resolution_idx def forward( self, hidden_states: torch.FloatTensor, res_hidden_states_tuple: Tuple[torch.FloatTensor, ...], temb: Optional[torch.FloatTensor] = None, upsample_size: Optional[int] = None, scale: float = 1.0, ) -> torch.FloatTensor: for resnet in self.resnets: # pop res hidden states res_hidden_states = res_hidden_states_tuple[-1] res_hidden_states_tuple = res_hidden_states_tuple[:-1] hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1) if self.training and self.gradient_checkpointing: def create_custom_forward(module): def custom_forward(*inputs): return module(*inputs) return custom_forward if is_torch_version(">=", "1.11.0"): hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb, use_reentrant=False ) else: hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb ) else: hidden_states = resnet(hidden_states, temb, scale=scale) if self.upsamplers is not None: for upsampler in self.upsamplers: hidden_states = upsampler(hidden_states, temb, scale=scale) return hidden_states class SimpleCrossAttnUpBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, prev_output_channel: int, temb_channels: int, resolution_idx: Optional[int] = None, dropout: float = 0.0, num_layers: int = 1, resnet_eps: float = 1e-6, resnet_time_scale_shift: str = "default", resnet_act_fn: str = "swish", resnet_groups: int = 32, resnet_pre_norm: bool = True, attention_head_dim: int = 1, cross_attention_dim: int = 1280, output_scale_factor: float = 1.0, add_upsample: bool = True, skip_time_act: bool = False, only_cross_attention: bool = False, cross_attention_norm: Optional[str] = None, ): super().__init__() resnets = [] attentions = [] self.has_cross_attention = True self.attention_head_dim = attention_head_dim self.num_heads = out_channels // self.attention_head_dim for i in range(num_layers): res_skip_channels = in_channels if (i == num_layers - 1) else out_channels resnet_in_channels = prev_output_channel if i == 0 else out_channels resnets.append( ResnetBlock2D( in_channels=resnet_in_channels + res_skip_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, skip_time_act=skip_time_act, ) ) processor = ( AttnAddedKVProcessor2_0() if hasattr(F, "scaled_dot_product_attention") else AttnAddedKVProcessor() ) attentions.append( Attention( query_dim=out_channels, cross_attention_dim=out_channels, heads=self.num_heads, dim_head=self.attention_head_dim, added_kv_proj_dim=cross_attention_dim, norm_num_groups=resnet_groups, bias=True, upcast_softmax=True, only_cross_attention=only_cross_attention, cross_attention_norm=cross_attention_norm, processor=processor, ) ) self.attentions = nn.ModuleList(attentions) self.resnets = nn.ModuleList(resnets) if add_upsample: self.upsamplers = nn.ModuleList( [ ResnetBlock2D( in_channels=out_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=resnet_groups, dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, skip_time_act=skip_time_act, up=True, ) ] ) else: self.upsamplers = None self.gradient_checkpointing = False self.resolution_idx = resolution_idx def forward( self, hidden_states: torch.FloatTensor, res_hidden_states_tuple: Tuple[torch.FloatTensor, ...], temb: Optional[torch.FloatTensor] = None, encoder_hidden_states: Optional[torch.FloatTensor] = None, upsample_size: Optional[int] = None, attention_mask: Optional[torch.FloatTensor] = None, cross_attention_kwargs: Optional[Dict[str, Any]] = None, encoder_attention_mask: Optional[torch.FloatTensor] = None, ) -> torch.FloatTensor: cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {} lora_scale = cross_attention_kwargs.get("scale", 1.0) if attention_mask is None: # if encoder_hidden_states is defined: we are doing cross-attn, so we should use cross-attn mask. mask = None if encoder_hidden_states is None else encoder_attention_mask else: # when attention_mask is defined: we don't even check for encoder_attention_mask. # this is to maintain compatibility with UnCLIP, which uses 'attention_mask' param for cross-attn masks. # TODO: UnCLIP should express cross-attn mask via encoder_attention_mask param instead of via attention_mask. # then we can simplify this whole if/else block to: # mask = attention_mask if encoder_hidden_states is None else encoder_attention_mask mask = attention_mask for resnet, attn in zip(self.resnets, self.attentions): # resnet # pop res hidden states res_hidden_states = res_hidden_states_tuple[-1] res_hidden_states_tuple = res_hidden_states_tuple[:-1] hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1) if self.training and self.gradient_checkpointing: def create_custom_forward(module, return_dict=None): def custom_forward(*inputs): if return_dict is not None: return module(*inputs, return_dict=return_dict) else: return module(*inputs) return custom_forward hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb) hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, attention_mask=mask, **cross_attention_kwargs, ) else: hidden_states = resnet(hidden_states, temb, scale=lora_scale) hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, attention_mask=mask, **cross_attention_kwargs, ) if self.upsamplers is not None: for upsampler in self.upsamplers: hidden_states = upsampler(hidden_states, temb, scale=lora_scale) return hidden_states class KUpBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, temb_channels: int, resolution_idx: int, dropout: float = 0.0, num_layers: int = 5, resnet_eps: float = 1e-5, resnet_act_fn: str = "gelu", resnet_group_size: Optional[int] = 32, add_upsample: bool = True, ): super().__init__() resnets = [] k_in_channels = 2 * out_channels k_out_channels = in_channels num_layers = num_layers - 1 for i in range(num_layers): in_channels = k_in_channels if i == 0 else out_channels groups = in_channels // resnet_group_size groups_out = out_channels // resnet_group_size resnets.append( ResnetBlockCondNorm2D( in_channels=in_channels, out_channels=k_out_channels if (i == num_layers - 1) else out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=groups, groups_out=groups_out, dropout=dropout, non_linearity=resnet_act_fn, time_embedding_norm="ada_group", conv_shortcut_bias=False, ) ) self.resnets = nn.ModuleList(resnets) if add_upsample: self.upsamplers = nn.ModuleList([KUpsample2D()]) else: self.upsamplers = None self.gradient_checkpointing = False self.resolution_idx = resolution_idx def forward( self, hidden_states: torch.FloatTensor, res_hidden_states_tuple: Tuple[torch.FloatTensor, ...], temb: Optional[torch.FloatTensor] = None, upsample_size: Optional[int] = None, scale: float = 1.0, ) -> torch.FloatTensor: res_hidden_states_tuple = res_hidden_states_tuple[-1] if res_hidden_states_tuple is not None: hidden_states = torch.cat([hidden_states, res_hidden_states_tuple], dim=1) for resnet in self.resnets: if self.training and self.gradient_checkpointing: def create_custom_forward(module): def custom_forward(*inputs): return module(*inputs) return custom_forward if is_torch_version(">=", "1.11.0"): hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb, use_reentrant=False ) else: hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb ) else: hidden_states = resnet(hidden_states, temb, scale=scale) if self.upsamplers is not None: for upsampler in self.upsamplers: hidden_states = upsampler(hidden_states) return hidden_states class KCrossAttnUpBlock2D(nn.Module): def __init__( self, in_channels: int, out_channels: int, temb_channels: int, resolution_idx: int, dropout: float = 0.0, num_layers: int = 4, resnet_eps: float = 1e-5, resnet_act_fn: str = "gelu", resnet_group_size: int = 32, attention_head_dim: int = 1, # attention dim_head cross_attention_dim: int = 768, add_upsample: bool = True, upcast_attention: bool = False, ): super().__init__() resnets = [] attentions = [] is_first_block = in_channels == out_channels == temb_channels is_middle_block = in_channels != out_channels add_self_attention = True if is_first_block else False self.has_cross_attention = True self.attention_head_dim = attention_head_dim # in_channels, and out_channels for the block (k-unet) k_in_channels = out_channels if is_first_block else 2 * out_channels k_out_channels = in_channels num_layers = num_layers - 1 for i in range(num_layers): in_channels = k_in_channels if i == 0 else out_channels groups = in_channels // resnet_group_size groups_out = out_channels // resnet_group_size if is_middle_block and (i == num_layers - 1): conv_2d_out_channels = k_out_channels else: conv_2d_out_channels = None resnets.append( ResnetBlockCondNorm2D( in_channels=in_channels, out_channels=out_channels, conv_2d_out_channels=conv_2d_out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=groups, groups_out=groups_out, dropout=dropout, non_linearity=resnet_act_fn, time_embedding_norm="ada_group", conv_shortcut_bias=False, ) ) attentions.append( KAttentionBlock( k_out_channels if (i == num_layers - 1) else out_channels, k_out_channels // attention_head_dim if (i == num_layers - 1) else out_channels // attention_head_dim, attention_head_dim, cross_attention_dim=cross_attention_dim, temb_channels=temb_channels, attention_bias=True, add_self_attention=add_self_attention, cross_attention_norm="layer_norm", upcast_attention=upcast_attention, ) ) self.resnets = nn.ModuleList(resnets) self.attentions = nn.ModuleList(attentions) if add_upsample: self.upsamplers = nn.ModuleList([KUpsample2D()]) else: self.upsamplers = None self.gradient_checkpointing = False self.resolution_idx = resolution_idx def forward( self, hidden_states: torch.FloatTensor, res_hidden_states_tuple: Tuple[torch.FloatTensor, ...], temb: Optional[torch.FloatTensor] = None, encoder_hidden_states: Optional[torch.FloatTensor] = None, cross_attention_kwargs: Optional[Dict[str, Any]] = None, upsample_size: Optional[int] = None, attention_mask: Optional[torch.FloatTensor] = None, encoder_attention_mask: Optional[torch.FloatTensor] = None, ) -> torch.FloatTensor: res_hidden_states_tuple = res_hidden_states_tuple[-1] if res_hidden_states_tuple is not None: hidden_states = torch.cat([hidden_states, res_hidden_states_tuple], dim=1) lora_scale = cross_attention_kwargs.get("scale", 1.0) if cross_attention_kwargs is not None else 1.0 for resnet, attn in zip(self.resnets, self.attentions): if self.training and self.gradient_checkpointing: def create_custom_forward(module, return_dict=None): def custom_forward(*inputs): if return_dict is not None: return module(*inputs, return_dict=return_dict) else: return module(*inputs) return custom_forward ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {} hidden_states = torch.utils.checkpoint.checkpoint( create_custom_forward(resnet), hidden_states, temb, **ckpt_kwargs, ) hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, emb=temb, attention_mask=attention_mask, cross_attention_kwargs=cross_attention_kwargs, encoder_attention_mask=encoder_attention_mask, ) else: hidden_states = resnet(hidden_states, temb, scale=lora_scale) hidden_states = attn( hidden_states, encoder_hidden_states=encoder_hidden_states, emb=temb, attention_mask=attention_mask, cross_attention_kwargs=cross_attention_kwargs, encoder_attention_mask=encoder_attention_mask, ) if self.upsamplers is not None: for upsampler in self.upsamplers: hidden_states = upsampler(hidden_states) return hidden_states # can potentially later be renamed to `No-feed-forward` attention class KAttentionBlock(nn.Module): r""" A basic Transformer block. Parameters: dim (`int`): The number of channels in the input and output. num_attention_heads (`int`): The number of heads to use for multi-head attention. attention_head_dim (`int`): The number of channels in each head. dropout (`float`, *optional*, defaults to 0.0): The dropout probability to use. cross_attention_dim (`int`, *optional*): The size of the encoder_hidden_states vector for cross attention. attention_bias (`bool`, *optional*, defaults to `False`): Configure if the attention layers should contain a bias parameter. upcast_attention (`bool`, *optional*, defaults to `False`): Set to `True` to upcast the attention computation to `float32`. temb_channels (`int`, *optional*, defaults to 768): The number of channels in the token embedding. add_self_attention (`bool`, *optional*, defaults to `False`): Set to `True` to add self-attention to the block. cross_attention_norm (`str`, *optional*, defaults to `None`): The type of normalization to use for the cross attention. Can be `None`, `layer_norm`, or `group_norm`. group_size (`int`, *optional*, defaults to 32): The number of groups to separate the channels into for group normalization. """ def __init__( self, dim: int, num_attention_heads: int, attention_head_dim: int, dropout: float = 0.0, cross_attention_dim: Optional[int] = None, attention_bias: bool = False, upcast_attention: bool = False, temb_channels: int = 768, # for ada_group_norm add_self_attention: bool = False, cross_attention_norm: Optional[str] = None, group_size: int = 32, ): super().__init__() self.add_self_attention = add_self_attention # 1. Self-Attn if add_self_attention: self.norm1 = AdaGroupNorm(temb_channels, dim, max(1, dim // group_size)) self.attn1 = Attention( query_dim=dim, heads=num_attention_heads, dim_head=attention_head_dim, dropout=dropout, bias=attention_bias, cross_attention_dim=None, cross_attention_norm=None, ) # 2. Cross-Attn self.norm2 = AdaGroupNorm(temb_channels, dim, max(1, dim // group_size)) self.attn2 = Attention( query_dim=dim, cross_attention_dim=cross_attention_dim, heads=num_attention_heads, dim_head=attention_head_dim, dropout=dropout, bias=attention_bias, upcast_attention=upcast_attention, cross_attention_norm=cross_attention_norm, ) def _to_3d(self, hidden_states: torch.FloatTensor, height: int, weight: int) -> torch.FloatTensor: return hidden_states.permute(0, 2, 3, 1).reshape(hidden_states.shape[0], height * weight, -1) def _to_4d(self, hidden_states: torch.FloatTensor, height: int, weight: int) -> torch.FloatTensor: return hidden_states.permute(0, 2, 1).reshape(hidden_states.shape[0], -1, height, weight) def forward( self, hidden_states: torch.FloatTensor, encoder_hidden_states: Optional[torch.FloatTensor] = None, # TODO: mark emb as non-optional (self.norm2 requires it). # requires assessing impact of change to positional param interface. emb: Optional[torch.FloatTensor] = None, attention_mask: Optional[torch.FloatTensor] = None, cross_attention_kwargs: Optional[Dict[str, Any]] = None, encoder_attention_mask: Optional[torch.FloatTensor] = None, ) -> torch.FloatTensor: cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {} # 1. Self-Attention if self.add_self_attention: norm_hidden_states = self.norm1(hidden_states, emb) height, weight = norm_hidden_states.shape[2:] norm_hidden_states = self._to_3d(norm_hidden_states, height, weight) attn_output = self.attn1( norm_hidden_states, encoder_hidden_states=None, attention_mask=attention_mask, **cross_attention_kwargs, ) attn_output = self._to_4d(attn_output, height, weight) hidden_states = attn_output + hidden_states # 2. Cross-Attention/None norm_hidden_states = self.norm2(hidden_states, emb) height, weight = norm_hidden_states.shape[2:] norm_hidden_states = self._to_3d(norm_hidden_states, height, weight) attn_output = self.attn2( norm_hidden_states, encoder_hidden_states=encoder_hidden_states, attention_mask=attention_mask if encoder_hidden_states is None else encoder_attention_mask, **cross_attention_kwargs, ) attn_output = self._to_4d(attn_output, height, weight) hidden_states = attn_output + hidden_states return hidden_states

diffusers/src/diffusers/models/unets/unet_2d_blocks.py/0

from typing import TYPE_CHECKING from ..utils import ( DIFFUSERS_SLOW_IMPORT, OptionalDependencyNotAvailable, _LazyModule, get_objects_from_module, is_flax_available, is_k_diffusion_available, is_librosa_available, is_note_seq_available, is_onnx_available, is_torch_available, is_transformers_available, ) # These modules contain pipelines from multiple libraries/frameworks _dummy_objects = {} _import_structure = { "controlnet": [], "controlnet_xs": [], "deprecated": [], "latent_diffusion": [], "stable_diffusion": [], "stable_diffusion_xl": [], } try: if not is_torch_available(): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils import dummy_pt_objects # noqa F403 _dummy_objects.update(get_objects_from_module(dummy_pt_objects)) else: _import_structure["auto_pipeline"] = [ "AutoPipelineForImage2Image", "AutoPipelineForInpainting", "AutoPipelineForText2Image", ] _import_structure["consistency_models"] = ["ConsistencyModelPipeline"] _import_structure["dance_diffusion"] = ["DanceDiffusionPipeline"] _import_structure["ddim"] = ["DDIMPipeline"] _import_structure["ddpm"] = ["DDPMPipeline"] _import_structure["dit"] = ["DiTPipeline"] _import_structure["latent_diffusion"].extend(["LDMSuperResolutionPipeline"]) _import_structure["pipeline_utils"] = [ "AudioPipelineOutput", "DiffusionPipeline", "ImagePipelineOutput", ] _import_structure["deprecated"].extend( [ "PNDMPipeline", "LDMPipeline", "RePaintPipeline", "ScoreSdeVePipeline", "KarrasVePipeline", ] ) try: if not (is_torch_available() and is_librosa_available()): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils import dummy_torch_and_librosa_objects # noqa F403 _dummy_objects.update(get_objects_from_module(dummy_torch_and_librosa_objects)) else: _import_structure["deprecated"].extend(["AudioDiffusionPipeline", "Mel"]) try: if not (is_transformers_available() and is_torch_available() and is_note_seq_available()): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils import dummy_transformers_and_torch_and_note_seq_objects # noqa F403 _dummy_objects.update(get_objects_from_module(dummy_transformers_and_torch_and_note_seq_objects)) else: _import_structure["deprecated"].extend( [ "MidiProcessor", "SpectrogramDiffusionPipeline", ] ) try: if not (is_torch_available() and is_transformers_available()): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils import dummy_torch_and_transformers_objects # noqa F403 _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects)) else: _import_structure["deprecated"].extend( [ "VQDiffusionPipeline", "AltDiffusionPipeline", "AltDiffusionImg2ImgPipeline", "CycleDiffusionPipeline", "StableDiffusionInpaintPipelineLegacy", "StableDiffusionPix2PixZeroPipeline", "StableDiffusionParadigmsPipeline", "StableDiffusionModelEditingPipeline", "VersatileDiffusionDualGuidedPipeline", "VersatileDiffusionImageVariationPipeline", "VersatileDiffusionPipeline", "VersatileDiffusionTextToImagePipeline", ] ) _import_structure["amused"] = ["AmusedImg2ImgPipeline", "AmusedInpaintPipeline", "AmusedPipeline"] _import_structure["animatediff"] = [ "AnimateDiffPipeline", "AnimateDiffVideoToVideoPipeline", ] _import_structure["audioldm"] = ["AudioLDMPipeline"] _import_structure["audioldm2"] = [ "AudioLDM2Pipeline", "AudioLDM2ProjectionModel", "AudioLDM2UNet2DConditionModel", ] _import_structure["blip_diffusion"] = ["BlipDiffusionPipeline"] _import_structure["controlnet"].extend( [ "BlipDiffusionControlNetPipeline", "StableDiffusionControlNetImg2ImgPipeline", "StableDiffusionControlNetInpaintPipeline", "StableDiffusionControlNetPipeline", "StableDiffusionXLControlNetImg2ImgPipeline", "StableDiffusionXLControlNetInpaintPipeline", "StableDiffusionXLControlNetPipeline", ] ) _import_structure["deepfloyd_if"] = [ "IFImg2ImgPipeline", "IFImg2ImgSuperResolutionPipeline", "IFInpaintingPipeline", "IFInpaintingSuperResolutionPipeline", "IFPipeline", "IFSuperResolutionPipeline", ] _import_structure["kandinsky"] = [ "KandinskyCombinedPipeline", "KandinskyImg2ImgCombinedPipeline", "KandinskyImg2ImgPipeline", "KandinskyInpaintCombinedPipeline", "KandinskyInpaintPipeline", "KandinskyPipeline", "KandinskyPriorPipeline", ] _import_structure["kandinsky2_2"] = [ "KandinskyV22CombinedPipeline", "KandinskyV22ControlnetImg2ImgPipeline", "KandinskyV22ControlnetPipeline", "KandinskyV22Img2ImgCombinedPipeline", "KandinskyV22Img2ImgPipeline", "KandinskyV22InpaintCombinedPipeline", "KandinskyV22InpaintPipeline", "KandinskyV22Pipeline", "KandinskyV22PriorEmb2EmbPipeline", "KandinskyV22PriorPipeline", ] _import_structure["kandinsky3"] = [ "Kandinsky3Img2ImgPipeline", "Kandinsky3Pipeline", ] _import_structure["latent_consistency_models"] = [ "LatentConsistencyModelImg2ImgPipeline", "LatentConsistencyModelPipeline", ] _import_structure["latent_diffusion"].extend(["LDMTextToImagePipeline"]) _import_structure["musicldm"] = ["MusicLDMPipeline"] _import_structure["paint_by_example"] = ["PaintByExamplePipeline"] _import_structure["pia"] = ["PIAPipeline"] _import_structure["pixart_alpha"] = ["PixArtAlphaPipeline"] _import_structure["semantic_stable_diffusion"] = ["SemanticStableDiffusionPipeline"] _import_structure["shap_e"] = ["ShapEImg2ImgPipeline", "ShapEPipeline"] _import_structure["stable_diffusion"].extend( [ "CLIPImageProjection", "StableDiffusionDepth2ImgPipeline", "StableDiffusionImageVariationPipeline", "StableDiffusionImg2ImgPipeline", "StableDiffusionInpaintPipeline", "StableDiffusionInstructPix2PixPipeline", "StableDiffusionLatentUpscalePipeline", "StableDiffusionPipeline", "StableDiffusionUpscalePipeline", "StableUnCLIPImg2ImgPipeline", "StableUnCLIPPipeline", "StableDiffusionLDM3DPipeline", ] ) _import_structure["stable_diffusion_attend_and_excite"] = ["StableDiffusionAttendAndExcitePipeline"] _import_structure["stable_diffusion_safe"] = ["StableDiffusionPipelineSafe"] _import_structure["stable_diffusion_sag"] = ["StableDiffusionSAGPipeline"] _import_structure["stable_diffusion_gligen"] = [ "StableDiffusionGLIGENPipeline", "StableDiffusionGLIGENTextImagePipeline", ] _import_structure["stable_video_diffusion"] = ["StableVideoDiffusionPipeline"] _import_structure["stable_diffusion_xl"].extend( [ "StableDiffusionXLImg2ImgPipeline", "StableDiffusionXLInpaintPipeline", "StableDiffusionXLInstructPix2PixPipeline", "StableDiffusionXLPipeline", ] ) _import_structure["stable_diffusion_diffedit"] = ["StableDiffusionDiffEditPipeline"] _import_structure["stable_diffusion_ldm3d"] = ["StableDiffusionLDM3DPipeline"] _import_structure["stable_diffusion_panorama"] = ["StableDiffusionPanoramaPipeline"] _import_structure["t2i_adapter"] = [ "StableDiffusionAdapterPipeline", "StableDiffusionXLAdapterPipeline", ] _import_structure["text_to_video_synthesis"] = [ "TextToVideoSDPipeline", "TextToVideoZeroPipeline", "TextToVideoZeroSDXLPipeline", "VideoToVideoSDPipeline", ] _import_structure["i2vgen_xl"] = ["I2VGenXLPipeline"] _import_structure["unclip"] = ["UnCLIPImageVariationPipeline", "UnCLIPPipeline"] _import_structure["unidiffuser"] = [ "ImageTextPipelineOutput", "UniDiffuserModel", "UniDiffuserPipeline", "UniDiffuserTextDecoder", ] _import_structure["wuerstchen"] = [ "WuerstchenCombinedPipeline", "WuerstchenDecoderPipeline", "WuerstchenPriorPipeline", ] try: if not is_onnx_available(): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils import dummy_onnx_objects # noqa F403 _dummy_objects.update(get_objects_from_module(dummy_onnx_objects)) else: _import_structure["onnx_utils"] = ["OnnxRuntimeModel"] try: if not (is_torch_available() and is_transformers_available() and is_onnx_available()): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils import dummy_torch_and_transformers_and_onnx_objects # noqa F403 _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_and_onnx_objects)) else: _import_structure["stable_diffusion"].extend( [ "OnnxStableDiffusionImg2ImgPipeline", "OnnxStableDiffusionInpaintPipeline", "OnnxStableDiffusionPipeline", "OnnxStableDiffusionUpscalePipeline", "StableDiffusionOnnxPipeline", ] ) try: if not (is_torch_available() and is_transformers_available() and is_k_diffusion_available()): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils import ( dummy_torch_and_transformers_and_k_diffusion_objects, ) _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_and_k_diffusion_objects)) else: _import_structure["stable_diffusion_k_diffusion"] = [ "StableDiffusionKDiffusionPipeline", "StableDiffusionXLKDiffusionPipeline", ] try: if not is_flax_available(): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils import dummy_flax_objects # noqa F403 _dummy_objects.update(get_objects_from_module(dummy_flax_objects)) else: _import_structure["pipeline_flax_utils"] = ["FlaxDiffusionPipeline"] try: if not (is_flax_available() and is_transformers_available()): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils import dummy_flax_and_transformers_objects # noqa F403 _dummy_objects.update(get_objects_from_module(dummy_flax_and_transformers_objects)) else: _import_structure["controlnet"].extend(["FlaxStableDiffusionControlNetPipeline"]) _import_structure["stable_diffusion"].extend( [ "FlaxStableDiffusionImg2ImgPipeline", "FlaxStableDiffusionInpaintPipeline", "FlaxStableDiffusionPipeline", ] ) _import_structure["stable_diffusion_xl"].extend( [ "FlaxStableDiffusionXLPipeline", ] ) if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT: try: if not is_torch_available(): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils.dummy_pt_objects import * # noqa F403 else: from .auto_pipeline import ( AutoPipelineForImage2Image, AutoPipelineForInpainting, AutoPipelineForText2Image, ) from .consistency_models import ConsistencyModelPipeline from .dance_diffusion import DanceDiffusionPipeline from .ddim import DDIMPipeline from .ddpm import DDPMPipeline from .deprecated import KarrasVePipeline, LDMPipeline, PNDMPipeline, RePaintPipeline, ScoreSdeVePipeline from .dit import DiTPipeline from .latent_diffusion import LDMSuperResolutionPipeline from .pipeline_utils import ( AudioPipelineOutput, DiffusionPipeline, ImagePipelineOutput, ) try: if not (is_torch_available() and is_librosa_available()): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils.dummy_torch_and_librosa_objects import * else: from .deprecated import AudioDiffusionPipeline, Mel try: if not (is_torch_available() and is_transformers_available()): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils.dummy_torch_and_transformers_objects import * else: from .amused import AmusedImg2ImgPipeline, AmusedInpaintPipeline, AmusedPipeline from .animatediff import AnimateDiffPipeline, AnimateDiffVideoToVideoPipeline from .audioldm import AudioLDMPipeline from .audioldm2 import ( AudioLDM2Pipeline, AudioLDM2ProjectionModel, AudioLDM2UNet2DConditionModel, ) from .blip_diffusion import BlipDiffusionPipeline from .controlnet import ( BlipDiffusionControlNetPipeline, StableDiffusionControlNetImg2ImgPipeline, StableDiffusionControlNetInpaintPipeline, StableDiffusionControlNetPipeline, StableDiffusionXLControlNetImg2ImgPipeline, StableDiffusionXLControlNetInpaintPipeline, StableDiffusionXLControlNetPipeline, ) from .deepfloyd_if import ( IFImg2ImgPipeline, IFImg2ImgSuperResolutionPipeline, IFInpaintingPipeline, IFInpaintingSuperResolutionPipeline, IFPipeline, IFSuperResolutionPipeline, ) from .deprecated import ( AltDiffusionImg2ImgPipeline, AltDiffusionPipeline, CycleDiffusionPipeline, StableDiffusionInpaintPipelineLegacy, StableDiffusionModelEditingPipeline, StableDiffusionParadigmsPipeline, StableDiffusionPix2PixZeroPipeline, VersatileDiffusionDualGuidedPipeline, VersatileDiffusionImageVariationPipeline, VersatileDiffusionPipeline, VersatileDiffusionTextToImagePipeline, VQDiffusionPipeline, ) from .i2vgen_xl import I2VGenXLPipeline from .kandinsky import ( KandinskyCombinedPipeline, KandinskyImg2ImgCombinedPipeline, KandinskyImg2ImgPipeline, KandinskyInpaintCombinedPipeline, KandinskyInpaintPipeline, KandinskyPipeline, KandinskyPriorPipeline, ) from .kandinsky2_2 import ( KandinskyV22CombinedPipeline, KandinskyV22ControlnetImg2ImgPipeline, KandinskyV22ControlnetPipeline, KandinskyV22Img2ImgCombinedPipeline, KandinskyV22Img2ImgPipeline, KandinskyV22InpaintCombinedPipeline, KandinskyV22InpaintPipeline, KandinskyV22Pipeline, KandinskyV22PriorEmb2EmbPipeline, KandinskyV22PriorPipeline, ) from .kandinsky3 import ( Kandinsky3Img2ImgPipeline, Kandinsky3Pipeline, ) from .latent_consistency_models import ( LatentConsistencyModelImg2ImgPipeline, LatentConsistencyModelPipeline, ) from .latent_diffusion import LDMTextToImagePipeline from .musicldm import MusicLDMPipeline from .paint_by_example import PaintByExamplePipeline from .pia import PIAPipeline from .pixart_alpha import PixArtAlphaPipeline from .semantic_stable_diffusion import SemanticStableDiffusionPipeline from .shap_e import ShapEImg2ImgPipeline, ShapEPipeline from .stable_diffusion import ( CLIPImageProjection, StableDiffusionDepth2ImgPipeline, StableDiffusionImageVariationPipeline, StableDiffusionImg2ImgPipeline, StableDiffusionInpaintPipeline, StableDiffusionInstructPix2PixPipeline, StableDiffusionLatentUpscalePipeline, StableDiffusionPipeline, StableDiffusionUpscalePipeline, StableUnCLIPImg2ImgPipeline, StableUnCLIPPipeline, ) from .stable_diffusion_attend_and_excite import StableDiffusionAttendAndExcitePipeline from .stable_diffusion_diffedit import StableDiffusionDiffEditPipeline from .stable_diffusion_gligen import StableDiffusionGLIGENPipeline, StableDiffusionGLIGENTextImagePipeline from .stable_diffusion_ldm3d import StableDiffusionLDM3DPipeline from .stable_diffusion_panorama import StableDiffusionPanoramaPipeline from .stable_diffusion_safe import StableDiffusionPipelineSafe from .stable_diffusion_sag import StableDiffusionSAGPipeline from .stable_diffusion_xl import ( StableDiffusionXLImg2ImgPipeline, StableDiffusionXLInpaintPipeline, StableDiffusionXLInstructPix2PixPipeline, StableDiffusionXLPipeline, ) from .stable_video_diffusion import StableVideoDiffusionPipeline from .t2i_adapter import ( StableDiffusionAdapterPipeline, StableDiffusionXLAdapterPipeline, ) from .text_to_video_synthesis import ( TextToVideoSDPipeline, TextToVideoZeroPipeline, TextToVideoZeroSDXLPipeline, VideoToVideoSDPipeline, ) from .unclip import UnCLIPImageVariationPipeline, UnCLIPPipeline from .unidiffuser import ( ImageTextPipelineOutput, UniDiffuserModel, UniDiffuserPipeline, UniDiffuserTextDecoder, ) from .wuerstchen import ( WuerstchenCombinedPipeline, WuerstchenDecoderPipeline, WuerstchenPriorPipeline, ) try: if not is_onnx_available(): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils.dummy_onnx_objects import * # noqa F403 else: from .onnx_utils import OnnxRuntimeModel try: if not (is_torch_available() and is_transformers_available() and is_onnx_available()): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils.dummy_torch_and_transformers_and_onnx_objects import * else: from .stable_diffusion import ( OnnxStableDiffusionImg2ImgPipeline, OnnxStableDiffusionInpaintPipeline, OnnxStableDiffusionPipeline, OnnxStableDiffusionUpscalePipeline, StableDiffusionOnnxPipeline, ) try: if not (is_torch_available() and is_transformers_available() and is_k_diffusion_available()): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils.dummy_torch_and_transformers_and_k_diffusion_objects import * else: from .stable_diffusion_k_diffusion import ( StableDiffusionKDiffusionPipeline, StableDiffusionXLKDiffusionPipeline, ) try: if not is_flax_available(): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils.dummy_flax_objects import * # noqa F403 else: from .pipeline_flax_utils import FlaxDiffusionPipeline try: if not (is_flax_available() and is_transformers_available()): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils.dummy_flax_and_transformers_objects import * else: from .controlnet import FlaxStableDiffusionControlNetPipeline from .stable_diffusion import ( FlaxStableDiffusionImg2ImgPipeline, FlaxStableDiffusionInpaintPipeline, FlaxStableDiffusionPipeline, ) from .stable_diffusion_xl import ( FlaxStableDiffusionXLPipeline, ) try: if not (is_transformers_available() and is_torch_available() and is_note_seq_available()): raise OptionalDependencyNotAvailable() except OptionalDependencyNotAvailable: from ..utils.dummy_transformers_and_torch_and_note_seq_objects import * # noqa F403 else: from .deprecated import ( MidiProcessor, SpectrogramDiffusionPipeline, ) else: import sys sys.modules[__name__] = _LazyModule( __name__, globals()["__file__"], _import_structure, module_spec=__spec__, ) for name, value in _dummy_objects.items(): setattr(sys.modules[__name__], name, value)

diffusers/src/diffusers/pipelines/__init__.py/0

from typing import TYPE_CHECKING from ...utils import DIFFUSERS_SLOW_IMPORT, _LazyModule _import_structure = {"pipeline_dance_diffusion": ["DanceDiffusionPipeline"]} if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT: from .pipeline_dance_diffusion import DanceDiffusionPipeline else: import sys sys.modules[__name__] = _LazyModule( __name__, globals()["__file__"], _import_structure, module_spec=__spec__, )

diffusers/src/diffusers/pipelines/dance_diffusion/__init__.py/0

from typing import List import PIL.Image import torch from PIL import Image from ...configuration_utils import ConfigMixin from ...models.modeling_utils import ModelMixin from ...utils import PIL_INTERPOLATION class IFWatermarker(ModelMixin, ConfigMixin): def __init__(self): super().__init__() self.register_buffer("watermark_image", torch.zeros((62, 62, 4))) self.watermark_image_as_pil = None def apply_watermark(self, images: List[PIL.Image.Image], sample_size=None): # copied from https://github.com/deep-floyd/IF/blob/b77482e36ca2031cb94dbca1001fc1e6400bf4ab/deepfloyd_if/modules/base.py#L287 h = images[0].height w = images[0].width sample_size = sample_size or h coef = min(h / sample_size, w / sample_size) img_h, img_w = (int(h / coef), int(w / coef)) if coef < 1 else (h, w) S1, S2 = 1024**2, img_w * img_h K = (S2 / S1) ** 0.5 wm_size, wm_x, wm_y = int(K * 62), img_w - int(14 * K), img_h - int(14 * K) if self.watermark_image_as_pil is None: watermark_image = self.watermark_image.to(torch.uint8).cpu().numpy() watermark_image = Image.fromarray(watermark_image, mode="RGBA") self.watermark_image_as_pil = watermark_image wm_img = self.watermark_image_as_pil.resize( (wm_size, wm_size), PIL_INTERPOLATION["bicubic"], reducing_gap=None ) for pil_img in images: pil_img.paste(wm_img, box=(wm_x - wm_size, wm_y - wm_size, wm_x, wm_y), mask=wm_img.split()[-1]) return images

diffusers/src/diffusers/pipelines/deepfloyd_if/watermark.py/0

# Copyright 2023 ETH Zurich Computer Vision Lab and The HuggingFace Team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from typing import List, Optional, Tuple, Union import numpy as np import PIL.Image import torch from ....models import UNet2DModel from ....schedulers import RePaintScheduler from ....utils import PIL_INTERPOLATION, deprecate, logging from ....utils.torch_utils import randn_tensor from ...pipeline_utils import DiffusionPipeline, ImagePipelineOutput logger = logging.get_logger(__name__) # pylint: disable=invalid-name # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.preprocess def _preprocess_image(image: Union[List, PIL.Image.Image, torch.Tensor]): deprecation_message = "The preprocess method is deprecated and will be removed in diffusers 1.0.0. Please use VaeImageProcessor.preprocess(...) instead" deprecate("preprocess", "1.0.0", deprecation_message, standard_warn=False) if isinstance(image, torch.Tensor): return image elif isinstance(image, PIL.Image.Image): image = [image] if isinstance(image[0], PIL.Image.Image): w, h = image[0].size w, h = (x - x % 8 for x in (w, h)) # resize to integer multiple of 8 image = [np.array(i.resize((w, h), resample=PIL_INTERPOLATION["lanczos"]))[None, :] for i in image] image = np.concatenate(image, axis=0) image = np.array(image).astype(np.float32) / 255.0 image = image.transpose(0, 3, 1, 2) image = 2.0 * image - 1.0 image = torch.from_numpy(image) elif isinstance(image[0], torch.Tensor): image = torch.cat(image, dim=0) return image def _preprocess_mask(mask: Union[List, PIL.Image.Image, torch.Tensor]): if isinstance(mask, torch.Tensor): return mask elif isinstance(mask, PIL.Image.Image): mask = [mask] if isinstance(mask[0], PIL.Image.Image): w, h = mask[0].size w, h = (x - x % 32 for x in (w, h)) # resize to integer multiple of 32 mask = [np.array(m.convert("L").resize((w, h), resample=PIL_INTERPOLATION["nearest"]))[None, :] for m in mask] mask = np.concatenate(mask, axis=0) mask = mask.astype(np.float32) / 255.0 mask[mask < 0.5] = 0 mask[mask >= 0.5] = 1 mask = torch.from_numpy(mask) elif isinstance(mask[0], torch.Tensor): mask = torch.cat(mask, dim=0) return mask class RePaintPipeline(DiffusionPipeline): r""" Pipeline for image inpainting using RePaint. This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a particular device, etc.). Parameters: unet ([`UNet2DModel`]): A `UNet2DModel` to denoise the encoded image latents. scheduler ([`RePaintScheduler`]): A `RePaintScheduler` to be used in combination with `unet` to denoise the encoded image. """ unet: UNet2DModel scheduler: RePaintScheduler model_cpu_offload_seq = "unet" def __init__(self, unet, scheduler): super().__init__() self.register_modules(unet=unet, scheduler=scheduler) @torch.no_grad() def __call__( self, image: Union[torch.Tensor, PIL.Image.Image], mask_image: Union[torch.Tensor, PIL.Image.Image], num_inference_steps: int = 250, eta: float = 0.0, jump_length: int = 10, jump_n_sample: int = 10, generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, output_type: Optional[str] = "pil", return_dict: bool = True, ) -> Union[ImagePipelineOutput, Tuple]: r""" The call function to the pipeline for generation. Args: image (`torch.FloatTensor` or `PIL.Image.Image`): The original image to inpaint on. mask_image (`torch.FloatTensor` or `PIL.Image.Image`): The mask_image where 0.0 define which part of the original image to inpaint. num_inference_steps (`int`, *optional*, defaults to 1000): The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. eta (`float`): The weight of the added noise in a diffusion step. Its value is between 0.0 and 1.0; 0.0 corresponds to DDIM and 1.0 is the DDPM scheduler. jump_length (`int`, *optional*, defaults to 10): The number of steps taken forward in time before going backward in time for a single jump ("j" in RePaint paper). Take a look at Figure 9 and 10 in the [paper](https://arxiv.org/pdf/2201.09865.pdf). jump_n_sample (`int`, *optional*, defaults to 10): The number of times to make a forward time jump for a given chosen time sample. Take a look at Figure 9 and 10 in the [paper](https://arxiv.org/pdf/2201.09865.pdf). generator (`torch.Generator`, *optional*): A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation deterministic. output_type (`str`, `optional`, defaults to `"pil"`): The output format of the generated image. Choose between `PIL.Image` or `np.array`. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`ImagePipelineOutput`] instead of a plain tuple. Example: ```py >>> from io import BytesIO >>> import torch >>> import PIL >>> import requests >>> from diffusers import RePaintPipeline, RePaintScheduler >>> def download_image(url): ... response = requests.get(url) ... return PIL.Image.open(BytesIO(response.content)).convert("RGB") >>> img_url = "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/repaint/celeba_hq_256.png" >>> mask_url = "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/repaint/mask_256.png" >>> # Load the original image and the mask as PIL images >>> original_image = download_image(img_url).resize((256, 256)) >>> mask_image = download_image(mask_url).resize((256, 256)) >>> # Load the RePaint scheduler and pipeline based on a pretrained DDPM model >>> scheduler = RePaintScheduler.from_pretrained("google/ddpm-ema-celebahq-256") >>> pipe = RePaintPipeline.from_pretrained("google/ddpm-ema-celebahq-256", scheduler=scheduler) >>> pipe = pipe.to("cuda") >>> generator = torch.Generator(device="cuda").manual_seed(0) >>> output = pipe( ... image=original_image, ... mask_image=mask_image, ... num_inference_steps=250, ... eta=0.0, ... jump_length=10, ... jump_n_sample=10, ... generator=generator, ... ) >>> inpainted_image = output.images[0] ``` Returns: [`~pipelines.ImagePipelineOutput`] or `tuple`: If `return_dict` is `True`, [`~pipelines.ImagePipelineOutput`] is returned, otherwise a `tuple` is returned where the first element is a list with the generated images. """ original_image = image original_image = _preprocess_image(original_image) original_image = original_image.to(device=self._execution_device, dtype=self.unet.dtype) mask_image = _preprocess_mask(mask_image) mask_image = mask_image.to(device=self._execution_device, dtype=self.unet.dtype) batch_size = original_image.shape[0] # sample gaussian noise to begin the loop if isinstance(generator, list) and len(generator) != batch_size: raise ValueError( f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" f" size of {batch_size}. Make sure the batch size matches the length of the generators." ) image_shape = original_image.shape image = randn_tensor(image_shape, generator=generator, device=self._execution_device, dtype=self.unet.dtype) # set step values self.scheduler.set_timesteps(num_inference_steps, jump_length, jump_n_sample, self._execution_device) self.scheduler.eta = eta t_last = self.scheduler.timesteps[0] + 1 generator = generator[0] if isinstance(generator, list) else generator for i, t in enumerate(self.progress_bar(self.scheduler.timesteps)): if t < t_last: # predict the noise residual model_output = self.unet(image, t).sample # compute previous image: x_t -> x_t-1 image = self.scheduler.step(model_output, t, image, original_image, mask_image, generator).prev_sample else: # compute the reverse: x_t-1 -> x_t image = self.scheduler.undo_step(image, t_last, generator) t_last = t image = (image / 2 + 0.5).clamp(0, 1) image = image.cpu().permute(0, 2, 3, 1).numpy() if output_type == "pil": image = self.numpy_to_pil(image) if not return_dict: return (image,) return ImagePipelineOutput(images=image)

diffusers/src/diffusers/pipelines/deprecated/repaint/pipeline_repaint.py/0

# Copyright 2023 The HuggingFace Team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from typing import List, Optional, Tuple, Union import torch from ....models import UNet2DModel from ....schedulers import KarrasVeScheduler from ....utils.torch_utils import randn_tensor from ...pipeline_utils import DiffusionPipeline, ImagePipelineOutput class KarrasVePipeline(DiffusionPipeline): r""" Pipeline for unconditional image generation. Parameters: unet ([`UNet2DModel`]): A `UNet2DModel` to denoise the encoded image. scheduler ([`KarrasVeScheduler`]): A scheduler to be used in combination with `unet` to denoise the encoded image. """ # add type hints for linting unet: UNet2DModel scheduler: KarrasVeScheduler def __init__(self, unet: UNet2DModel, scheduler: KarrasVeScheduler): super().__init__() self.register_modules(unet=unet, scheduler=scheduler) @torch.no_grad() def __call__( self, batch_size: int = 1, num_inference_steps: int = 50, generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, output_type: Optional[str] = "pil", return_dict: bool = True, **kwargs, ) -> Union[Tuple, ImagePipelineOutput]: r""" The call function to the pipeline for generation. Args: batch_size (`int`, *optional*, defaults to 1): The number of images to generate. generator (`torch.Generator`, *optional*): A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation deterministic. num_inference_steps (`int`, *optional*, defaults to 50): The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. output_type (`str`, *optional*, defaults to `"pil"`): The output format of the generated image. Choose between `PIL.Image` or `np.array`. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`ImagePipelineOutput`] instead of a plain tuple. Example: Returns: [`~pipelines.ImagePipelineOutput`] or `tuple`: If `return_dict` is `True`, [`~pipelines.ImagePipelineOutput`] is returned, otherwise a `tuple` is returned where the first element is a list with the generated images. """ img_size = self.unet.config.sample_size shape = (batch_size, 3, img_size, img_size) model = self.unet # sample x_0 ~ N(0, sigma_0^2 * I) sample = randn_tensor(shape, generator=generator, device=self.device) * self.scheduler.init_noise_sigma self.scheduler.set_timesteps(num_inference_steps) for t in self.progress_bar(self.scheduler.timesteps): # here sigma_t == t_i from the paper sigma = self.scheduler.schedule[t] sigma_prev = self.scheduler.schedule[t - 1] if t > 0 else 0 # 1. Select temporarily increased noise level sigma_hat # 2. Add new noise to move from sample_i to sample_hat sample_hat, sigma_hat = self.scheduler.add_noise_to_input(sample, sigma, generator=generator) # 3. Predict the noise residual given the noise magnitude `sigma_hat` # The model inputs and output are adjusted by following eq. (213) in [1]. model_output = (sigma_hat / 2) * model((sample_hat + 1) / 2, sigma_hat / 2).sample # 4. Evaluate dx/dt at sigma_hat # 5. Take Euler step from sigma to sigma_prev step_output = self.scheduler.step(model_output, sigma_hat, sigma_prev, sample_hat) if sigma_prev != 0: # 6. Apply 2nd order correction # The model inputs and output are adjusted by following eq. (213) in [1]. model_output = (sigma_prev / 2) * model((step_output.prev_sample + 1) / 2, sigma_prev / 2).sample step_output = self.scheduler.step_correct( model_output, sigma_hat, sigma_prev, sample_hat, step_output.prev_sample, step_output["derivative"], ) sample = step_output.prev_sample sample = (sample / 2 + 0.5).clamp(0, 1) image = sample.cpu().permute(0, 2, 3, 1).numpy() if output_type == "pil": image = self.numpy_to_pil(image) if not return_dict: return (image,) return ImagePipelineOutput(images=image)

diffusers/src/diffusers/pipelines/deprecated/stochastic_karras_ve/pipeline_stochastic_karras_ve.py/0

# Copyright 2023 The HuggingFace Team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from typing import Callable, List, Optional, Union import numpy as np import PIL.Image import torch from PIL import Image from transformers import ( XLMRobertaTokenizer, ) from ...models import UNet2DConditionModel, VQModel from ...schedulers import DDIMScheduler from ...utils import ( logging, replace_example_docstring, ) from ...utils.torch_utils import randn_tensor from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput from .text_encoder import MultilingualCLIP logger = logging.get_logger(__name__) # pylint: disable=invalid-name EXAMPLE_DOC_STRING = """ Examples: ```py >>> from diffusers import KandinskyImg2ImgPipeline, KandinskyPriorPipeline >>> from diffusers.utils import load_image >>> import torch >>> pipe_prior = KandinskyPriorPipeline.from_pretrained( ... "kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16 ... ) >>> pipe_prior.to("cuda") >>> prompt = "A red cartoon frog, 4k" >>> image_emb, zero_image_emb = pipe_prior(prompt, return_dict=False) >>> pipe = KandinskyImg2ImgPipeline.from_pretrained( ... "kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16 ... ) >>> pipe.to("cuda") >>> init_image = load_image( ... "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main" ... "/kandinsky/frog.png" ... ) >>> image = pipe( ... prompt, ... image=init_image, ... image_embeds=image_emb, ... negative_image_embeds=zero_image_emb, ... height=768, ... width=768, ... num_inference_steps=100, ... strength=0.2, ... ).images >>> image[0].save("red_frog.png") ``` """ def get_new_h_w(h, w, scale_factor=8): new_h = h // scale_factor**2 if h % scale_factor**2 != 0: new_h += 1 new_w = w // scale_factor**2 if w % scale_factor**2 != 0: new_w += 1 return new_h * scale_factor, new_w * scale_factor def prepare_image(pil_image, w=512, h=512): pil_image = pil_image.resize((w, h), resample=Image.BICUBIC, reducing_gap=1) arr = np.array(pil_image.convert("RGB")) arr = arr.astype(np.float32) / 127.5 - 1 arr = np.transpose(arr, [2, 0, 1]) image = torch.from_numpy(arr).unsqueeze(0) return image class KandinskyImg2ImgPipeline(DiffusionPipeline): """ Pipeline for image-to-image generation using Kandinsky This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.) Args: text_encoder ([`MultilingualCLIP`]): Frozen text-encoder. tokenizer ([`XLMRobertaTokenizer`]): Tokenizer of class scheduler ([`DDIMScheduler`]): A scheduler to be used in combination with `unet` to generate image latents. unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the image embedding. movq ([`VQModel`]): MoVQ image encoder and decoder """ model_cpu_offload_seq = "text_encoder->unet->movq" def __init__( self, text_encoder: MultilingualCLIP, movq: VQModel, tokenizer: XLMRobertaTokenizer, unet: UNet2DConditionModel, scheduler: DDIMScheduler, ): super().__init__() self.register_modules( text_encoder=text_encoder, tokenizer=tokenizer, unet=unet, scheduler=scheduler, movq=movq, ) self.movq_scale_factor = 2 ** (len(self.movq.config.block_out_channels) - 1) def get_timesteps(self, num_inference_steps, strength, device): # get the original timestep using init_timestep init_timestep = min(int(num_inference_steps * strength), num_inference_steps) t_start = max(num_inference_steps - init_timestep, 0) timesteps = self.scheduler.timesteps[t_start:] return timesteps, num_inference_steps - t_start def prepare_latents(self, latents, latent_timestep, shape, dtype, device, generator, scheduler): if latents is None: latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype) else: if latents.shape != shape: raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {shape}") latents = latents.to(device) latents = latents * scheduler.init_noise_sigma shape = latents.shape noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype) latents = self.add_noise(latents, noise, latent_timestep) return latents def _encode_prompt( self, prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt=None, ): batch_size = len(prompt) if isinstance(prompt, list) else 1 # get prompt text embeddings text_inputs = self.tokenizer( prompt, padding="max_length", max_length=77, truncation=True, return_attention_mask=True, add_special_tokens=True, return_tensors="pt", ) text_input_ids = text_inputs.input_ids untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids): removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]) logger.warning( "The following part of your input was truncated because CLIP can only handle sequences up to" f" {self.tokenizer.model_max_length} tokens: {removed_text}" ) text_input_ids = text_input_ids.to(device) text_mask = text_inputs.attention_mask.to(device) prompt_embeds, text_encoder_hidden_states = self.text_encoder( input_ids=text_input_ids, attention_mask=text_mask ) prompt_embeds = prompt_embeds.repeat_interleave(num_images_per_prompt, dim=0) text_encoder_hidden_states = text_encoder_hidden_states.repeat_interleave(num_images_per_prompt, dim=0) text_mask = text_mask.repeat_interleave(num_images_per_prompt, dim=0) if do_classifier_free_guidance: uncond_tokens: List[str] if negative_prompt is None: uncond_tokens = [""] * batch_size elif type(prompt) is not type(negative_prompt): raise TypeError( f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !=" f" {type(prompt)}." ) elif isinstance(negative_prompt, str): uncond_tokens = [negative_prompt] elif batch_size != len(negative_prompt): raise ValueError( f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:" f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches" " the batch size of `prompt`." ) else: uncond_tokens = negative_prompt uncond_input = self.tokenizer( uncond_tokens, padding="max_length", max_length=77, truncation=True, return_attention_mask=True, add_special_tokens=True, return_tensors="pt", ) uncond_text_input_ids = uncond_input.input_ids.to(device) uncond_text_mask = uncond_input.attention_mask.to(device) negative_prompt_embeds, uncond_text_encoder_hidden_states = self.text_encoder( input_ids=uncond_text_input_ids, attention_mask=uncond_text_mask ) # duplicate unconditional embeddings for each generation per prompt, using mps friendly method seq_len = negative_prompt_embeds.shape[1] negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt) negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len) seq_len = uncond_text_encoder_hidden_states.shape[1] uncond_text_encoder_hidden_states = uncond_text_encoder_hidden_states.repeat(1, num_images_per_prompt, 1) uncond_text_encoder_hidden_states = uncond_text_encoder_hidden_states.view( batch_size * num_images_per_prompt, seq_len, -1 ) uncond_text_mask = uncond_text_mask.repeat_interleave(num_images_per_prompt, dim=0) # done duplicates # For classifier free guidance, we need to do two forward passes. # Here we concatenate the unconditional and text embeddings into a single batch # to avoid doing two forward passes prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds]) text_encoder_hidden_states = torch.cat([uncond_text_encoder_hidden_states, text_encoder_hidden_states]) text_mask = torch.cat([uncond_text_mask, text_mask]) return prompt_embeds, text_encoder_hidden_states, text_mask # add_noise method to overwrite the one in schedule because it use a different beta schedule for adding noise vs sampling def add_noise( self, original_samples: torch.FloatTensor, noise: torch.FloatTensor, timesteps: torch.IntTensor, ) -> torch.FloatTensor: betas = torch.linspace(0.0001, 0.02, 1000, dtype=torch.float32) alphas = 1.0 - betas alphas_cumprod = torch.cumprod(alphas, dim=0) alphas_cumprod = alphas_cumprod.to(device=original_samples.device, dtype=original_samples.dtype) timesteps = timesteps.to(original_samples.device) sqrt_alpha_prod = alphas_cumprod[timesteps] ** 0.5 sqrt_alpha_prod = sqrt_alpha_prod.flatten() while len(sqrt_alpha_prod.shape) < len(original_samples.shape): sqrt_alpha_prod = sqrt_alpha_prod.unsqueeze(-1) sqrt_one_minus_alpha_prod = (1 - alphas_cumprod[timesteps]) ** 0.5 sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.flatten() while len(sqrt_one_minus_alpha_prod.shape) < len(original_samples.shape): sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.unsqueeze(-1) noisy_samples = sqrt_alpha_prod * original_samples + sqrt_one_minus_alpha_prod * noise return noisy_samples @torch.no_grad() @replace_example_docstring(EXAMPLE_DOC_STRING) def __call__( self, prompt: Union[str, List[str]], image: Union[torch.FloatTensor, PIL.Image.Image, List[torch.FloatTensor], List[PIL.Image.Image]], image_embeds: torch.FloatTensor, negative_image_embeds: torch.FloatTensor, negative_prompt: Optional[Union[str, List[str]]] = None, height: int = 512, width: int = 512, num_inference_steps: int = 100, strength: float = 0.3, guidance_scale: float = 7.0, num_images_per_prompt: int = 1, generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, output_type: Optional[str] = "pil", callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None, callback_steps: int = 1, return_dict: bool = True, ): """ Function invoked when calling the pipeline for generation. Args: prompt (`str` or `List[str]`): The prompt or prompts to guide the image generation. image (`torch.FloatTensor`, `PIL.Image.Image`): `Image`, or tensor representing an image batch, that will be used as the starting point for the process. image_embeds (`torch.FloatTensor` or `List[torch.FloatTensor]`): The clip image embeddings for text prompt, that will be used to condition the image generation. negative_image_embeds (`torch.FloatTensor` or `List[torch.FloatTensor]`): The clip image embeddings for negative text prompt, will be used to condition the image generation. negative_prompt (`str` or `List[str]`, *optional*): The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`). height (`int`, *optional*, defaults to 512): The height in pixels of the generated image. width (`int`, *optional*, defaults to 512): The width in pixels of the generated image. num_inference_steps (`int`, *optional*, defaults to 100): The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. strength (`float`, *optional*, defaults to 0.3): Conceptually, indicates how much to transform the reference `image`. Must be between 0 and 1. `image` will be used as a starting point, adding more noise to it the larger the `strength`. The number of denoising steps depends on the amount of noise initially added. When `strength` is 1, added noise will be maximum and the denoising process will run for the full number of iterations specified in `num_inference_steps`. A value of 1, therefore, essentially ignores `image`. guidance_scale (`float`, *optional*, defaults to 4.0): Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598). `guidance_scale` is defined as `w` of equation 2. of [Imagen Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`, usually at the expense of lower image quality. num_images_per_prompt (`int`, *optional*, defaults to 1): The number of images to generate per prompt. generator (`torch.Generator` or `List[torch.Generator]`, *optional*): One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation deterministic. output_type (`str`, *optional*, defaults to `"pil"`): The output format of the generate image. Choose between: `"pil"` (`PIL.Image.Image`), `"np"` (`np.array`) or `"pt"` (`torch.Tensor`). callback (`Callable`, *optional*): A function that calls every `callback_steps` steps during inference. The function is called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`. callback_steps (`int`, *optional*, defaults to 1): The frequency at which the `callback` function is called. If not specified, the callback is called at every step. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`~pipelines.ImagePipelineOutput`] instead of a plain tuple. Examples: Returns: [`~pipelines.ImagePipelineOutput`] or `tuple` """ # 1. Define call parameters if isinstance(prompt, str): batch_size = 1 elif isinstance(prompt, list): batch_size = len(prompt) else: raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") device = self._execution_device batch_size = batch_size * num_images_per_prompt do_classifier_free_guidance = guidance_scale > 1.0 # 2. get text and image embeddings prompt_embeds, text_encoder_hidden_states, _ = self._encode_prompt( prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt ) if isinstance(image_embeds, list): image_embeds = torch.cat(image_embeds, dim=0) if isinstance(negative_image_embeds, list): negative_image_embeds = torch.cat(negative_image_embeds, dim=0) if do_classifier_free_guidance: image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0) negative_image_embeds = negative_image_embeds.repeat_interleave(num_images_per_prompt, dim=0) image_embeds = torch.cat([negative_image_embeds, image_embeds], dim=0).to( dtype=prompt_embeds.dtype, device=device ) # 3. pre-processing initial image if not isinstance(image, list): image = [image] if not all(isinstance(i, (PIL.Image.Image, torch.Tensor)) for i in image): raise ValueError( f"Input is in incorrect format: {[type(i) for i in image]}. Currently, we only support PIL image and pytorch tensor" ) image = torch.cat([prepare_image(i, width, height) for i in image], dim=0) image = image.to(dtype=prompt_embeds.dtype, device=device) latents = self.movq.encode(image)["latents"] latents = latents.repeat_interleave(num_images_per_prompt, dim=0) # 4. set timesteps self.scheduler.set_timesteps(num_inference_steps, device=device) timesteps_tensor, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device) # the formular to calculate timestep for add_noise is taken from the original kandinsky repo latent_timestep = int(self.scheduler.config.num_train_timesteps * strength) - 2 latent_timestep = torch.tensor([latent_timestep] * batch_size, dtype=timesteps_tensor.dtype, device=device) num_channels_latents = self.unet.config.in_channels height, width = get_new_h_w(height, width, self.movq_scale_factor) # 5. Create initial latent latents = self.prepare_latents( latents, latent_timestep, (batch_size, num_channels_latents, height, width), text_encoder_hidden_states.dtype, device, generator, self.scheduler, ) # 6. Denoising loop for i, t in enumerate(self.progress_bar(timesteps_tensor)): # expand the latents if we are doing classifier free guidance latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents added_cond_kwargs = {"text_embeds": prompt_embeds, "image_embeds": image_embeds} noise_pred = self.unet( sample=latent_model_input, timestep=t, encoder_hidden_states=text_encoder_hidden_states, added_cond_kwargs=added_cond_kwargs, return_dict=False, )[0] if do_classifier_free_guidance: noise_pred, variance_pred = noise_pred.split(latents.shape[1], dim=1) noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) _, variance_pred_text = variance_pred.chunk(2) noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond) noise_pred = torch.cat([noise_pred, variance_pred_text], dim=1) if not ( hasattr(self.scheduler.config, "variance_type") and self.scheduler.config.variance_type in ["learned", "learned_range"] ): noise_pred, _ = noise_pred.split(latents.shape[1], dim=1) # compute the previous noisy sample x_t -> x_t-1 latents = self.scheduler.step( noise_pred, t, latents, generator=generator, ).prev_sample if callback is not None and i % callback_steps == 0: step_idx = i // getattr(self.scheduler, "order", 1) callback(step_idx, t, latents) # 7. post-processing image = self.movq.decode(latents, force_not_quantize=True)["sample"] self.maybe_free_model_hooks() if output_type not in ["pt", "np", "pil"]: raise ValueError(f"Only the output types `pt`, `pil` and `np` are supported not output_type={output_type}") if output_type in ["np", "pil"]: image = image * 0.5 + 0.5 image = image.clamp(0, 1) image = image.cpu().permute(0, 2, 3, 1).float().numpy() if output_type == "pil": image = self.numpy_to_pil(image) if not return_dict: return (image,) return ImagePipelineOutput(images=image)

diffusers/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_img2img.py/0

import inspect from typing import Callable, Dict, List, Optional, Union import numpy as np import PIL import PIL.Image import torch from transformers import T5EncoderModel, T5Tokenizer from ...loaders import LoraLoaderMixin from ...models import Kandinsky3UNet, VQModel from ...schedulers import DDPMScheduler from ...utils import ( deprecate, is_accelerate_available, logging, replace_example_docstring, ) from ...utils.torch_utils import randn_tensor from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput logger = logging.get_logger(__name__) # pylint: disable=invalid-name EXAMPLE_DOC_STRING = """ Examples: ```py >>> from diffusers import AutoPipelineForImage2Image >>> from diffusers.utils import load_image >>> import torch >>> pipe = AutoPipelineForImage2Image.from_pretrained("kandinsky-community/kandinsky-3", variant="fp16", torch_dtype=torch.float16) >>> pipe.enable_model_cpu_offload() >>> prompt = "A painting of the inside of a subway train with tiny raccoons." >>> image = load_image("https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinsky3/t2i.png") >>> generator = torch.Generator(device="cpu").manual_seed(0) >>> image = pipe(prompt, image=image, strength=0.75, num_inference_steps=25, generator=generator).images[0] ``` """ def downscale_height_and_width(height, width, scale_factor=8): new_height = height // scale_factor**2 if height % scale_factor**2 != 0: new_height += 1 new_width = width // scale_factor**2 if width % scale_factor**2 != 0: new_width += 1 return new_height * scale_factor, new_width * scale_factor def prepare_image(pil_image): arr = np.array(pil_image.convert("RGB")) arr = arr.astype(np.float32) / 127.5 - 1 arr = np.transpose(arr, [2, 0, 1]) image = torch.from_numpy(arr).unsqueeze(0) return image class Kandinsky3Img2ImgPipeline(DiffusionPipeline, LoraLoaderMixin): model_cpu_offload_seq = "text_encoder->movq->unet->movq" _callback_tensor_inputs = [ "latents", "prompt_embeds", "negative_prompt_embeds", "negative_attention_mask", "attention_mask", ] def __init__( self, tokenizer: T5Tokenizer, text_encoder: T5EncoderModel, unet: Kandinsky3UNet, scheduler: DDPMScheduler, movq: VQModel, ): super().__init__() self.register_modules( tokenizer=tokenizer, text_encoder=text_encoder, unet=unet, scheduler=scheduler, movq=movq ) def get_timesteps(self, num_inference_steps, strength, device): # get the original timestep using init_timestep init_timestep = min(int(num_inference_steps * strength), num_inference_steps) t_start = max(num_inference_steps - init_timestep, 0) timesteps = self.scheduler.timesteps[t_start:] return timesteps, num_inference_steps - t_start def remove_all_hooks(self): if is_accelerate_available(): from accelerate.hooks import remove_hook_from_module else: raise ImportError("Please install accelerate via `pip install accelerate`") for model in [self.text_encoder, self.unet]: if model is not None: remove_hook_from_module(model, recurse=True) self.unet_offload_hook = None self.text_encoder_offload_hook = None self.final_offload_hook = None def _process_embeds(self, embeddings, attention_mask, cut_context): # return embeddings, attention_mask if cut_context: embeddings[attention_mask == 0] = torch.zeros_like(embeddings[attention_mask == 0]) max_seq_length = attention_mask.sum(-1).max() + 1 embeddings = embeddings[:, :max_seq_length] attention_mask = attention_mask[:, :max_seq_length] return embeddings, attention_mask @torch.no_grad() def encode_prompt( self, prompt, do_classifier_free_guidance=True, num_images_per_prompt=1, device=None, negative_prompt=None, prompt_embeds: Optional[torch.FloatTensor] = None, negative_prompt_embeds: Optional[torch.FloatTensor] = None, _cut_context=False, attention_mask: Optional[torch.FloatTensor] = None, negative_attention_mask: Optional[torch.FloatTensor] = None, ): r""" Encodes the prompt into text encoder hidden states. Args: prompt (`str` or `List[str]`, *optional*): prompt to be encoded device: (`torch.device`, *optional*): torch device to place the resulting embeddings on num_images_per_prompt (`int`, *optional*, defaults to 1): number of images that should be generated per prompt do_classifier_free_guidance (`bool`, *optional*, defaults to `True`): whether to use classifier free guidance or not negative_prompt (`str` or `List[str]`, *optional*): The prompt or prompts not to guide the image generation. If not defined, one has to pass `negative_prompt_embeds`. instead. If not defined, one has to pass `negative_prompt_embeds`. instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`). prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, text embeddings will be generated from `prompt` input argument. negative_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input argument. attention_mask (`torch.FloatTensor`, *optional*): Pre-generated attention mask. Must provide if passing `prompt_embeds` directly. negative_attention_mask (`torch.FloatTensor`, *optional*): Pre-generated negative attention mask. Must provide if passing `negative_prompt_embeds` directly. """ if prompt is not None and negative_prompt is not None: if type(prompt) is not type(negative_prompt): raise TypeError( f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !=" f" {type(prompt)}." ) if device is None: device = self._execution_device if prompt is not None and isinstance(prompt, str): batch_size = 1 elif prompt is not None and isinstance(prompt, list): batch_size = len(prompt) else: batch_size = prompt_embeds.shape[0] max_length = 128 if prompt_embeds is None: text_inputs = self.tokenizer( prompt, padding="max_length", max_length=max_length, truncation=True, return_tensors="pt", ) text_input_ids = text_inputs.input_ids.to(device) attention_mask = text_inputs.attention_mask.to(device) prompt_embeds = self.text_encoder( text_input_ids, attention_mask=attention_mask, ) prompt_embeds = prompt_embeds[0] prompt_embeds, attention_mask = self._process_embeds(prompt_embeds, attention_mask, _cut_context) prompt_embeds = prompt_embeds * attention_mask.unsqueeze(2) if self.text_encoder is not None: dtype = self.text_encoder.dtype else: dtype = None prompt_embeds = prompt_embeds.to(dtype=dtype, device=device) bs_embed, seq_len, _ = prompt_embeds.shape # duplicate text embeddings for each generation per prompt, using mps friendly method prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1) prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1) attention_mask = attention_mask.repeat(num_images_per_prompt, 1) # get unconditional embeddings for classifier free guidance if do_classifier_free_guidance and negative_prompt_embeds is None: uncond_tokens: List[str] if negative_prompt is None: uncond_tokens = [""] * batch_size elif isinstance(negative_prompt, str): uncond_tokens = [negative_prompt] elif batch_size != len(negative_prompt): raise ValueError( f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:" f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches" " the batch size of `prompt`." ) else: uncond_tokens = negative_prompt if negative_prompt is not None: uncond_input = self.tokenizer( uncond_tokens, padding="max_length", max_length=128, truncation=True, return_attention_mask=True, return_tensors="pt", ) text_input_ids = uncond_input.input_ids.to(device) negative_attention_mask = uncond_input.attention_mask.to(device) negative_prompt_embeds = self.text_encoder( text_input_ids, attention_mask=negative_attention_mask, ) negative_prompt_embeds = negative_prompt_embeds[0] negative_prompt_embeds = negative_prompt_embeds[:, : prompt_embeds.shape[1]] negative_attention_mask = negative_attention_mask[:, : prompt_embeds.shape[1]] negative_prompt_embeds = negative_prompt_embeds * negative_attention_mask.unsqueeze(2) else: negative_prompt_embeds = torch.zeros_like(prompt_embeds) negative_attention_mask = torch.zeros_like(attention_mask) if do_classifier_free_guidance: # duplicate unconditional embeddings for each generation per prompt, using mps friendly method seq_len = negative_prompt_embeds.shape[1] negative_prompt_embeds = negative_prompt_embeds.to(dtype=dtype, device=device) if negative_prompt_embeds.shape != prompt_embeds.shape: negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1) negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1) negative_attention_mask = negative_attention_mask.repeat(num_images_per_prompt, 1) # For classifier free guidance, we need to do two forward passes. # Here we concatenate the unconditional and text embeddings into a single batch # to avoid doing two forward passes else: negative_prompt_embeds = None negative_attention_mask = None return prompt_embeds, negative_prompt_embeds, attention_mask, negative_attention_mask def prepare_latents(self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None): if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)): raise ValueError( f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}" ) image = image.to(device=device, dtype=dtype) batch_size = batch_size * num_images_per_prompt if image.shape[1] == 4: init_latents = image else: if isinstance(generator, list) and len(generator) != batch_size: raise ValueError( f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" f" size of {batch_size}. Make sure the batch size matches the length of the generators." ) elif isinstance(generator, list): init_latents = [ self.movq.encode(image[i : i + 1]).latent_dist.sample(generator[i]) for i in range(batch_size) ] init_latents = torch.cat(init_latents, dim=0) else: init_latents = self.movq.encode(image).latent_dist.sample(generator) init_latents = self.movq.config.scaling_factor * init_latents init_latents = torch.cat([init_latents], dim=0) shape = init_latents.shape noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype) # get latents init_latents = self.scheduler.add_noise(init_latents, noise, timestep) latents = init_latents return latents # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs def prepare_extra_step_kwargs(self, generator, eta): # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers. # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502 # and should be between [0, 1] accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys()) extra_step_kwargs = {} if accepts_eta: extra_step_kwargs["eta"] = eta # check if the scheduler accepts generator accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys()) if accepts_generator: extra_step_kwargs["generator"] = generator return extra_step_kwargs def check_inputs( self, prompt, callback_steps, negative_prompt=None, prompt_embeds=None, negative_prompt_embeds=None, callback_on_step_end_tensor_inputs=None, attention_mask=None, negative_attention_mask=None, ): if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0): raise ValueError( f"`callback_steps` has to be a positive integer but is {callback_steps} of type" f" {type(callback_steps)}." ) if callback_on_step_end_tensor_inputs is not None and not all( k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs ): raise ValueError( f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" ) if prompt is not None and prompt_embeds is not None: raise ValueError( f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to" " only forward one of the two." ) elif prompt is None and prompt_embeds is None: raise ValueError( "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined." ) elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)): raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") if negative_prompt is not None and negative_prompt_embeds is not None: raise ValueError( f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:" f" {negative_prompt_embeds}. Please make sure to only forward one of the two." ) if prompt_embeds is not None and negative_prompt_embeds is not None: if prompt_embeds.shape != negative_prompt_embeds.shape: raise ValueError( "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but" f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`" f" {negative_prompt_embeds.shape}." ) if negative_prompt_embeds is not None and negative_attention_mask is None: raise ValueError("Please provide `negative_attention_mask` along with `negative_prompt_embeds`") if negative_prompt_embeds is not None and negative_attention_mask is not None: if negative_prompt_embeds.shape[:2] != negative_attention_mask.shape: raise ValueError( "`negative_prompt_embeds` and `negative_attention_mask` must have the same batch_size and token length when passed directly, but" f" got: `negative_prompt_embeds` {negative_prompt_embeds.shape[:2]} != `negative_attention_mask`" f" {negative_attention_mask.shape}." ) if prompt_embeds is not None and attention_mask is None: raise ValueError("Please provide `attention_mask` along with `prompt_embeds`") if prompt_embeds is not None and attention_mask is not None: if prompt_embeds.shape[:2] != attention_mask.shape: raise ValueError( "`prompt_embeds` and `attention_mask` must have the same batch_size and token length when passed directly, but" f" got: `prompt_embeds` {prompt_embeds.shape[:2]} != `attention_mask`" f" {attention_mask.shape}." ) @property def guidance_scale(self): return self._guidance_scale @property def do_classifier_free_guidance(self): return self._guidance_scale > 1 @property def num_timesteps(self): return self._num_timesteps @torch.no_grad() @replace_example_docstring(EXAMPLE_DOC_STRING) def __call__( self, prompt: Union[str, List[str]] = None, image: Union[torch.FloatTensor, PIL.Image.Image, List[torch.FloatTensor], List[PIL.Image.Image]] = None, strength: float = 0.3, num_inference_steps: int = 25, guidance_scale: float = 3.0, negative_prompt: Optional[Union[str, List[str]]] = None, num_images_per_prompt: Optional[int] = 1, generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, prompt_embeds: Optional[torch.FloatTensor] = None, negative_prompt_embeds: Optional[torch.FloatTensor] = None, attention_mask: Optional[torch.FloatTensor] = None, negative_attention_mask: Optional[torch.FloatTensor] = None, output_type: Optional[str] = "pil", return_dict: bool = True, callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None, callback_on_step_end_tensor_inputs: List[str] = ["latents"], **kwargs, ): """ Function invoked when calling the pipeline for generation. Args: prompt (`str` or `List[str]`, *optional*): The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`. instead. image (`torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.FloatTensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]`): `Image`, or tensor representing an image batch, that will be used as the starting point for the process. strength (`float`, *optional*, defaults to 0.8): Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a starting point and more noise is added the higher the `strength`. The number of denoising steps depends on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising process runs for the full number of iterations specified in `num_inference_steps`. A value of 1 essentially ignores `image`. num_inference_steps (`int`, *optional*, defaults to 50): The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. guidance_scale (`float`, *optional*, defaults to 3.0): Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598). `guidance_scale` is defined as `w` of equation 2. of [Imagen Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`, usually at the expense of lower image quality. negative_prompt (`str` or `List[str]`, *optional*): The prompt or prompts not to guide the image generation. If not defined, one has to pass `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`). num_images_per_prompt (`int`, *optional*, defaults to 1): The number of images to generate per prompt. generator (`torch.Generator` or `List[torch.Generator]`, *optional*): One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation deterministic. prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, text embeddings will be generated from `prompt` input argument. negative_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input argument. attention_mask (`torch.FloatTensor`, *optional*): Pre-generated attention mask. Must provide if passing `prompt_embeds` directly. negative_attention_mask (`torch.FloatTensor`, *optional*): Pre-generated negative attention mask. Must provide if passing `negative_prompt_embeds` directly. output_type (`str`, *optional*, defaults to `"pil"`): The output format of the generate image. Choose between [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`~pipelines.stable_diffusion.IFPipelineOutput`] instead of a plain tuple. callback_on_step_end (`Callable`, *optional*): A function that calls at the end of each denoising steps during the inference. The function is called with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by `callback_on_step_end_tensor_inputs`. callback_on_step_end_tensor_inputs (`List`, *optional*): The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the `._callback_tensor_inputs` attribute of your pipeline class. Examples: Returns: [`~pipelines.ImagePipelineOutput`] or `tuple` """ callback = kwargs.pop("callback", None) callback_steps = kwargs.pop("callback_steps", None) if callback is not None: deprecate( "callback", "1.0.0", "Passing `callback` as an input argument to `__call__` is deprecated, consider use `callback_on_step_end`", ) if callback_steps is not None: deprecate( "callback_steps", "1.0.0", "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider use `callback_on_step_end`", ) if callback_on_step_end_tensor_inputs is not None and not all( k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs ): raise ValueError( f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" ) cut_context = True # 1. Check inputs. Raise error if not correct self.check_inputs( prompt, callback_steps, negative_prompt, prompt_embeds, negative_prompt_embeds, callback_on_step_end_tensor_inputs, attention_mask, negative_attention_mask, ) self._guidance_scale = guidance_scale if prompt is not None and isinstance(prompt, str): batch_size = 1 elif prompt is not None and isinstance(prompt, list): batch_size = len(prompt) else: batch_size = prompt_embeds.shape[0] device = self._execution_device # 3. Encode input prompt prompt_embeds, negative_prompt_embeds, attention_mask, negative_attention_mask = self.encode_prompt( prompt, self.do_classifier_free_guidance, num_images_per_prompt=num_images_per_prompt, device=device, negative_prompt=negative_prompt, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, _cut_context=cut_context, attention_mask=attention_mask, negative_attention_mask=negative_attention_mask, ) if self.do_classifier_free_guidance: prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds]) attention_mask = torch.cat([negative_attention_mask, attention_mask]).bool() if not isinstance(image, list): image = [image] if not all(isinstance(i, (PIL.Image.Image, torch.Tensor)) for i in image): raise ValueError( f"Input is in incorrect format: {[type(i) for i in image]}. Currently, we only support PIL image and pytorch tensor" ) image = torch.cat([prepare_image(i) for i in image], dim=0) image = image.to(dtype=prompt_embeds.dtype, device=device) # 4. Prepare timesteps self.scheduler.set_timesteps(num_inference_steps, device=device) timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device) # 5. Prepare latents latents = self.movq.encode(image)["latents"] latents = latents.repeat_interleave(num_images_per_prompt, dim=0) latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt) latents = self.prepare_latents( latents, latent_timestep, batch_size, num_images_per_prompt, prompt_embeds.dtype, device, generator ) if hasattr(self, "text_encoder_offload_hook") and self.text_encoder_offload_hook is not None: self.text_encoder_offload_hook.offload() # 7. Denoising loop num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order self._num_timesteps = len(timesteps) with self.progress_bar(total=num_inference_steps) as progress_bar: for i, t in enumerate(timesteps): latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents # predict the noise residual noise_pred = self.unet( latent_model_input, t, encoder_hidden_states=prompt_embeds, encoder_attention_mask=attention_mask, )[0] if self.do_classifier_free_guidance: noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) noise_pred = (guidance_scale + 1.0) * noise_pred_text - guidance_scale * noise_pred_uncond # compute the previous noisy sample x_t -> x_t-1 latents = self.scheduler.step( noise_pred, t, latents, generator=generator, ).prev_sample if callback_on_step_end is not None: callback_kwargs = {} for k in callback_on_step_end_tensor_inputs: callback_kwargs[k] = locals()[k] callback_outputs = callback_on_step_end(self, i, t, callback_kwargs) latents = callback_outputs.pop("latents", latents) prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds) negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds) attention_mask = callback_outputs.pop("attention_mask", attention_mask) negative_attention_mask = callback_outputs.pop("negative_attention_mask", negative_attention_mask) if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): progress_bar.update() if callback is not None and i % callback_steps == 0: step_idx = i // getattr(self.scheduler, "order", 1) callback(step_idx, t, latents) # post-processing if output_type not in ["pt", "np", "pil", "latent"]: raise ValueError( f"Only the output types `pt`, `pil`, `np` and `latent` are supported not output_type={output_type}" ) if not output_type == "latent": image = self.movq.decode(latents, force_not_quantize=True)["sample"] if output_type in ["np", "pil"]: image = image * 0.5 + 0.5 image = image.clamp(0, 1) image = image.cpu().permute(0, 2, 3, 1).float().numpy() if output_type == "pil": image = self.numpy_to_pil(image) else: image = latents self.maybe_free_model_hooks() if not return_dict: return (image,) return ImagePipelineOutput(images=image)

diffusers/src/diffusers/pipelines/kandinsky3/pipeline_kandinsky3_img2img.py/0

# coding=utf-8 # Copyright 2023 The HuggingFace Inc. team. # Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import fnmatch import importlib import inspect import os import re import sys import warnings from dataclasses import dataclass from pathlib import Path from typing import Any, Callable, Dict, List, Optional, Union import numpy as np import PIL.Image import requests import torch from huggingface_hub import ( ModelCard, create_repo, hf_hub_download, model_info, snapshot_download, ) from huggingface_hub.utils import OfflineModeIsEnabled, validate_hf_hub_args from packaging import version from requests.exceptions import HTTPError from tqdm.auto import tqdm from .. import __version__ from ..configuration_utils import ConfigMixin from ..models.modeling_utils import _LOW_CPU_MEM_USAGE_DEFAULT from ..schedulers.scheduling_utils import SCHEDULER_CONFIG_NAME from ..utils import ( CONFIG_NAME, DEPRECATED_REVISION_ARGS, SAFETENSORS_WEIGHTS_NAME, WEIGHTS_NAME, BaseOutput, deprecate, get_class_from_dynamic_module, is_accelerate_available, is_accelerate_version, is_peft_available, is_torch_version, is_transformers_available, logging, numpy_to_pil, ) from ..utils.hub_utils import load_or_create_model_card, populate_model_card from ..utils.torch_utils import is_compiled_module if is_transformers_available(): import transformers from transformers import PreTrainedModel from transformers.utils import FLAX_WEIGHTS_NAME as TRANSFORMERS_FLAX_WEIGHTS_NAME from transformers.utils import SAFE_WEIGHTS_NAME as TRANSFORMERS_SAFE_WEIGHTS_NAME from transformers.utils import WEIGHTS_NAME as TRANSFORMERS_WEIGHTS_NAME from ..utils import FLAX_WEIGHTS_NAME, ONNX_EXTERNAL_WEIGHTS_NAME, ONNX_WEIGHTS_NAME, PushToHubMixin if is_accelerate_available(): import accelerate INDEX_FILE = "diffusion_pytorch_model.bin" CUSTOM_PIPELINE_FILE_NAME = "pipeline.py" DUMMY_MODULES_FOLDER = "diffusers.utils" TRANSFORMERS_DUMMY_MODULES_FOLDER = "transformers.utils" CONNECTED_PIPES_KEYS = ["prior"] logger = logging.get_logger(__name__) LOADABLE_CLASSES = { "diffusers": { "ModelMixin": ["save_pretrained", "from_pretrained"], "SchedulerMixin": ["save_pretrained", "from_pretrained"], "DiffusionPipeline": ["save_pretrained", "from_pretrained"], "OnnxRuntimeModel": ["save_pretrained", "from_pretrained"], }, "transformers": { "PreTrainedTokenizer": ["save_pretrained", "from_pretrained"], "PreTrainedTokenizerFast": ["save_pretrained", "from_pretrained"], "PreTrainedModel": ["save_pretrained", "from_pretrained"], "FeatureExtractionMixin": ["save_pretrained", "from_pretrained"], "ProcessorMixin": ["save_pretrained", "from_pretrained"], "ImageProcessingMixin": ["save_pretrained", "from_pretrained"], }, "onnxruntime.training": { "ORTModule": ["save_pretrained", "from_pretrained"], }, } ALL_IMPORTABLE_CLASSES = {} for library in LOADABLE_CLASSES: ALL_IMPORTABLE_CLASSES.update(LOADABLE_CLASSES[library]) @dataclass class ImagePipelineOutput(BaseOutput): """ Output class for image pipelines. Args: images (`List[PIL.Image.Image]` or `np.ndarray`) List of denoised PIL images of length `batch_size` or NumPy array of shape `(batch_size, height, width, num_channels)`. """ images: Union[List[PIL.Image.Image], np.ndarray] @dataclass class AudioPipelineOutput(BaseOutput): """ Output class for audio pipelines. Args: audios (`np.ndarray`) List of denoised audio samples of a NumPy array of shape `(batch_size, num_channels, sample_rate)`. """ audios: np.ndarray def is_safetensors_compatible(filenames, variant=None, passed_components=None) -> bool: """ Checking for safetensors compatibility: - By default, all models are saved with the default pytorch serialization, so we use the list of default pytorch files to know which safetensors files are needed. - The model is safetensors compatible only if there is a matching safetensors file for every default pytorch file. Converting default pytorch serialized filenames to safetensors serialized filenames: - For models from the diffusers library, just replace the ".bin" extension with ".safetensors" - For models from the transformers library, the filename changes from "pytorch_model" to "model", and the ".bin" extension is replaced with ".safetensors" """ pt_filenames = [] sf_filenames = set() passed_components = passed_components or [] for filename in filenames: _, extension = os.path.splitext(filename) if len(filename.split("/")) == 2 and filename.split("/")[0] in passed_components: continue if extension == ".bin": pt_filenames.append(os.path.normpath(filename)) elif extension == ".safetensors": sf_filenames.add(os.path.normpath(filename)) for filename in pt_filenames: # filename = 'foo/bar/baz.bam' -> path = 'foo/bar', filename = 'baz', extention = '.bam' path, filename = os.path.split(filename) filename, extension = os.path.splitext(filename) if filename.startswith("pytorch_model"): filename = filename.replace("pytorch_model", "model") else: filename = filename expected_sf_filename = os.path.normpath(os.path.join(path, filename)) expected_sf_filename = f"{expected_sf_filename}.safetensors" if expected_sf_filename not in sf_filenames: logger.warning(f"{expected_sf_filename} not found") return False return True def variant_compatible_siblings(filenames, variant=None) -> Union[List[os.PathLike], str]: weight_names = [ WEIGHTS_NAME, SAFETENSORS_WEIGHTS_NAME, FLAX_WEIGHTS_NAME, ONNX_WEIGHTS_NAME, ONNX_EXTERNAL_WEIGHTS_NAME, ] if is_transformers_available(): weight_names += [TRANSFORMERS_WEIGHTS_NAME, TRANSFORMERS_SAFE_WEIGHTS_NAME, TRANSFORMERS_FLAX_WEIGHTS_NAME] # model_pytorch, diffusion_model_pytorch, ... weight_prefixes = [w.split(".")[0] for w in weight_names] # .bin, .safetensors, ... weight_suffixs = [w.split(".")[-1] for w in weight_names] # -00001-of-00002 transformers_index_format = r"\d{5}-of-\d{5}" if variant is not None: # `diffusion_pytorch_model.fp16.bin` as well as `model.fp16-00001-of-00002.safetensors` variant_file_re = re.compile( rf"({'|'.join(weight_prefixes)})\.({variant}|{variant}-{transformers_index_format})\.({'|'.join(weight_suffixs)})$" ) # `text_encoder/pytorch_model.bin.index.fp16.json` variant_index_re = re.compile( rf"({'|'.join(weight_prefixes)})\.({'|'.join(weight_suffixs)})\.index\.{variant}\.json$" ) # `diffusion_pytorch_model.bin` as well as `model-00001-of-00002.safetensors` non_variant_file_re = re.compile( rf"({'|'.join(weight_prefixes)})(-{transformers_index_format})?\.({'|'.join(weight_suffixs)})$" ) # `text_encoder/pytorch_model.bin.index.json` non_variant_index_re = re.compile(rf"({'|'.join(weight_prefixes)})\.({'|'.join(weight_suffixs)})\.index\.json") if variant is not None: variant_weights = {f for f in filenames if variant_file_re.match(f.split("/")[-1]) is not None} variant_indexes = {f for f in filenames if variant_index_re.match(f.split("/")[-1]) is not None} variant_filenames = variant_weights | variant_indexes else: variant_filenames = set() non_variant_weights = {f for f in filenames if non_variant_file_re.match(f.split("/")[-1]) is not None} non_variant_indexes = {f for f in filenames if non_variant_index_re.match(f.split("/")[-1]) is not None} non_variant_filenames = non_variant_weights | non_variant_indexes # all variant filenames will be used by default usable_filenames = set(variant_filenames) def convert_to_variant(filename): if "index" in filename: variant_filename = filename.replace("index", f"index.{variant}") elif re.compile(f"^(.*?){transformers_index_format}").match(filename) is not None: variant_filename = f"{filename.split('-')[0]}.{variant}-{'-'.join(filename.split('-')[1:])}" else: variant_filename = f"{filename.split('.')[0]}.{variant}.{filename.split('.')[1]}" return variant_filename for f in non_variant_filenames: variant_filename = convert_to_variant(f) if variant_filename not in usable_filenames: usable_filenames.add(f) return usable_filenames, variant_filenames @validate_hf_hub_args def warn_deprecated_model_variant(pretrained_model_name_or_path, token, variant, revision, model_filenames): info = model_info( pretrained_model_name_or_path, token=token, revision=None, ) filenames = {sibling.rfilename for sibling in info.siblings} comp_model_filenames, _ = variant_compatible_siblings(filenames, variant=revision) comp_model_filenames = [".".join(f.split(".")[:1] + f.split(".")[2:]) for f in comp_model_filenames] if set(model_filenames).issubset(set(comp_model_filenames)): warnings.warn( f"You are loading the variant {revision} from {pretrained_model_name_or_path} via `revision='{revision}'` even though you can load it via `variant=`{revision}`. Loading model variants via `revision='{revision}'` is deprecated and will be removed in diffusers v1. Please use `variant='{revision}'` instead.", FutureWarning, ) else: warnings.warn( f"You are loading the variant {revision} from {pretrained_model_name_or_path} via `revision='{revision}'`. This behavior is deprecated and will be removed in diffusers v1. One should use `variant='{revision}'` instead. However, it appears that {pretrained_model_name_or_path} currently does not have the required variant filenames in the 'main' branch. \n The Diffusers team and community would be very grateful if you could open an issue: https://github.com/huggingface/diffusers/issues/new with the title '{pretrained_model_name_or_path} is missing {revision} files' so that the correct variant file can be added.", FutureWarning, ) def _unwrap_model(model): """Unwraps a model.""" if is_compiled_module(model): model = model._orig_mod if is_peft_available(): from peft import PeftModel if isinstance(model, PeftModel): model = model.base_model.model return model def maybe_raise_or_warn( library_name, library, class_name, importable_classes, passed_class_obj, name, is_pipeline_module ): """Simple helper method to raise or warn in case incorrect module has been passed""" if not is_pipeline_module: library = importlib.import_module(library_name) class_obj = getattr(library, class_name) class_candidates = {c: getattr(library, c, None) for c in importable_classes.keys()} expected_class_obj = None for class_name, class_candidate in class_candidates.items(): if class_candidate is not None and issubclass(class_obj, class_candidate): expected_class_obj = class_candidate # Dynamo wraps the original model in a private class. # I didn't find a public API to get the original class. sub_model = passed_class_obj[name] unwrapped_sub_model = _unwrap_model(sub_model) model_cls = unwrapped_sub_model.__class__ if not issubclass(model_cls, expected_class_obj): raise ValueError( f"{passed_class_obj[name]} is of type: {model_cls}, but should be" f" {expected_class_obj}" ) else: logger.warning( f"You have passed a non-standard module {passed_class_obj[name]}. We cannot verify whether it" " has the correct type" ) def get_class_obj_and_candidates( library_name, class_name, importable_classes, pipelines, is_pipeline_module, component_name=None, cache_dir=None ): """Simple helper method to retrieve class object of module as well as potential parent class objects""" component_folder = os.path.join(cache_dir, component_name) if is_pipeline_module: pipeline_module = getattr(pipelines, library_name) class_obj = getattr(pipeline_module, class_name) class_candidates = {c: class_obj for c in importable_classes.keys()} elif os.path.isfile(os.path.join(component_folder, library_name + ".py")): # load custom component class_obj = get_class_from_dynamic_module( component_folder, module_file=library_name + ".py", class_name=class_name ) class_candidates = {c: class_obj for c in importable_classes.keys()} else: # else we just import it from the library. library = importlib.import_module(library_name) class_obj = getattr(library, class_name) class_candidates = {c: getattr(library, c, None) for c in importable_classes.keys()} return class_obj, class_candidates def _get_pipeline_class( class_obj, config=None, load_connected_pipeline=False, custom_pipeline=None, repo_id=None, hub_revision=None, class_name=None, cache_dir=None, revision=None, ): if custom_pipeline is not None: if custom_pipeline.endswith(".py"): path = Path(custom_pipeline) # decompose into folder & file file_name = path.name custom_pipeline = path.parent.absolute() elif repo_id is not None: file_name = f"{custom_pipeline}.py" custom_pipeline = repo_id else: file_name = CUSTOM_PIPELINE_FILE_NAME if repo_id is not None and hub_revision is not None: # if we load the pipeline code from the Hub # make sure to overwrite the `revison` revision = hub_revision return get_class_from_dynamic_module( custom_pipeline, module_file=file_name, class_name=class_name, cache_dir=cache_dir, revision=revision, ) if class_obj != DiffusionPipeline: return class_obj diffusers_module = importlib.import_module(class_obj.__module__.split(".")[0]) class_name = class_name or config["_class_name"] if not class_name: raise ValueError( "The class name could not be found in the configuration file. Please make sure to pass the correct `class_name`." ) class_name = class_name[4:] if class_name.startswith("Flax") else class_name pipeline_cls = getattr(diffusers_module, class_name) if load_connected_pipeline: from .auto_pipeline import _get_connected_pipeline connected_pipeline_cls = _get_connected_pipeline(pipeline_cls) if connected_pipeline_cls is not None: logger.info( f"Loading connected pipeline {connected_pipeline_cls.__name__} instead of {pipeline_cls.__name__} as specified via `load_connected_pipeline=True`" ) else: logger.info(f"{pipeline_cls.__name__} has no connected pipeline class. Loading {pipeline_cls.__name__}.") pipeline_cls = connected_pipeline_cls or pipeline_cls return pipeline_cls def load_sub_model( library_name: str, class_name: str, importable_classes: List[Any], pipelines: Any, is_pipeline_module: bool, pipeline_class: Any, torch_dtype: torch.dtype, provider: Any, sess_options: Any, device_map: Optional[Union[Dict[str, torch.device], str]], max_memory: Optional[Dict[Union[int, str], Union[int, str]]], offload_folder: Optional[Union[str, os.PathLike]], offload_state_dict: bool, model_variants: Dict[str, str], name: str, from_flax: bool, variant: str, low_cpu_mem_usage: bool, cached_folder: Union[str, os.PathLike], revision: str = None, ): """Helper method to load the module `name` from `library_name` and `class_name`""" # retrieve class candidates class_obj, class_candidates = get_class_obj_and_candidates( library_name, class_name, importable_classes, pipelines, is_pipeline_module, component_name=name, cache_dir=cached_folder, ) load_method_name = None # retrive load method name for class_name, class_candidate in class_candidates.items(): if class_candidate is not None and issubclass(class_obj, class_candidate): load_method_name = importable_classes[class_name][1] # if load method name is None, then we have a dummy module -> raise Error if load_method_name is None: none_module = class_obj.__module__ is_dummy_path = none_module.startswith(DUMMY_MODULES_FOLDER) or none_module.startswith( TRANSFORMERS_DUMMY_MODULES_FOLDER ) if is_dummy_path and "dummy" in none_module: # call class_obj for nice error message of missing requirements class_obj() raise ValueError( f"The component {class_obj} of {pipeline_class} cannot be loaded as it does not seem to have" f" any of the loading methods defined in {ALL_IMPORTABLE_CLASSES}." ) load_method = getattr(class_obj, load_method_name) # add kwargs to loading method diffusers_module = importlib.import_module(__name__.split(".")[0]) loading_kwargs = {} if issubclass(class_obj, torch.nn.Module): loading_kwargs["torch_dtype"] = torch_dtype if issubclass(class_obj, diffusers_module.OnnxRuntimeModel): loading_kwargs["provider"] = provider loading_kwargs["sess_options"] = sess_options is_diffusers_model = issubclass(class_obj, diffusers_module.ModelMixin) if is_transformers_available(): transformers_version = version.parse(version.parse(transformers.__version__).base_version) else: transformers_version = "N/A" is_transformers_model = ( is_transformers_available() and issubclass(class_obj, PreTrainedModel) and transformers_version >= version.parse("4.20.0") ) # When loading a transformers model, if the device_map is None, the weights will be initialized as opposed to diffusers. # To make default loading faster we set the `low_cpu_mem_usage=low_cpu_mem_usage` flag which is `True` by default. # This makes sure that the weights won't be initialized which significantly speeds up loading. if is_diffusers_model or is_transformers_model: loading_kwargs["device_map"] = device_map loading_kwargs["max_memory"] = max_memory loading_kwargs["offload_folder"] = offload_folder loading_kwargs["offload_state_dict"] = offload_state_dict loading_kwargs["variant"] = model_variants.pop(name, None) if from_flax: loading_kwargs["from_flax"] = True # the following can be deleted once the minimum required `transformers` version # is higher than 4.27 if ( is_transformers_model and loading_kwargs["variant"] is not None and transformers_version < version.parse("4.27.0") ): raise ImportError( f"When passing `variant='{variant}'`, please make sure to upgrade your `transformers` version to at least 4.27.0.dev0" ) elif is_transformers_model and loading_kwargs["variant"] is None: loading_kwargs.pop("variant") # if `from_flax` and model is transformer model, can currently not load with `low_cpu_mem_usage` if not (from_flax and is_transformers_model): loading_kwargs["low_cpu_mem_usage"] = low_cpu_mem_usage else: loading_kwargs["low_cpu_mem_usage"] = False # check if the module is in a subdirectory if os.path.isdir(os.path.join(cached_folder, name)): loaded_sub_model = load_method(os.path.join(cached_folder, name), **loading_kwargs) else: # else load from the root directory loaded_sub_model = load_method(cached_folder, **loading_kwargs) return loaded_sub_model def _fetch_class_library_tuple(module): # import it here to avoid circular import diffusers_module = importlib.import_module(__name__.split(".")[0]) pipelines = getattr(diffusers_module, "pipelines") # register the config from the original module, not the dynamo compiled one not_compiled_module = _unwrap_model(module) library = not_compiled_module.__module__.split(".")[0] # check if the module is a pipeline module module_path_items = not_compiled_module.__module__.split(".") pipeline_dir = module_path_items[-2] if len(module_path_items) > 2 else None path = not_compiled_module.__module__.split(".") is_pipeline_module = pipeline_dir in path and hasattr(pipelines, pipeline_dir) # if library is not in LOADABLE_CLASSES, then it is a custom module. # Or if it's a pipeline module, then the module is inside the pipeline # folder so we set the library to module name. if is_pipeline_module: library = pipeline_dir elif library not in LOADABLE_CLASSES: library = not_compiled_module.__module__ # retrieve class_name class_name = not_compiled_module.__class__.__name__ return (library, class_name) class DiffusionPipeline(ConfigMixin, PushToHubMixin): r""" Base class for all pipelines. [`DiffusionPipeline`] stores all components (models, schedulers, and processors) for diffusion pipelines and provides methods for loading, downloading and saving models. It also includes methods to: - move all PyTorch modules to the device of your choice - enable/disable the progress bar for the denoising iteration Class attributes: - **config_name** (`str`) -- The configuration filename that stores the class and module names of all the diffusion pipeline's components. - **_optional_components** (`List[str]`) -- List of all optional components that don't have to be passed to the pipeline to function (should be overridden by subclasses). """ config_name = "model_index.json" model_cpu_offload_seq = None _optional_components = [] _exclude_from_cpu_offload = [] _load_connected_pipes = False _is_onnx = False def register_modules(self, **kwargs): for name, module in kwargs.items(): # retrieve library if module is None or isinstance(module, (tuple, list)) and module[0] is None: register_dict = {name: (None, None)} else: library, class_name = _fetch_class_library_tuple(module) register_dict = {name: (library, class_name)} # save model index config self.register_to_config(**register_dict) # set models setattr(self, name, module) def __setattr__(self, name: str, value: Any): if name in self.__dict__ and hasattr(self.config, name): # We need to overwrite the config if name exists in config if isinstance(getattr(self.config, name), (tuple, list)): if value is not None and self.config[name][0] is not None: class_library_tuple = _fetch_class_library_tuple(value) else: class_library_tuple = (None, None) self.register_to_config(**{name: class_library_tuple}) else: self.register_to_config(**{name: value}) super().__setattr__(name, value) def save_pretrained( self, save_directory: Union[str, os.PathLike], safe_serialization: bool = True, variant: Optional[str] = None, push_to_hub: bool = False, **kwargs, ): """ Save all saveable variables of the pipeline to a directory. A pipeline variable can be saved and loaded if its class implements both a save and loading method. The pipeline is easily reloaded using the [`~DiffusionPipeline.from_pretrained`] class method. Arguments: save_directory (`str` or `os.PathLike`): Directory to save a pipeline to. Will be created if it doesn't exist. safe_serialization (`bool`, *optional*, defaults to `True`): Whether to save the model using `safetensors` or the traditional PyTorch way with `pickle`. variant (`str`, *optional*): If specified, weights are saved in the format `pytorch_model.<variant>.bin`. push_to_hub (`bool`, *optional*, defaults to `False`): Whether or not to push your model to the Hugging Face model hub after saving it. You can specify the repository you want to push to with `repo_id` (will default to the name of `save_directory` in your namespace). kwargs (`Dict[str, Any]`, *optional*): Additional keyword arguments passed along to the [`~utils.PushToHubMixin.push_to_hub`] method. """ model_index_dict = dict(self.config) model_index_dict.pop("_class_name", None) model_index_dict.pop("_diffusers_version", None) model_index_dict.pop("_module", None) model_index_dict.pop("_name_or_path", None) if push_to_hub: commit_message = kwargs.pop("commit_message", None) private = kwargs.pop("private", False) create_pr = kwargs.pop("create_pr", False) token = kwargs.pop("token", None) repo_id = kwargs.pop("repo_id", save_directory.split(os.path.sep)[-1]) repo_id = create_repo(repo_id, exist_ok=True, private=private, token=token).repo_id expected_modules, optional_kwargs = self._get_signature_keys(self) def is_saveable_module(name, value): if name not in expected_modules: return False if name in self._optional_components and value[0] is None: return False return True model_index_dict = {k: v for k, v in model_index_dict.items() if is_saveable_module(k, v)} for pipeline_component_name in model_index_dict.keys(): sub_model = getattr(self, pipeline_component_name) model_cls = sub_model.__class__ # Dynamo wraps the original model in a private class. # I didn't find a public API to get the original class. if is_compiled_module(sub_model): sub_model = _unwrap_model(sub_model) model_cls = sub_model.__class__ save_method_name = None # search for the model's base class in LOADABLE_CLASSES for library_name, library_classes in LOADABLE_CLASSES.items(): if library_name in sys.modules: library = importlib.import_module(library_name) else: logger.info( f"{library_name} is not installed. Cannot save {pipeline_component_name} as {library_classes} from {library_name}" ) for base_class, save_load_methods in library_classes.items(): class_candidate = getattr(library, base_class, None) if class_candidate is not None and issubclass(model_cls, class_candidate): # if we found a suitable base class in LOADABLE_CLASSES then grab its save method save_method_name = save_load_methods[0] break if save_method_name is not None: break if save_method_name is None: logger.warn(f"self.{pipeline_component_name}={sub_model} of type {type(sub_model)} cannot be saved.") # make sure that unsaveable components are not tried to be loaded afterward self.register_to_config(**{pipeline_component_name: (None, None)}) continue save_method = getattr(sub_model, save_method_name) # Call the save method with the argument safe_serialization only if it's supported save_method_signature = inspect.signature(save_method) save_method_accept_safe = "safe_serialization" in save_method_signature.parameters save_method_accept_variant = "variant" in save_method_signature.parameters save_kwargs = {} if save_method_accept_safe: save_kwargs["safe_serialization"] = safe_serialization if save_method_accept_variant: save_kwargs["variant"] = variant save_method(os.path.join(save_directory, pipeline_component_name), **save_kwargs) # finally save the config self.save_config(save_directory) if push_to_hub: # Create a new empty model card and eventually tag it model_card = load_or_create_model_card(repo_id, token=token, is_pipeline=True) model_card = populate_model_card(model_card) model_card.save(os.path.join(save_directory, "README.md")) self._upload_folder( save_directory, repo_id, token=token, commit_message=commit_message, create_pr=create_pr, ) def to(self, *args, **kwargs): r""" Performs Pipeline dtype and/or device conversion. A torch.dtype and torch.device are inferred from the arguments of `self.to(*args, **kwargs).` <Tip> If the pipeline already has the correct torch.dtype and torch.device, then it is returned as is. Otherwise, the returned pipeline is a copy of self with the desired torch.dtype and torch.device. </Tip> Here are the ways to call `to`: - `to(dtype, silence_dtype_warnings=False) → DiffusionPipeline` to return a pipeline with the specified [`dtype`](https://pytorch.org/docs/stable/tensor_attributes.html#torch.dtype) - `to(device, silence_dtype_warnings=False) → DiffusionPipeline` to return a pipeline with the specified [`device`](https://pytorch.org/docs/stable/tensor_attributes.html#torch.device) - `to(device=None, dtype=None, silence_dtype_warnings=False) → DiffusionPipeline` to return a pipeline with the specified [`device`](https://pytorch.org/docs/stable/tensor_attributes.html#torch.device) and [`dtype`](https://pytorch.org/docs/stable/tensor_attributes.html#torch.dtype) Arguments: dtype (`torch.dtype`, *optional*): Returns a pipeline with the specified [`dtype`](https://pytorch.org/docs/stable/tensor_attributes.html#torch.dtype) device (`torch.Device`, *optional*): Returns a pipeline with the specified [`device`](https://pytorch.org/docs/stable/tensor_attributes.html#torch.device) silence_dtype_warnings (`str`, *optional*, defaults to `False`): Whether to omit warnings if the target `dtype` is not compatible with the target `device`. Returns: [`DiffusionPipeline`]: The pipeline converted to specified `dtype` and/or `dtype`. """ torch_dtype = kwargs.pop("torch_dtype", None) if torch_dtype is not None: deprecate("torch_dtype", "0.27.0", "") torch_device = kwargs.pop("torch_device", None) if torch_device is not None: deprecate("torch_device", "0.27.0", "") dtype_kwarg = kwargs.pop("dtype", None) device_kwarg = kwargs.pop("device", None) silence_dtype_warnings = kwargs.pop("silence_dtype_warnings", False) if torch_dtype is not None and dtype_kwarg is not None: raise ValueError( "You have passed both `torch_dtype` and `dtype` as a keyword argument. Please make sure to only pass `dtype`." ) dtype = torch_dtype or dtype_kwarg if torch_device is not None and device_kwarg is not None: raise ValueError( "You have passed both `torch_device` and `device` as a keyword argument. Please make sure to only pass `device`." ) device = torch_device or device_kwarg dtype_arg = None device_arg = None if len(args) == 1: if isinstance(args[0], torch.dtype): dtype_arg = args[0] else: device_arg = torch.device(args[0]) if args[0] is not None else None elif len(args) == 2: if isinstance(args[0], torch.dtype): raise ValueError( "When passing two arguments, make sure the first corresponds to `device` and the second to `dtype`." ) device_arg = torch.device(args[0]) if args[0] is not None else None dtype_arg = args[1] elif len(args) > 2: raise ValueError("Please make sure to pass at most two arguments (`device` and `dtype`) `.to(...)`") if dtype is not None and dtype_arg is not None: raise ValueError( "You have passed `dtype` both as an argument and as a keyword argument. Please only pass one of the two." ) dtype = dtype or dtype_arg if device is not None and device_arg is not None: raise ValueError( "You have passed `device` both as an argument and as a keyword argument. Please only pass one of the two." ) device = device or device_arg # throw warning if pipeline is in "offloaded"-mode but user tries to manually set to GPU. def module_is_sequentially_offloaded(module): if not is_accelerate_available() or is_accelerate_version("<", "0.14.0"): return False return hasattr(module, "_hf_hook") and not isinstance( module._hf_hook, (accelerate.hooks.CpuOffload, accelerate.hooks.AlignDevicesHook) ) def module_is_offloaded(module): if not is_accelerate_available() or is_accelerate_version("<", "0.17.0.dev0"): return False return hasattr(module, "_hf_hook") and isinstance(module._hf_hook, accelerate.hooks.CpuOffload) # .to("cuda") would raise an error if the pipeline is sequentially offloaded, so we raise our own to make it clearer pipeline_is_sequentially_offloaded = any( module_is_sequentially_offloaded(module) for _, module in self.components.items() ) if pipeline_is_sequentially_offloaded and device and torch.device(device).type == "cuda": raise ValueError( "It seems like you have activated sequential model offloading by calling `enable_sequential_cpu_offload`, but are now attempting to move the pipeline to GPU. This is not compatible with offloading. Please, move your pipeline `.to('cpu')` or consider removing the move altogether if you use sequential offloading." ) # Display a warning in this case (the operation succeeds but the benefits are lost) pipeline_is_offloaded = any(module_is_offloaded(module) for _, module in self.components.items()) if pipeline_is_offloaded and device and torch.device(device).type == "cuda": logger.warning( f"It seems like you have activated model offloading by calling `enable_model_cpu_offload`, but are now manually moving the pipeline to GPU. It is strongly recommended against doing so as memory gains from offloading are likely to be lost. Offloading automatically takes care of moving the individual components {', '.join(self.components.keys())} to GPU when needed. To make sure offloading works as expected, you should consider moving the pipeline back to CPU: `pipeline.to('cpu')` or removing the move altogether if you use offloading." ) module_names, _ = self._get_signature_keys(self) modules = [getattr(self, n, None) for n in module_names] modules = [m for m in modules if isinstance(m, torch.nn.Module)] is_offloaded = pipeline_is_offloaded or pipeline_is_sequentially_offloaded for module in modules: is_loaded_in_8bit = hasattr(module, "is_loaded_in_8bit") and module.is_loaded_in_8bit if is_loaded_in_8bit and dtype is not None: logger.warning( f"The module '{module.__class__.__name__}' has been loaded in 8bit and conversion to {torch_dtype} is not yet supported. Module is still in 8bit precision." ) if is_loaded_in_8bit and device is not None: logger.warning( f"The module '{module.__class__.__name__}' has been loaded in 8bit and moving it to {torch_dtype} via `.to()` is not yet supported. Module is still on {module.device}." ) else: module.to(device, dtype) if ( module.dtype == torch.float16 and str(device) in ["cpu"] and not silence_dtype_warnings and not is_offloaded ): logger.warning( "Pipelines loaded with `dtype=torch.float16` cannot run with `cpu` device. It" " is not recommended to move them to `cpu` as running them will fail. Please make" " sure to use an accelerator to run the pipeline in inference, due to the lack of" " support for`float16` operations on this device in PyTorch. Please, remove the" " `torch_dtype=torch.float16` argument, or use another device for inference." ) return self @property def device(self) -> torch.device: r""" Returns: `torch.device`: The torch device on which the pipeline is located. """ module_names, _ = self._get_signature_keys(self) modules = [getattr(self, n, None) for n in module_names] modules = [m for m in modules if isinstance(m, torch.nn.Module)] for module in modules: return module.device return torch.device("cpu") @property def dtype(self) -> torch.dtype: r""" Returns: `torch.dtype`: The torch dtype on which the pipeline is located. """ module_names, _ = self._get_signature_keys(self) modules = [getattr(self, n, None) for n in module_names] modules = [m for m in modules if isinstance(m, torch.nn.Module)] for module in modules: return module.dtype return torch.float32 @classmethod @validate_hf_hub_args def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.PathLike]], **kwargs): r""" Instantiate a PyTorch diffusion pipeline from pretrained pipeline weights. The pipeline is set in evaluation mode (`model.eval()`) by default. If you get the error message below, you need to finetune the weights for your downstream task: ``` Some weights of UNet2DConditionModel were not initialized from the model checkpoint at runwayml/stable-diffusion-v1-5 and are newly initialized because the shapes did not match: - conv_in.weight: found shape torch.Size([320, 4, 3, 3]) in the checkpoint and torch.Size([320, 9, 3, 3]) in the model instantiated You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference. ``` Parameters: pretrained_model_name_or_path (`str` or `os.PathLike`, *optional*): Can be either: - A string, the *repo id* (for example `CompVis/ldm-text2im-large-256`) of a pretrained pipeline hosted on the Hub. - A path to a *directory* (for example `./my_pipeline_directory/`) containing pipeline weights saved using [`~DiffusionPipeline.save_pretrained`]. torch_dtype (`str` or `torch.dtype`, *optional*): Override the default `torch.dtype` and load the model with another dtype. If "auto" is passed, the dtype is automatically derived from the model's weights. custom_pipeline (`str`, *optional*): <Tip warning={true}> 🧪 This is an experimental feature and may change in the future. </Tip> Can be either: - A string, the *repo id* (for example `hf-internal-testing/diffusers-dummy-pipeline`) of a custom pipeline hosted on the Hub. The repository must contain a file called pipeline.py that defines the custom pipeline. - A string, the *file name* of a community pipeline hosted on GitHub under [Community](https://github.com/huggingface/diffusers/tree/main/examples/community). Valid file names must match the file name and not the pipeline script (`clip_guided_stable_diffusion` instead of `clip_guided_stable_diffusion.py`). Community pipelines are always loaded from the current main branch of GitHub. - A path to a directory (`./my_pipeline_directory/`) containing a custom pipeline. The directory must contain a file called `pipeline.py` that defines the custom pipeline. For more information on how to load and create custom pipelines, please have a look at [Loading and Adding Custom Pipelines](https://huggingface.co/docs/diffusers/using-diffusers/custom_pipeline_overview) force_download (`bool`, *optional*, defaults to `False`): Whether or not to force the (re-)download of the model weights and configuration files, overriding the cached versions if they exist. cache_dir (`Union[str, os.PathLike]`, *optional*): Path to a directory where a downloaded pretrained model configuration is cached if the standard cache is not used. resume_download (`bool`, *optional*, defaults to `False`): Whether or not to resume downloading the model weights and configuration files. If set to `False`, any incompletely downloaded files are deleted. proxies (`Dict[str, str]`, *optional*): A dictionary of proxy servers to use by protocol or endpoint, for example, `{'http': 'foo.bar:3128', 'http://hostname': 'foo.bar:4012'}`. The proxies are used on each request. output_loading_info(`bool`, *optional*, defaults to `False`): Whether or not to also return a dictionary containing missing keys, unexpected keys and error messages. local_files_only (`bool`, *optional*, defaults to `False`): Whether to only load local model weights and configuration files or not. If set to `True`, the model won't be downloaded from the Hub. token (`str` or *bool*, *optional*): The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from `diffusers-cli login` (stored in `~/.huggingface`) is used. revision (`str`, *optional*, defaults to `"main"`): The specific model version to use. It can be a branch name, a tag name, a commit id, or any identifier allowed by Git. custom_revision (`str`, *optional*, defaults to `"main"`): The specific model version to use. It can be a branch name, a tag name, or a commit id similar to `revision` when loading a custom pipeline from the Hub. It can be a 🤗 Diffusers version when loading a custom pipeline from GitHub, otherwise it defaults to `"main"` when loading from the Hub. mirror (`str`, *optional*): Mirror source to resolve accessibility issues if you’re downloading a model in China. We do not guarantee the timeliness or safety of the source, and you should refer to the mirror site for more information. device_map (`str` or `Dict[str, Union[int, str, torch.device]]`, *optional*): A map that specifies where each submodule should go. It doesn’t need to be defined for each parameter/buffer name; once a given module name is inside, every submodule of it will be sent to the same device. Set `device_map="auto"` to have 🤗 Accelerate automatically compute the most optimized `device_map`. For more information about each option see [designing a device map](https://hf.co/docs/accelerate/main/en/usage_guides/big_modeling#designing-a-device-map). max_memory (`Dict`, *optional*): A dictionary device identifier for the maximum memory. Will default to the maximum memory available for each GPU and the available CPU RAM if unset. offload_folder (`str` or `os.PathLike`, *optional*): The path to offload weights if device_map contains the value `"disk"`. offload_state_dict (`bool`, *optional*): If `True`, temporarily offloads the CPU state dict to the hard drive to avoid running out of CPU RAM if the weight of the CPU state dict + the biggest shard of the checkpoint does not fit. Defaults to `True` when there is some disk offload. low_cpu_mem_usage (`bool`, *optional*, defaults to `True` if torch version >= 1.9.0 else `False`): Speed up model loading only loading the pretrained weights and not initializing the weights. This also tries to not use more than 1x model size in CPU memory (including peak memory) while loading the model. Only supported for PyTorch >= 1.9.0. If you are using an older version of PyTorch, setting this argument to `True` will raise an error. use_safetensors (`bool`, *optional*, defaults to `None`): If set to `None`, the safetensors weights are downloaded if they're available **and** if the safetensors library is installed. If set to `True`, the model is forcibly loaded from safetensors weights. If set to `False`, safetensors weights are not loaded. use_onnx (`bool`, *optional*, defaults to `None`): If set to `True`, ONNX weights will always be downloaded if present. If set to `False`, ONNX weights will never be downloaded. By default `use_onnx` defaults to the `_is_onnx` class attribute which is `False` for non-ONNX pipelines and `True` for ONNX pipelines. ONNX weights include both files ending with `.onnx` and `.pb`. kwargs (remaining dictionary of keyword arguments, *optional*): Can be used to overwrite load and saveable variables (the pipeline components of the specific pipeline class). The overwritten components are passed directly to the pipelines `__init__` method. See example below for more information. variant (`str`, *optional*): Load weights from a specified variant filename such as `"fp16"` or `"ema"`. This is ignored when loading `from_flax`. <Tip> To use private or [gated](https://huggingface.co/docs/hub/models-gated#gated-models) models, log-in with `huggingface-cli login`. </Tip> Examples: ```py >>> from diffusers import DiffusionPipeline >>> # Download pipeline from huggingface.co and cache. >>> pipeline = DiffusionPipeline.from_pretrained("CompVis/ldm-text2im-large-256") >>> # Download pipeline that requires an authorization token >>> # For more information on access tokens, please refer to this section >>> # of the documentation](https://huggingface.co/docs/hub/security-tokens) >>> pipeline = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5") >>> # Use a different scheduler >>> from diffusers import LMSDiscreteScheduler >>> scheduler = LMSDiscreteScheduler.from_config(pipeline.scheduler.config) >>> pipeline.scheduler = scheduler ``` """ cache_dir = kwargs.pop("cache_dir", None) resume_download = kwargs.pop("resume_download", False) force_download = kwargs.pop("force_download", False) proxies = kwargs.pop("proxies", None) local_files_only = kwargs.pop("local_files_only", None) token = kwargs.pop("token", None) revision = kwargs.pop("revision", None) from_flax = kwargs.pop("from_flax", False) torch_dtype = kwargs.pop("torch_dtype", None) custom_pipeline = kwargs.pop("custom_pipeline", None) custom_revision = kwargs.pop("custom_revision", None) provider = kwargs.pop("provider", None) sess_options = kwargs.pop("sess_options", None) device_map = kwargs.pop("device_map", None) max_memory = kwargs.pop("max_memory", None) offload_folder = kwargs.pop("offload_folder", None) offload_state_dict = kwargs.pop("offload_state_dict", False) low_cpu_mem_usage = kwargs.pop("low_cpu_mem_usage", _LOW_CPU_MEM_USAGE_DEFAULT) variant = kwargs.pop("variant", None) use_safetensors = kwargs.pop("use_safetensors", None) use_onnx = kwargs.pop("use_onnx", None) load_connected_pipeline = kwargs.pop("load_connected_pipeline", False) # 1. Download the checkpoints and configs # use snapshot download here to get it working from from_pretrained if not os.path.isdir(pretrained_model_name_or_path): if pretrained_model_name_or_path.count("/") > 1: raise ValueError( f'The provided pretrained_model_name_or_path "{pretrained_model_name_or_path}"' " is neither a valid local path nor a valid repo id. Please check the parameter." ) cached_folder = cls.download( pretrained_model_name_or_path, cache_dir=cache_dir, resume_download=resume_download, force_download=force_download, proxies=proxies, local_files_only=local_files_only, token=token, revision=revision, from_flax=from_flax, use_safetensors=use_safetensors, use_onnx=use_onnx, custom_pipeline=custom_pipeline, custom_revision=custom_revision, variant=variant, load_connected_pipeline=load_connected_pipeline, **kwargs, ) else: cached_folder = pretrained_model_name_or_path config_dict = cls.load_config(cached_folder) # pop out "_ignore_files" as it is only needed for download config_dict.pop("_ignore_files", None) # 2. Define which model components should load variants # We retrieve the information by matching whether variant # model checkpoints exist in the subfolders model_variants = {} if variant is not None: for folder in os.listdir(cached_folder): folder_path = os.path.join(cached_folder, folder) is_folder = os.path.isdir(folder_path) and folder in config_dict variant_exists = is_folder and any( p.split(".")[1].startswith(variant) for p in os.listdir(folder_path) ) if variant_exists: model_variants[folder] = variant # 3. Load the pipeline class, if using custom module then load it from the hub # if we load from explicit class, let's use it custom_class_name = None if os.path.isfile(os.path.join(cached_folder, f"{custom_pipeline}.py")): custom_pipeline = os.path.join(cached_folder, f"{custom_pipeline}.py") elif isinstance(config_dict["_class_name"], (list, tuple)) and os.path.isfile( os.path.join(cached_folder, f"{config_dict['_class_name'][0]}.py") ): custom_pipeline = os.path.join(cached_folder, f"{config_dict['_class_name'][0]}.py") custom_class_name = config_dict["_class_name"][1] pipeline_class = _get_pipeline_class( cls, config_dict, load_connected_pipeline=load_connected_pipeline, custom_pipeline=custom_pipeline, class_name=custom_class_name, cache_dir=cache_dir, revision=custom_revision, ) # DEPRECATED: To be removed in 1.0.0 if pipeline_class.__name__ == "StableDiffusionInpaintPipeline" and version.parse( version.parse(config_dict["_diffusers_version"]).base_version ) <= version.parse("0.5.1"): from diffusers import StableDiffusionInpaintPipeline, StableDiffusionInpaintPipelineLegacy pipeline_class = StableDiffusionInpaintPipelineLegacy deprecation_message = ( "You are using a legacy checkpoint for inpainting with Stable Diffusion, therefore we are loading the" f" {StableDiffusionInpaintPipelineLegacy} class instead of {StableDiffusionInpaintPipeline}. For" " better inpainting results, we strongly suggest using Stable Diffusion's official inpainting" " checkpoint: https://huggingface.co/runwayml/stable-diffusion-inpainting instead or adapting your" f" checkpoint {pretrained_model_name_or_path} to the format of" " https://huggingface.co/runwayml/stable-diffusion-inpainting. Note that we do not actively maintain" " the {StableDiffusionInpaintPipelineLegacy} class and will likely remove it in version 1.0.0." ) deprecate("StableDiffusionInpaintPipelineLegacy", "1.0.0", deprecation_message, standard_warn=False) # 4. Define expected modules given pipeline signature # and define non-None initialized modules (=`init_kwargs`) # some modules can be passed directly to the init # in this case they are already instantiated in `kwargs` # extract them here expected_modules, optional_kwargs = cls._get_signature_keys(pipeline_class) passed_class_obj = {k: kwargs.pop(k) for k in expected_modules if k in kwargs} passed_pipe_kwargs = {k: kwargs.pop(k) for k in optional_kwargs if k in kwargs} init_dict, unused_kwargs, _ = pipeline_class.extract_init_dict(config_dict, **kwargs) # define init kwargs and make sure that optional component modules are filtered out init_kwargs = { k: init_dict.pop(k) for k in optional_kwargs if k in init_dict and k not in pipeline_class._optional_components } init_kwargs = {**init_kwargs, **passed_pipe_kwargs} # remove `null` components def load_module(name, value): if value[0] is None: return False if name in passed_class_obj and passed_class_obj[name] is None: return False return True init_dict = {k: v for k, v in init_dict.items() if load_module(k, v)} # Special case: safety_checker must be loaded separately when using `from_flax` if from_flax and "safety_checker" in init_dict and "safety_checker" not in passed_class_obj: raise NotImplementedError( "The safety checker cannot be automatically loaded when loading weights `from_flax`." " Please, pass `safety_checker=None` to `from_pretrained`, and load the safety checker" " separately if you need it." ) # 5. Throw nice warnings / errors for fast accelerate loading if len(unused_kwargs) > 0: logger.warning( f"Keyword arguments {unused_kwargs} are not expected by {pipeline_class.__name__} and will be ignored." ) if low_cpu_mem_usage and not is_accelerate_available(): low_cpu_mem_usage = False logger.warning( "Cannot initialize model with low cpu memory usage because `accelerate` was not found in the" " environment. Defaulting to `low_cpu_mem_usage=False`. It is strongly recommended to install" " `accelerate` for faster and less memory-intense model loading. You can do so with: \n```\npip" " install accelerate\n```\n." ) if device_map is not None and not is_torch_version(">=", "1.9.0"): raise NotImplementedError( "Loading and dispatching requires torch >= 1.9.0. Please either update your PyTorch version or set" " `device_map=None`." ) if low_cpu_mem_usage is True and not is_torch_version(">=", "1.9.0"): raise NotImplementedError( "Low memory initialization requires torch >= 1.9.0. Please either update your PyTorch version or set" " `low_cpu_mem_usage=False`." ) if low_cpu_mem_usage is False and device_map is not None: raise ValueError( f"You cannot set `low_cpu_mem_usage` to False while using device_map={device_map} for loading and" " dispatching. Please make sure to set `low_cpu_mem_usage=True`." ) # import it here to avoid circular import from diffusers import pipelines # 6. Load each module in the pipeline for name, (library_name, class_name) in logging.tqdm(init_dict.items(), desc="Loading pipeline components..."): # 6.1 - now that JAX/Flax is an official framework of the library, we might load from Flax names class_name = class_name[4:] if class_name.startswith("Flax") else class_name # 6.2 Define all importable classes is_pipeline_module = hasattr(pipelines, library_name) importable_classes = ALL_IMPORTABLE_CLASSES loaded_sub_model = None # 6.3 Use passed sub model or load class_name from library_name if name in passed_class_obj: # if the model is in a pipeline module, then we load it from the pipeline # check that passed_class_obj has correct parent class maybe_raise_or_warn( library_name, library, class_name, importable_classes, passed_class_obj, name, is_pipeline_module ) loaded_sub_model = passed_class_obj[name] else: # load sub model loaded_sub_model = load_sub_model( library_name=library_name, class_name=class_name, importable_classes=importable_classes, pipelines=pipelines, is_pipeline_module=is_pipeline_module, pipeline_class=pipeline_class, torch_dtype=torch_dtype, provider=provider, sess_options=sess_options, device_map=device_map, max_memory=max_memory, offload_folder=offload_folder, offload_state_dict=offload_state_dict, model_variants=model_variants, name=name, from_flax=from_flax, variant=variant, low_cpu_mem_usage=low_cpu_mem_usage, cached_folder=cached_folder, revision=revision, ) logger.info( f"Loaded {name} as {class_name} from `{name}` subfolder of {pretrained_model_name_or_path}." ) init_kwargs[name] = loaded_sub_model # UNet(...), # DiffusionSchedule(...) if pipeline_class._load_connected_pipes and os.path.isfile(os.path.join(cached_folder, "README.md")): modelcard = ModelCard.load(os.path.join(cached_folder, "README.md")) connected_pipes = {prefix: getattr(modelcard.data, prefix, [None])[0] for prefix in CONNECTED_PIPES_KEYS} load_kwargs = { "cache_dir": cache_dir, "resume_download": resume_download, "force_download": force_download, "proxies": proxies, "local_files_only": local_files_only, "token": token, "revision": revision, "torch_dtype": torch_dtype, "custom_pipeline": custom_pipeline, "custom_revision": custom_revision, "provider": provider, "sess_options": sess_options, "device_map": device_map, "max_memory": max_memory, "offload_folder": offload_folder, "offload_state_dict": offload_state_dict, "low_cpu_mem_usage": low_cpu_mem_usage, "variant": variant, "use_safetensors": use_safetensors, } def get_connected_passed_kwargs(prefix): connected_passed_class_obj = { k.replace(f"{prefix}_", ""): w for k, w in passed_class_obj.items() if k.split("_")[0] == prefix } connected_passed_pipe_kwargs = { k.replace(f"{prefix}_", ""): w for k, w in passed_pipe_kwargs.items() if k.split("_")[0] == prefix } connected_passed_kwargs = {**connected_passed_class_obj, **connected_passed_pipe_kwargs} return connected_passed_kwargs connected_pipes = { prefix: DiffusionPipeline.from_pretrained( repo_id, **load_kwargs.copy(), **get_connected_passed_kwargs(prefix) ) for prefix, repo_id in connected_pipes.items() if repo_id is not None } for prefix, connected_pipe in connected_pipes.items(): # add connected pipes to `init_kwargs` with <prefix>_<component_name>, e.g. "prior_text_encoder" init_kwargs.update( {"_".join([prefix, name]): component for name, component in connected_pipe.components.items()} ) # 7. Potentially add passed objects if expected missing_modules = set(expected_modules) - set(init_kwargs.keys()) passed_modules = list(passed_class_obj.keys()) optional_modules = pipeline_class._optional_components if len(missing_modules) > 0 and missing_modules <= set(passed_modules + optional_modules): for module in missing_modules: init_kwargs[module] = passed_class_obj.get(module, None) elif len(missing_modules) > 0: passed_modules = set(list(init_kwargs.keys()) + list(passed_class_obj.keys())) - optional_kwargs raise ValueError( f"Pipeline {pipeline_class} expected {expected_modules}, but only {passed_modules} were passed." ) # 8. Instantiate the pipeline model = pipeline_class(**init_kwargs) # 9. Save where the model was instantiated from model.register_to_config(_name_or_path=pretrained_model_name_or_path) return model @property def name_or_path(self) -> str: return getattr(self.config, "_name_or_path", None) @property def _execution_device(self): r""" Returns the device on which the pipeline's models will be executed. After calling [`~DiffusionPipeline.enable_sequential_cpu_offload`] the execution device can only be inferred from Accelerate's module hooks. """ for name, model in self.components.items(): if not isinstance(model, torch.nn.Module) or name in self._exclude_from_cpu_offload: continue if not hasattr(model, "_hf_hook"): return self.device for module in model.modules(): if ( hasattr(module, "_hf_hook") and hasattr(module._hf_hook, "execution_device") and module._hf_hook.execution_device is not None ): return torch.device(module._hf_hook.execution_device) return self.device def enable_model_cpu_offload(self, gpu_id: Optional[int] = None, device: Union[torch.device, str] = "cuda"): r""" Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared to `enable_sequential_cpu_offload`, this method moves one whole model at a time to the GPU when its `forward` method is called, and the model remains in GPU until the next model runs. Memory savings are lower than with `enable_sequential_cpu_offload`, but performance is much better due to the iterative execution of the `unet`. Arguments: gpu_id (`int`, *optional*): The ID of the accelerator that shall be used in inference. If not specified, it will default to 0. device (`torch.Device` or `str`, *optional*, defaults to "cuda"): The PyTorch device type of the accelerator that shall be used in inference. If not specified, it will default to "cuda". """ if self.model_cpu_offload_seq is None: raise ValueError( "Model CPU offload cannot be enabled because no `model_cpu_offload_seq` class attribute is set." ) if is_accelerate_available() and is_accelerate_version(">=", "0.17.0.dev0"): from accelerate import cpu_offload_with_hook else: raise ImportError("`enable_model_cpu_offload` requires `accelerate v0.17.0` or higher.") torch_device = torch.device(device) device_index = torch_device.index if gpu_id is not None and device_index is not None: raise ValueError( f"You have passed both `gpu_id`={gpu_id} and an index as part of the passed device `device`={device}" f"Cannot pass both. Please make sure to either not define `gpu_id` or not pass the index as part of the device: `device`={torch_device.type}" ) # _offload_gpu_id should be set to passed gpu_id (or id in passed `device`) or default to previously set id or default to 0 self._offload_gpu_id = gpu_id or torch_device.index or getattr(self, "_offload_gpu_id", 0) device_type = torch_device.type device = torch.device(f"{device_type}:{self._offload_gpu_id}") if self.device.type != "cpu": self.to("cpu", silence_dtype_warnings=True) device_mod = getattr(torch, self.device.type, None) if hasattr(device_mod, "empty_cache") and device_mod.is_available(): device_mod.empty_cache() # otherwise we don't see the memory savings (but they probably exist) all_model_components = {k: v for k, v in self.components.items() if isinstance(v, torch.nn.Module)} self._all_hooks = [] hook = None for model_str in self.model_cpu_offload_seq.split("->"): model = all_model_components.pop(model_str, None) if not isinstance(model, torch.nn.Module): continue _, hook = cpu_offload_with_hook(model, device, prev_module_hook=hook) self._all_hooks.append(hook) # CPU offload models that are not in the seq chain unless they are explicitly excluded # these models will stay on CPU until maybe_free_model_hooks is called # some models cannot be in the seq chain because they are iteratively called, such as controlnet for name, model in all_model_components.items(): if not isinstance(model, torch.nn.Module): continue if name in self._exclude_from_cpu_offload: model.to(device) else: _, hook = cpu_offload_with_hook(model, device) self._all_hooks.append(hook) def maybe_free_model_hooks(self): r""" Function that offloads all components, removes all model hooks that were added when using `enable_model_cpu_offload` and then applies them again. In case the model has not been offloaded this function is a no-op. Make sure to add this function to the end of the `__call__` function of your pipeline so that it functions correctly when applying enable_model_cpu_offload. """ if not hasattr(self, "_all_hooks") or len(self._all_hooks) == 0: # `enable_model_cpu_offload` has not be called, so silently do nothing return for hook in self._all_hooks: # offload model and remove hook from model hook.offload() hook.remove() # make sure the model is in the same state as before calling it self.enable_model_cpu_offload() def enable_sequential_cpu_offload(self, gpu_id: Optional[int] = None, device: Union[torch.device, str] = "cuda"): r""" Offloads all models to CPU using 🤗 Accelerate, significantly reducing memory usage. When called, the state dicts of all `torch.nn.Module` components (except those in `self._exclude_from_cpu_offload`) are saved to CPU and then moved to `torch.device('meta')` and loaded to GPU only when their specific submodule has its `forward` method called. Offloading happens on a submodule basis. Memory savings are higher than with `enable_model_cpu_offload`, but performance is lower. Arguments: gpu_id (`int`, *optional*): The ID of the accelerator that shall be used in inference. If not specified, it will default to 0. device (`torch.Device` or `str`, *optional*, defaults to "cuda"): The PyTorch device type of the accelerator that shall be used in inference. If not specified, it will default to "cuda". """ if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"): from accelerate import cpu_offload else: raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher") torch_device = torch.device(device) device_index = torch_device.index if gpu_id is not None and device_index is not None: raise ValueError( f"You have passed both `gpu_id`={gpu_id} and an index as part of the passed device `device`={device}" f"Cannot pass both. Please make sure to either not define `gpu_id` or not pass the index as part of the device: `device`={torch_device.type}" ) # _offload_gpu_id should be set to passed gpu_id (or id in passed `device`) or default to previously set id or default to 0 self._offload_gpu_id = gpu_id or torch_device.index or getattr(self, "_offload_gpu_id", 0) device_type = torch_device.type device = torch.device(f"{device_type}:{self._offload_gpu_id}") if self.device.type != "cpu": self.to("cpu", silence_dtype_warnings=True) device_mod = getattr(torch, self.device.type, None) if hasattr(device_mod, "empty_cache") and device_mod.is_available(): device_mod.empty_cache() # otherwise we don't see the memory savings (but they probably exist) for name, model in self.components.items(): if not isinstance(model, torch.nn.Module): continue if name in self._exclude_from_cpu_offload: model.to(device) else: # make sure to offload buffers if not all high level weights # are of type nn.Module offload_buffers = len(model._parameters) > 0 cpu_offload(model, device, offload_buffers=offload_buffers) @classmethod @validate_hf_hub_args def download(cls, pretrained_model_name, **kwargs) -> Union[str, os.PathLike]: r""" Download and cache a PyTorch diffusion pipeline from pretrained pipeline weights. Parameters: pretrained_model_name (`str` or `os.PathLike`, *optional*): A string, the *repository id* (for example `CompVis/ldm-text2im-large-256`) of a pretrained pipeline hosted on the Hub. custom_pipeline (`str`, *optional*): Can be either: - A string, the *repository id* (for example `CompVis/ldm-text2im-large-256`) of a pretrained pipeline hosted on the Hub. The repository must contain a file called `pipeline.py` that defines the custom pipeline. - A string, the *file name* of a community pipeline hosted on GitHub under [Community](https://github.com/huggingface/diffusers/tree/main/examples/community). Valid file names must match the file name and not the pipeline script (`clip_guided_stable_diffusion` instead of `clip_guided_stable_diffusion.py`). Community pipelines are always loaded from the current `main` branch of GitHub. - A path to a *directory* (`./my_pipeline_directory/`) containing a custom pipeline. The directory must contain a file called `pipeline.py` that defines the custom pipeline. <Tip warning={true}> 🧪 This is an experimental feature and may change in the future. </Tip> For more information on how to load and create custom pipelines, take a look at [How to contribute a community pipeline](https://huggingface.co/docs/diffusers/main/en/using-diffusers/contribute_pipeline). force_download (`bool`, *optional*, defaults to `False`): Whether or not to force the (re-)download of the model weights and configuration files, overriding the cached versions if they exist. resume_download (`bool`, *optional*, defaults to `False`): Whether or not to resume downloading the model weights and configuration files. If set to `False`, any incompletely downloaded files are deleted. proxies (`Dict[str, str]`, *optional*): A dictionary of proxy servers to use by protocol or endpoint, for example, `{'http': 'foo.bar:3128', 'http://hostname': 'foo.bar:4012'}`. The proxies are used on each request. output_loading_info(`bool`, *optional*, defaults to `False`): Whether or not to also return a dictionary containing missing keys, unexpected keys and error messages. local_files_only (`bool`, *optional*, defaults to `False`): Whether to only load local model weights and configuration files or not. If set to `True`, the model won't be downloaded from the Hub. token (`str` or *bool*, *optional*): The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from `diffusers-cli login` (stored in `~/.huggingface`) is used. revision (`str`, *optional*, defaults to `"main"`): The specific model version to use. It can be a branch name, a tag name, a commit id, or any identifier allowed by Git. custom_revision (`str`, *optional*, defaults to `"main"`): The specific model version to use. It can be a branch name, a tag name, or a commit id similar to `revision` when loading a custom pipeline from the Hub. It can be a 🤗 Diffusers version when loading a custom pipeline from GitHub, otherwise it defaults to `"main"` when loading from the Hub. mirror (`str`, *optional*): Mirror source to resolve accessibility issues if you're downloading a model in China. We do not guarantee the timeliness or safety of the source, and you should refer to the mirror site for more information. variant (`str`, *optional*): Load weights from a specified variant filename such as `"fp16"` or `"ema"`. This is ignored when loading `from_flax`. use_safetensors (`bool`, *optional*, defaults to `None`): If set to `None`, the safetensors weights are downloaded if they're available **and** if the safetensors library is installed. If set to `True`, the model is forcibly loaded from safetensors weights. If set to `False`, safetensors weights are not loaded. use_onnx (`bool`, *optional*, defaults to `False`): If set to `True`, ONNX weights will always be downloaded if present. If set to `False`, ONNX weights will never be downloaded. By default `use_onnx` defaults to the `_is_onnx` class attribute which is `False` for non-ONNX pipelines and `True` for ONNX pipelines. ONNX weights include both files ending with `.onnx` and `.pb`. trust_remote_code (`bool`, *optional*, defaults to `False`): Whether or not to allow for custom pipelines and components defined on the Hub in their own files. This option should only be set to `True` for repositories you trust and in which you have read the code, as it will execute code present on the Hub on your local machine. Returns: `os.PathLike`: A path to the downloaded pipeline. <Tip> To use private or [gated models](https://huggingface.co/docs/hub/models-gated#gated-models), log-in with `huggingface-cli login`. </Tip> """ cache_dir = kwargs.pop("cache_dir", None) resume_download = kwargs.pop("resume_download", False) force_download = kwargs.pop("force_download", False) proxies = kwargs.pop("proxies", None) local_files_only = kwargs.pop("local_files_only", None) token = kwargs.pop("token", None) revision = kwargs.pop("revision", None) from_flax = kwargs.pop("from_flax", False) custom_pipeline = kwargs.pop("custom_pipeline", None) custom_revision = kwargs.pop("custom_revision", None) variant = kwargs.pop("variant", None) use_safetensors = kwargs.pop("use_safetensors", None) use_onnx = kwargs.pop("use_onnx", None) load_connected_pipeline = kwargs.pop("load_connected_pipeline", False) trust_remote_code = kwargs.pop("trust_remote_code", False) allow_pickle = False if use_safetensors is None: use_safetensors = True allow_pickle = True allow_patterns = None ignore_patterns = None model_info_call_error: Optional[Exception] = None if not local_files_only: try: info = model_info(pretrained_model_name, token=token, revision=revision) except (HTTPError, OfflineModeIsEnabled, requests.ConnectionError) as e: logger.warn(f"Couldn't connect to the Hub: {e}.\nWill try to load from local cache.") local_files_only = True model_info_call_error = e # save error to reraise it if model is not cached locally if not local_files_only: config_file = hf_hub_download( pretrained_model_name, cls.config_name, cache_dir=cache_dir, revision=revision, proxies=proxies, force_download=force_download, resume_download=resume_download, token=token, ) config_dict = cls._dict_from_json_file(config_file) ignore_filenames = config_dict.pop("_ignore_files", []) # retrieve all folder_names that contain relevant files folder_names = [k for k, v in config_dict.items() if isinstance(v, list) and k != "_class_name"] filenames = {sibling.rfilename for sibling in info.siblings} model_filenames, variant_filenames = variant_compatible_siblings(filenames, variant=variant) diffusers_module = importlib.import_module(__name__.split(".")[0]) pipelines = getattr(diffusers_module, "pipelines") # optionally create a custom component <> custom file mapping custom_components = {} for component in folder_names: module_candidate = config_dict[component][0] if module_candidate is None or not isinstance(module_candidate, str): continue # We compute candidate file path on the Hub. Do not use `os.path.join`. candidate_file = f"{component}/{module_candidate}.py" if candidate_file in filenames: custom_components[component] = module_candidate elif module_candidate not in LOADABLE_CLASSES and not hasattr(pipelines, module_candidate): raise ValueError( f"{candidate_file} as defined in `model_index.json` does not exist in {pretrained_model_name} and is not a module in 'diffusers/pipelines'." ) if len(variant_filenames) == 0 and variant is not None: deprecation_message = ( f"You are trying to load the model files of the `variant={variant}`, but no such modeling files are available." f"The default model files: {model_filenames} will be loaded instead. Make sure to not load from `variant={variant}`" "if such variant modeling files are not available. Doing so will lead to an error in v0.24.0 as defaulting to non-variant" "modeling files is deprecated." ) deprecate("no variant default", "0.24.0", deprecation_message, standard_warn=False) # remove ignored filenames model_filenames = set(model_filenames) - set(ignore_filenames) variant_filenames = set(variant_filenames) - set(ignore_filenames) # if the whole pipeline is cached we don't have to ping the Hub if revision in DEPRECATED_REVISION_ARGS and version.parse( version.parse(__version__).base_version ) >= version.parse("0.22.0"): warn_deprecated_model_variant(pretrained_model_name, token, variant, revision, model_filenames) model_folder_names = {os.path.split(f)[0] for f in model_filenames if os.path.split(f)[0] in folder_names} custom_class_name = None if custom_pipeline is None and isinstance(config_dict["_class_name"], (list, tuple)): custom_pipeline = config_dict["_class_name"][0] custom_class_name = config_dict["_class_name"][1] # all filenames compatible with variant will be added allow_patterns = list(model_filenames) # allow all patterns from non-model folders # this enables downloading schedulers, tokenizers, ... allow_patterns += [f"{k}/*" for k in folder_names if k not in model_folder_names] # add custom component files allow_patterns += [f"{k}/{f}.py" for k, f in custom_components.items()] # add custom pipeline file allow_patterns += [f"{custom_pipeline}.py"] if f"{custom_pipeline}.py" in filenames else [] # also allow downloading config.json files with the model allow_patterns += [os.path.join(k, "config.json") for k in model_folder_names] allow_patterns += [ SCHEDULER_CONFIG_NAME, CONFIG_NAME, cls.config_name, CUSTOM_PIPELINE_FILE_NAME, ] load_pipe_from_hub = custom_pipeline is not None and f"{custom_pipeline}.py" in filenames load_components_from_hub = len(custom_components) > 0 if load_pipe_from_hub and not trust_remote_code: raise ValueError( f"The repository for {pretrained_model_name} contains custom code in {custom_pipeline}.py which must be executed to correctly " f"load the model. You can inspect the repository content at https://hf.co/{pretrained_model_name}/blob/main/{custom_pipeline}.py.\n" f"Please pass the argument `trust_remote_code=True` to allow custom code to be run." ) if load_components_from_hub and not trust_remote_code: raise ValueError( f"The repository for {pretrained_model_name} contains custom code in {'.py, '.join([os.path.join(k, v) for k,v in custom_components.items()])} which must be executed to correctly " f"load the model. You can inspect the repository content at {', '.join([f'https://hf.co/{pretrained_model_name}/{k}/{v}.py' for k,v in custom_components.items()])}.\n" f"Please pass the argument `trust_remote_code=True` to allow custom code to be run." ) # retrieve passed components that should not be downloaded pipeline_class = _get_pipeline_class( cls, config_dict, load_connected_pipeline=load_connected_pipeline, custom_pipeline=custom_pipeline, repo_id=pretrained_model_name if load_pipe_from_hub else None, hub_revision=revision, class_name=custom_class_name, cache_dir=cache_dir, revision=custom_revision, ) expected_components, _ = cls._get_signature_keys(pipeline_class) passed_components = [k for k in expected_components if k in kwargs] if ( use_safetensors and not allow_pickle and not is_safetensors_compatible( model_filenames, variant=variant, passed_components=passed_components ) ): raise EnvironmentError( f"Could not find the necessary `safetensors` weights in {model_filenames} (variant={variant})" ) if from_flax: ignore_patterns = ["*.bin", "*.safetensors", "*.onnx", "*.pb"] elif use_safetensors and is_safetensors_compatible( model_filenames, variant=variant, passed_components=passed_components ): ignore_patterns = ["*.bin", "*.msgpack"] use_onnx = use_onnx if use_onnx is not None else pipeline_class._is_onnx if not use_onnx: ignore_patterns += ["*.onnx", "*.pb"] safetensors_variant_filenames = {f for f in variant_filenames if f.endswith(".safetensors")} safetensors_model_filenames = {f for f in model_filenames if f.endswith(".safetensors")} if ( len(safetensors_variant_filenames) > 0 and safetensors_model_filenames != safetensors_variant_filenames ): logger.warn( f"\nA mixture of {variant} and non-{variant} filenames will be loaded.\nLoaded {variant} filenames:\n[{', '.join(safetensors_variant_filenames)}]\nLoaded non-{variant} filenames:\n[{', '.join(safetensors_model_filenames - safetensors_variant_filenames)}\nIf this behavior is not expected, please check your folder structure." ) else: ignore_patterns = ["*.safetensors", "*.msgpack"] use_onnx = use_onnx if use_onnx is not None else pipeline_class._is_onnx if not use_onnx: ignore_patterns += ["*.onnx", "*.pb"] bin_variant_filenames = {f for f in variant_filenames if f.endswith(".bin")} bin_model_filenames = {f for f in model_filenames if f.endswith(".bin")} if len(bin_variant_filenames) > 0 and bin_model_filenames != bin_variant_filenames: logger.warn( f"\nA mixture of {variant} and non-{variant} filenames will be loaded.\nLoaded {variant} filenames:\n[{', '.join(bin_variant_filenames)}]\nLoaded non-{variant} filenames:\n[{', '.join(bin_model_filenames - bin_variant_filenames)}\nIf this behavior is not expected, please check your folder structure." ) # Don't download any objects that are passed allow_patterns = [ p for p in allow_patterns if not (len(p.split("/")) == 2 and p.split("/")[0] in passed_components) ] if pipeline_class._load_connected_pipes: allow_patterns.append("README.md") # Don't download index files of forbidden patterns either ignore_patterns = ignore_patterns + [f"{i}.index.*json" for i in ignore_patterns] re_ignore_pattern = [re.compile(fnmatch.translate(p)) for p in ignore_patterns] re_allow_pattern = [re.compile(fnmatch.translate(p)) for p in allow_patterns] expected_files = [f for f in filenames if not any(p.match(f) for p in re_ignore_pattern)] expected_files = [f for f in expected_files if any(p.match(f) for p in re_allow_pattern)] snapshot_folder = Path(config_file).parent pipeline_is_cached = all((snapshot_folder / f).is_file() for f in expected_files) if pipeline_is_cached and not force_download: # if the pipeline is cached, we can directly return it # else call snapshot_download return snapshot_folder user_agent = {"pipeline_class": cls.__name__} if custom_pipeline is not None and not custom_pipeline.endswith(".py"): user_agent["custom_pipeline"] = custom_pipeline # download all allow_patterns - ignore_patterns try: cached_folder = snapshot_download( pretrained_model_name, cache_dir=cache_dir, resume_download=resume_download, proxies=proxies, local_files_only=local_files_only, token=token, revision=revision, allow_patterns=allow_patterns, ignore_patterns=ignore_patterns, user_agent=user_agent, ) # retrieve pipeline class from local file cls_name = cls.load_config(os.path.join(cached_folder, "model_index.json")).get("_class_name", None) cls_name = cls_name[4:] if isinstance(cls_name, str) and cls_name.startswith("Flax") else cls_name diffusers_module = importlib.import_module(__name__.split(".")[0]) pipeline_class = getattr(diffusers_module, cls_name, None) if isinstance(cls_name, str) else None if pipeline_class is not None and pipeline_class._load_connected_pipes: modelcard = ModelCard.load(os.path.join(cached_folder, "README.md")) connected_pipes = sum([getattr(modelcard.data, k, []) for k in CONNECTED_PIPES_KEYS], []) for connected_pipe_repo_id in connected_pipes: download_kwargs = { "cache_dir": cache_dir, "resume_download": resume_download, "force_download": force_download, "proxies": proxies, "local_files_only": local_files_only, "token": token, "variant": variant, "use_safetensors": use_safetensors, } DiffusionPipeline.download(connected_pipe_repo_id, **download_kwargs) return cached_folder except FileNotFoundError: # Means we tried to load pipeline with `local_files_only=True` but the files have not been found in local cache. # This can happen in two cases: # 1. If the user passed `local_files_only=True` => we raise the error directly # 2. If we forced `local_files_only=True` when `model_info` failed => we raise the initial error if model_info_call_error is None: # 1. user passed `local_files_only=True` raise else: # 2. we forced `local_files_only=True` when `model_info` failed raise EnvironmentError( f"Cannot load model {pretrained_model_name}: model is not cached locally and an error occured" " while trying to fetch metadata from the Hub. Please check out the root cause in the stacktrace" " above." ) from model_info_call_error @classmethod def _get_signature_keys(cls, obj): parameters = inspect.signature(obj.__init__).parameters required_parameters = {k: v for k, v in parameters.items() if v.default == inspect._empty} optional_parameters = set({k for k, v in parameters.items() if v.default != inspect._empty}) expected_modules = set(required_parameters.keys()) - {"self"} optional_names = list(optional_parameters) for name in optional_names: if name in cls._optional_components: expected_modules.add(name) optional_parameters.remove(name) return expected_modules, optional_parameters @property def components(self) -> Dict[str, Any]: r""" The `self.components` property can be useful to run different pipelines with the same weights and configurations without reallocating additional memory. Returns (`dict`): A dictionary containing all the modules needed to initialize the pipeline. Examples: ```py >>> from diffusers import ( ... StableDiffusionPipeline, ... StableDiffusionImg2ImgPipeline, ... StableDiffusionInpaintPipeline, ... ) >>> text2img = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5") >>> img2img = StableDiffusionImg2ImgPipeline(**text2img.components) >>> inpaint = StableDiffusionInpaintPipeline(**text2img.components) ``` """ expected_modules, optional_parameters = self._get_signature_keys(self) components = { k: getattr(self, k) for k in self.config.keys() if not k.startswith("_") and k not in optional_parameters } if set(components.keys()) != expected_modules: raise ValueError( f"{self} has been incorrectly initialized or {self.__class__} is incorrectly implemented. Expected" f" {expected_modules} to be defined, but {components.keys()} are defined." ) return components @staticmethod def numpy_to_pil(images): """ Convert a NumPy image or a batch of images to a PIL image. """ return numpy_to_pil(images) def progress_bar(self, iterable=None, total=None): if not hasattr(self, "_progress_bar_config"): self._progress_bar_config = {} elif not isinstance(self._progress_bar_config, dict): raise ValueError( f"`self._progress_bar_config` should be of type `dict`, but is {type(self._progress_bar_config)}." ) if iterable is not None: return tqdm(iterable, **self._progress_bar_config) elif total is not None: return tqdm(total=total, **self._progress_bar_config) else: raise ValueError("Either `total` or `iterable` has to be defined.") def set_progress_bar_config(self, **kwargs): self._progress_bar_config = kwargs def enable_xformers_memory_efficient_attention(self, attention_op: Optional[Callable] = None): r""" Enable memory efficient attention from [xFormers](https://facebookresearch.github.io/xformers/). When this option is enabled, you should observe lower GPU memory usage and a potential speed up during inference. Speed up during training is not guaranteed. <Tip warning={true}> ⚠️ When memory efficient attention and sliced attention are both enabled, memory efficient attention takes precedent. </Tip> Parameters: attention_op (`Callable`, *optional*): Override the default `None` operator for use as `op` argument to the [`memory_efficient_attention()`](https://facebookresearch.github.io/xformers/components/ops.html#xformers.ops.memory_efficient_attention) function of xFormers. Examples: ```py >>> import torch >>> from diffusers import DiffusionPipeline >>> from xformers.ops import MemoryEfficientAttentionFlashAttentionOp >>> pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-2-1", torch_dtype=torch.float16) >>> pipe = pipe.to("cuda") >>> pipe.enable_xformers_memory_efficient_attention(attention_op=MemoryEfficientAttentionFlashAttentionOp) >>> # Workaround for not accepting attention shape using VAE for Flash Attention >>> pipe.vae.enable_xformers_memory_efficient_attention(attention_op=None) ``` """ self.set_use_memory_efficient_attention_xformers(True, attention_op) def disable_xformers_memory_efficient_attention(self): r""" Disable memory efficient attention from [xFormers](https://facebookresearch.github.io/xformers/). """ self.set_use_memory_efficient_attention_xformers(False) def set_use_memory_efficient_attention_xformers( self, valid: bool, attention_op: Optional[Callable] = None ) -> None: # Recursively walk through all the children. # Any children which exposes the set_use_memory_efficient_attention_xformers method # gets the message def fn_recursive_set_mem_eff(module: torch.nn.Module): if hasattr(module, "set_use_memory_efficient_attention_xformers"): module.set_use_memory_efficient_attention_xformers(valid, attention_op) for child in module.children(): fn_recursive_set_mem_eff(child) module_names, _ = self._get_signature_keys(self) modules = [getattr(self, n, None) for n in module_names] modules = [m for m in modules if isinstance(m, torch.nn.Module)] for module in modules: fn_recursive_set_mem_eff(module) def enable_attention_slicing(self, slice_size: Optional[Union[str, int]] = "auto"): r""" Enable sliced attention computation. When this option is enabled, the attention module splits the input tensor in slices to compute attention in several steps. For more than one attention head, the computation is performed sequentially over each head. This is useful to save some memory in exchange for a small speed decrease. <Tip warning={true}> ⚠️ Don't enable attention slicing if you're already using `scaled_dot_product_attention` (SDPA) from PyTorch 2.0 or xFormers. These attention computations are already very memory efficient so you won't need to enable this function. If you enable attention slicing with SDPA or xFormers, it can lead to serious slow downs! </Tip> Args: slice_size (`str` or `int`, *optional*, defaults to `"auto"`): When `"auto"`, halves the input to the attention heads, so attention will be computed in two steps. If `"max"`, maximum amount of memory will be saved by running only one slice at a time. If a number is provided, uses as many slices as `attention_head_dim // slice_size`. In this case, `attention_head_dim` must be a multiple of `slice_size`. Examples: ```py >>> import torch >>> from diffusers import StableDiffusionPipeline >>> pipe = StableDiffusionPipeline.from_pretrained( ... "runwayml/stable-diffusion-v1-5", ... torch_dtype=torch.float16, ... use_safetensors=True, ... ) >>> prompt = "a photo of an astronaut riding a horse on mars" >>> pipe.enable_attention_slicing() >>> image = pipe(prompt).images[0] ``` """ self.set_attention_slice(slice_size) def disable_attention_slicing(self): r""" Disable sliced attention computation. If `enable_attention_slicing` was previously called, attention is computed in one step. """ # set slice_size = `None` to disable `attention slicing` self.enable_attention_slicing(None) def set_attention_slice(self, slice_size: Optional[int]): module_names, _ = self._get_signature_keys(self) modules = [getattr(self, n, None) for n in module_names] modules = [m for m in modules if isinstance(m, torch.nn.Module) and hasattr(m, "set_attention_slice")] for module in modules: module.set_attention_slice(slice_size)

diffusers/src/diffusers/pipelines/pipeline_utils.py/0

# Copyright 2023 The HuggingFace Team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import warnings from functools import partial from typing import Dict, List, Optional, Union import jax import jax.numpy as jnp import numpy as np from flax.core.frozen_dict import FrozenDict from flax.jax_utils import unreplicate from flax.training.common_utils import shard from PIL import Image from transformers import CLIPImageProcessor, CLIPTokenizer, FlaxCLIPTextModel from ...models import FlaxAutoencoderKL, FlaxUNet2DConditionModel from ...schedulers import ( FlaxDDIMScheduler, FlaxDPMSolverMultistepScheduler, FlaxLMSDiscreteScheduler, FlaxPNDMScheduler, ) from ...utils import PIL_INTERPOLATION, logging, replace_example_docstring from ..pipeline_flax_utils import FlaxDiffusionPipeline from .pipeline_output import FlaxStableDiffusionPipelineOutput from .safety_checker_flax import FlaxStableDiffusionSafetyChecker logger = logging.get_logger(__name__) # pylint: disable=invalid-name # Set to True to use python for loop instead of jax.fori_loop for easier debugging DEBUG = False EXAMPLE_DOC_STRING = """ Examples: ```py >>> import jax >>> import numpy as np >>> import jax.numpy as jnp >>> from flax.jax_utils import replicate >>> from flax.training.common_utils import shard >>> import requests >>> from io import BytesIO >>> from PIL import Image >>> from diffusers import FlaxStableDiffusionImg2ImgPipeline >>> def create_key(seed=0): ... return jax.random.PRNGKey(seed) >>> rng = create_key(0) >>> url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg" >>> response = requests.get(url) >>> init_img = Image.open(BytesIO(response.content)).convert("RGB") >>> init_img = init_img.resize((768, 512)) >>> prompts = "A fantasy landscape, trending on artstation" >>> pipeline, params = FlaxStableDiffusionImg2ImgPipeline.from_pretrained( ... "CompVis/stable-diffusion-v1-4", ... revision="flax", ... dtype=jnp.bfloat16, ... ) >>> num_samples = jax.device_count() >>> rng = jax.random.split(rng, jax.device_count()) >>> prompt_ids, processed_image = pipeline.prepare_inputs( ... prompt=[prompts] * num_samples, image=[init_img] * num_samples ... ) >>> p_params = replicate(params) >>> prompt_ids = shard(prompt_ids) >>> processed_image = shard(processed_image) >>> output = pipeline( ... prompt_ids=prompt_ids, ... image=processed_image, ... params=p_params, ... prng_seed=rng, ... strength=0.75, ... num_inference_steps=50, ... jit=True, ... height=512, ... width=768, ... ).images >>> output_images = pipeline.numpy_to_pil(np.asarray(output.reshape((num_samples,) + output.shape[-3:]))) ``` """ class FlaxStableDiffusionImg2ImgPipeline(FlaxDiffusionPipeline): r""" Flax-based pipeline for text-guided image-to-image generation using Stable Diffusion. This model inherits from [`FlaxDiffusionPipeline`]. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a particular device, etc.). Args: vae ([`FlaxAutoencoderKL`]): Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations. text_encoder ([`~transformers.FlaxCLIPTextModel`]): Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)). tokenizer ([`~transformers.CLIPTokenizer`]): A `CLIPTokenizer` to tokenize text. unet ([`FlaxUNet2DConditionModel`]): A `FlaxUNet2DConditionModel` to denoise the encoded image latents. scheduler ([`SchedulerMixin`]): A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of [`FlaxDDIMScheduler`], [`FlaxLMSDiscreteScheduler`], [`FlaxPNDMScheduler`], or [`FlaxDPMSolverMultistepScheduler`]. safety_checker ([`FlaxStableDiffusionSafetyChecker`]): Classification module that estimates whether generated images could be considered offensive or harmful. Please refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for more details about a model's potential harms. feature_extractor ([`~transformers.CLIPImageProcessor`]): A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`. """ def __init__( self, vae: FlaxAutoencoderKL, text_encoder: FlaxCLIPTextModel, tokenizer: CLIPTokenizer, unet: FlaxUNet2DConditionModel, scheduler: Union[ FlaxDDIMScheduler, FlaxPNDMScheduler, FlaxLMSDiscreteScheduler, FlaxDPMSolverMultistepScheduler ], safety_checker: FlaxStableDiffusionSafetyChecker, feature_extractor: CLIPImageProcessor, dtype: jnp.dtype = jnp.float32, ): super().__init__() self.dtype = dtype if safety_checker is None: logger.warn( f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure" " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered" " results in services or applications open to the public. Both the diffusers team and Hugging Face" " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling" " it only for use-cases that involve analyzing network behavior or auditing its results. For more" " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ." ) self.register_modules( vae=vae, text_encoder=text_encoder, tokenizer=tokenizer, unet=unet, scheduler=scheduler, safety_checker=safety_checker, feature_extractor=feature_extractor, ) self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) def prepare_inputs(self, prompt: Union[str, List[str]], image: Union[Image.Image, List[Image.Image]]): if not isinstance(prompt, (str, list)): raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") if not isinstance(image, (Image.Image, list)): raise ValueError(f"image has to be of type `PIL.Image.Image` or list but is {type(image)}") if isinstance(image, Image.Image): image = [image] processed_images = jnp.concatenate([preprocess(img, jnp.float32) for img in image]) text_input = self.tokenizer( prompt, padding="max_length", max_length=self.tokenizer.model_max_length, truncation=True, return_tensors="np", ) return text_input.input_ids, processed_images def _get_has_nsfw_concepts(self, features, params): has_nsfw_concepts = self.safety_checker(features, params) return has_nsfw_concepts def _run_safety_checker(self, images, safety_model_params, jit=False): # safety_model_params should already be replicated when jit is True pil_images = [Image.fromarray(image) for image in images] features = self.feature_extractor(pil_images, return_tensors="np").pixel_values if jit: features = shard(features) has_nsfw_concepts = _p_get_has_nsfw_concepts(self, features, safety_model_params) has_nsfw_concepts = unshard(has_nsfw_concepts) safety_model_params = unreplicate(safety_model_params) else: has_nsfw_concepts = self._get_has_nsfw_concepts(features, safety_model_params) images_was_copied = False for idx, has_nsfw_concept in enumerate(has_nsfw_concepts): if has_nsfw_concept: if not images_was_copied: images_was_copied = True images = images.copy() images[idx] = np.zeros(images[idx].shape, dtype=np.uint8) # black image if any(has_nsfw_concepts): warnings.warn( "Potential NSFW content was detected in one or more images. A black image will be returned" " instead. Try again with a different prompt and/or seed." ) return images, has_nsfw_concepts def get_timestep_start(self, num_inference_steps, strength): # get the original timestep using init_timestep init_timestep = min(int(num_inference_steps * strength), num_inference_steps) t_start = max(num_inference_steps - init_timestep, 0) return t_start def _generate( self, prompt_ids: jnp.ndarray, image: jnp.ndarray, params: Union[Dict, FrozenDict], prng_seed: jax.Array, start_timestep: int, num_inference_steps: int, height: int, width: int, guidance_scale: float, noise: Optional[jnp.ndarray] = None, neg_prompt_ids: Optional[jnp.ndarray] = None, ): if height % 8 != 0 or width % 8 != 0: raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.") # get prompt text embeddings prompt_embeds = self.text_encoder(prompt_ids, params=params["text_encoder"])[0] # TODO: currently it is assumed `do_classifier_free_guidance = guidance_scale > 1.0` # implement this conditional `do_classifier_free_guidance = guidance_scale > 1.0` batch_size = prompt_ids.shape[0] max_length = prompt_ids.shape[-1] if neg_prompt_ids is None: uncond_input = self.tokenizer( [""] * batch_size, padding="max_length", max_length=max_length, return_tensors="np" ).input_ids else: uncond_input = neg_prompt_ids negative_prompt_embeds = self.text_encoder(uncond_input, params=params["text_encoder"])[0] context = jnp.concatenate([negative_prompt_embeds, prompt_embeds]) latents_shape = ( batch_size, self.unet.config.in_channels, height // self.vae_scale_factor, width // self.vae_scale_factor, ) if noise is None: noise = jax.random.normal(prng_seed, shape=latents_shape, dtype=jnp.float32) else: if noise.shape != latents_shape: raise ValueError(f"Unexpected latents shape, got {noise.shape}, expected {latents_shape}") # Create init_latents init_latent_dist = self.vae.apply({"params": params["vae"]}, image, method=self.vae.encode).latent_dist init_latents = init_latent_dist.sample(key=prng_seed).transpose((0, 3, 1, 2)) init_latents = self.vae.config.scaling_factor * init_latents def loop_body(step, args): latents, scheduler_state = args # For classifier free guidance, we need to do two forward passes. # Here we concatenate the unconditional and text embeddings into a single batch # to avoid doing two forward passes latents_input = jnp.concatenate([latents] * 2) t = jnp.array(scheduler_state.timesteps, dtype=jnp.int32)[step] timestep = jnp.broadcast_to(t, latents_input.shape[0]) latents_input = self.scheduler.scale_model_input(scheduler_state, latents_input, t) # predict the noise residual noise_pred = self.unet.apply( {"params": params["unet"]}, jnp.array(latents_input), jnp.array(timestep, dtype=jnp.int32), encoder_hidden_states=context, ).sample # perform guidance noise_pred_uncond, noise_prediction_text = jnp.split(noise_pred, 2, axis=0) noise_pred = noise_pred_uncond + guidance_scale * (noise_prediction_text - noise_pred_uncond) # compute the previous noisy sample x_t -> x_t-1 latents, scheduler_state = self.scheduler.step(scheduler_state, noise_pred, t, latents).to_tuple() return latents, scheduler_state scheduler_state = self.scheduler.set_timesteps( params["scheduler"], num_inference_steps=num_inference_steps, shape=latents_shape ) latent_timestep = scheduler_state.timesteps[start_timestep : start_timestep + 1].repeat(batch_size) latents = self.scheduler.add_noise(params["scheduler"], init_latents, noise, latent_timestep) # scale the initial noise by the standard deviation required by the scheduler latents = latents * params["scheduler"].init_noise_sigma if DEBUG: # run with python for loop for i in range(start_timestep, num_inference_steps): latents, scheduler_state = loop_body(i, (latents, scheduler_state)) else: latents, _ = jax.lax.fori_loop(start_timestep, num_inference_steps, loop_body, (latents, scheduler_state)) # scale and decode the image latents with vae latents = 1 / self.vae.config.scaling_factor * latents image = self.vae.apply({"params": params["vae"]}, latents, method=self.vae.decode).sample image = (image / 2 + 0.5).clip(0, 1).transpose(0, 2, 3, 1) return image @replace_example_docstring(EXAMPLE_DOC_STRING) def __call__( self, prompt_ids: jnp.ndarray, image: jnp.ndarray, params: Union[Dict, FrozenDict], prng_seed: jax.Array, strength: float = 0.8, num_inference_steps: int = 50, height: Optional[int] = None, width: Optional[int] = None, guidance_scale: Union[float, jnp.ndarray] = 7.5, noise: jnp.ndarray = None, neg_prompt_ids: jnp.ndarray = None, return_dict: bool = True, jit: bool = False, ): r""" The call function to the pipeline for generation. Args: prompt_ids (`jnp.ndarray`): The prompt or prompts to guide image generation. image (`jnp.ndarray`): Array representing an image batch to be used as the starting point. params (`Dict` or `FrozenDict`): Dictionary containing the model parameters/weights. prng_seed (`jax.Array` or `jax.Array`): Array containing random number generator key. strength (`float`, *optional*, defaults to 0.8): Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a starting point and more noise is added the higher the `strength`. The number of denoising steps depends on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising process runs for the full number of iterations specified in `num_inference_steps`. A value of 1 essentially ignores `image`. num_inference_steps (`int`, *optional*, defaults to 50): The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. This parameter is modulated by `strength`. height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`): The height in pixels of the generated image. width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`): The width in pixels of the generated image. guidance_scale (`float`, *optional*, defaults to 7.5): A higher guidance scale value encourages the model to generate images closely linked to the text `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`. noise (`jnp.ndarray`, *optional*): Pre-generated noisy latents sampled from a Gaussian distribution to be used as inputs for image generation. Can be used to tweak the same generation with different prompts. The array is generated by sampling using the supplied random `generator`. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`~pipelines.stable_diffusion.FlaxStableDiffusionPipelineOutput`] instead of a plain tuple. jit (`bool`, defaults to `False`): Whether to run `pmap` versions of the generation and safety scoring functions. <Tip warning={true}> This argument exists because `__call__` is not yet end-to-end pmap-able. It will be removed in a future release. </Tip> Examples: Returns: [`~pipelines.stable_diffusion.FlaxStableDiffusionPipelineOutput`] or `tuple`: If `return_dict` is `True`, [`~pipelines.stable_diffusion.FlaxStableDiffusionPipelineOutput`] is returned, otherwise a `tuple` is returned where the first element is a list with the generated images and the second element is a list of `bool`s indicating whether the corresponding generated image contains "not-safe-for-work" (nsfw) content. """ # 0. Default height and width to unet height = height or self.unet.config.sample_size * self.vae_scale_factor width = width or self.unet.config.sample_size * self.vae_scale_factor if isinstance(guidance_scale, float): # Convert to a tensor so each device gets a copy. Follow the prompt_ids for # shape information, as they may be sharded (when `jit` is `True`), or not. guidance_scale = jnp.array([guidance_scale] * prompt_ids.shape[0]) if len(prompt_ids.shape) > 2: # Assume sharded guidance_scale = guidance_scale[:, None] start_timestep = self.get_timestep_start(num_inference_steps, strength) if jit: images = _p_generate( self, prompt_ids, image, params, prng_seed, start_timestep, num_inference_steps, height, width, guidance_scale, noise, neg_prompt_ids, ) else: images = self._generate( prompt_ids, image, params, prng_seed, start_timestep, num_inference_steps, height, width, guidance_scale, noise, neg_prompt_ids, ) if self.safety_checker is not None: safety_params = params["safety_checker"] images_uint8_casted = (images * 255).round().astype("uint8") num_devices, batch_size = images.shape[:2] images_uint8_casted = np.asarray(images_uint8_casted).reshape(num_devices * batch_size, height, width, 3) images_uint8_casted, has_nsfw_concept = self._run_safety_checker(images_uint8_casted, safety_params, jit) images = np.asarray(images) # block images if any(has_nsfw_concept): for i, is_nsfw in enumerate(has_nsfw_concept): if is_nsfw: images[i] = np.asarray(images_uint8_casted[i]) images = images.reshape(num_devices, batch_size, height, width, 3) else: images = np.asarray(images) has_nsfw_concept = False if not return_dict: return (images, has_nsfw_concept) return FlaxStableDiffusionPipelineOutput(images=images, nsfw_content_detected=has_nsfw_concept) # Static argnums are pipe, start_timestep, num_inference_steps, height, width. A change would trigger recompilation. # Non-static args are (sharded) input tensors mapped over their first dimension (hence, `0`). @partial( jax.pmap, in_axes=(None, 0, 0, 0, 0, None, None, None, None, 0, 0, 0), static_broadcasted_argnums=(0, 5, 6, 7, 8), ) def _p_generate( pipe, prompt_ids, image, params, prng_seed, start_timestep, num_inference_steps, height, width, guidance_scale, noise, neg_prompt_ids, ): return pipe._generate( prompt_ids, image, params, prng_seed, start_timestep, num_inference_steps, height, width, guidance_scale, noise, neg_prompt_ids, ) @partial(jax.pmap, static_broadcasted_argnums=(0,)) def _p_get_has_nsfw_concepts(pipe, features, params): return pipe._get_has_nsfw_concepts(features, params) def unshard(x: jnp.ndarray): # einops.rearrange(x, 'd b ... -> (d b) ...') num_devices, batch_size = x.shape[:2] rest = x.shape[2:] return x.reshape(num_devices * batch_size, *rest) def preprocess(image, dtype): w, h = image.size w, h = (x - x % 32 for x in (w, h)) # resize to integer multiple of 32 image = image.resize((w, h), resample=PIL_INTERPOLATION["lanczos"]) image = jnp.array(image).astype(dtype) / 255.0 image = image[None].transpose(0, 3, 1, 2) return 2.0 * image - 1.0

diffusers/src/diffusers/pipelines/stable_diffusion/pipeline_flax_stable_diffusion_img2img.py/0

# Copyright 2023 The HuggingFace Team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import inspect from typing import Any, Callable, Dict, List, Optional, Union import PIL.Image import torch from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection from ...image_processor import VaeImageProcessor from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin from ...models import AutoencoderKL, UNet2DConditionModel from ...models.embeddings import get_timestep_embedding from ...models.lora import adjust_lora_scale_text_encoder from ...schedulers import KarrasDiffusionSchedulers from ...utils import ( USE_PEFT_BACKEND, deprecate, logging, replace_example_docstring, scale_lora_layers, unscale_lora_layers, ) from ...utils.torch_utils import randn_tensor from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput from .stable_unclip_image_normalizer import StableUnCLIPImageNormalizer logger = logging.get_logger(__name__) # pylint: disable=invalid-name EXAMPLE_DOC_STRING = """ Examples: ```py >>> import requests >>> import torch >>> from PIL import Image >>> from io import BytesIO >>> from diffusers import StableUnCLIPImg2ImgPipeline >>> pipe = StableUnCLIPImg2ImgPipeline.from_pretrained( ... "fusing/stable-unclip-2-1-l-img2img", torch_dtype=torch.float16 ... ) # TODO update model path >>> pipe = pipe.to("cuda") >>> url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg" >>> response = requests.get(url) >>> init_image = Image.open(BytesIO(response.content)).convert("RGB") >>> init_image = init_image.resize((768, 512)) >>> prompt = "A fantasy landscape, trending on artstation" >>> images = pipe(prompt, init_image).images >>> images[0].save("fantasy_landscape.png") ``` """ class StableUnCLIPImg2ImgPipeline(DiffusionPipeline, TextualInversionLoaderMixin, LoraLoaderMixin): """ Pipeline for text-guided image-to-image generation using stable unCLIP. This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a particular device, etc.). The pipeline also inherits the following loading methods: - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings - [`~loaders.LoraLoaderMixin.load_lora_weights`] for loading LoRA weights - [`~loaders.LoraLoaderMixin.save_lora_weights`] for saving LoRA weights Args: feature_extractor ([`CLIPImageProcessor`]): Feature extractor for image pre-processing before being encoded. image_encoder ([`CLIPVisionModelWithProjection`]): CLIP vision model for encoding images. image_normalizer ([`StableUnCLIPImageNormalizer`]): Used to normalize the predicted image embeddings before the noise is applied and un-normalize the image embeddings after the noise has been applied. image_noising_scheduler ([`KarrasDiffusionSchedulers`]): Noise schedule for adding noise to the predicted image embeddings. The amount of noise to add is determined by the `noise_level`. tokenizer (`~transformers.CLIPTokenizer`): A [`~transformers.CLIPTokenizer`)]. text_encoder ([`~transformers.CLIPTextModel`]): Frozen [`~transformers.CLIPTextModel`] text-encoder. unet ([`UNet2DConditionModel`]): A [`UNet2DConditionModel`] to denoise the encoded image latents. scheduler ([`KarrasDiffusionSchedulers`]): A scheduler to be used in combination with `unet` to denoise the encoded image latents. vae ([`AutoencoderKL`]): Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. """ model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae" _exclude_from_cpu_offload = ["image_normalizer"] # image encoding components feature_extractor: CLIPImageProcessor image_encoder: CLIPVisionModelWithProjection # image noising components image_normalizer: StableUnCLIPImageNormalizer image_noising_scheduler: KarrasDiffusionSchedulers # regular denoising components tokenizer: CLIPTokenizer text_encoder: CLIPTextModel unet: UNet2DConditionModel scheduler: KarrasDiffusionSchedulers vae: AutoencoderKL def __init__( self, # image encoding components feature_extractor: CLIPImageProcessor, image_encoder: CLIPVisionModelWithProjection, # image noising components image_normalizer: StableUnCLIPImageNormalizer, image_noising_scheduler: KarrasDiffusionSchedulers, # regular denoising components tokenizer: CLIPTokenizer, text_encoder: CLIPTextModel, unet: UNet2DConditionModel, scheduler: KarrasDiffusionSchedulers, # vae vae: AutoencoderKL, ): super().__init__() self.register_modules( feature_extractor=feature_extractor, image_encoder=image_encoder, image_normalizer=image_normalizer, image_noising_scheduler=image_noising_scheduler, tokenizer=tokenizer, text_encoder=text_encoder, unet=unet, scheduler=scheduler, vae=vae, ) self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor) # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_vae_slicing def enable_vae_slicing(self): r""" Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to compute decoding in several steps. This is useful to save some memory and allow larger batch sizes. """ self.vae.enable_slicing() # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_vae_slicing def disable_vae_slicing(self): r""" Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to computing decoding in one step. """ self.vae.disable_slicing() # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt def _encode_prompt( self, prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt=None, prompt_embeds: Optional[torch.FloatTensor] = None, negative_prompt_embeds: Optional[torch.FloatTensor] = None, lora_scale: Optional[float] = None, **kwargs, ): deprecation_message = "`_encode_prompt()` is deprecated and it will be removed in a future version. Use `encode_prompt()` instead. Also, be aware that the output format changed from a concatenated tensor to a tuple." deprecate("_encode_prompt()", "1.0.0", deprecation_message, standard_warn=False) prompt_embeds_tuple = self.encode_prompt( prompt=prompt, device=device, num_images_per_prompt=num_images_per_prompt, do_classifier_free_guidance=do_classifier_free_guidance, negative_prompt=negative_prompt, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, lora_scale=lora_scale, **kwargs, ) # concatenate for backwards comp prompt_embeds = torch.cat([prompt_embeds_tuple[1], prompt_embeds_tuple[0]]) return prompt_embeds def _encode_image( self, image, device, batch_size, num_images_per_prompt, do_classifier_free_guidance, noise_level, generator, image_embeds, ): dtype = next(self.image_encoder.parameters()).dtype if isinstance(image, PIL.Image.Image): # the image embedding should repeated so it matches the total batch size of the prompt repeat_by = batch_size else: # assume the image input is already properly batched and just needs to be repeated so # it matches the num_images_per_prompt. # # NOTE(will) this is probably missing a few number of side cases. I.e. batched/non-batched # `image_embeds`. If those happen to be common use cases, let's think harder about # what the expected dimensions of inputs should be and how we handle the encoding. repeat_by = num_images_per_prompt if image_embeds is None: if not isinstance(image, torch.Tensor): image = self.feature_extractor(images=image, return_tensors="pt").pixel_values image = image.to(device=device, dtype=dtype) image_embeds = self.image_encoder(image).image_embeds image_embeds = self.noise_image_embeddings( image_embeds=image_embeds, noise_level=noise_level, generator=generator, ) # duplicate image embeddings for each generation per prompt, using mps friendly method image_embeds = image_embeds.unsqueeze(1) bs_embed, seq_len, _ = image_embeds.shape image_embeds = image_embeds.repeat(1, repeat_by, 1) image_embeds = image_embeds.view(bs_embed * repeat_by, seq_len, -1) image_embeds = image_embeds.squeeze(1) if do_classifier_free_guidance: negative_prompt_embeds = torch.zeros_like(image_embeds) # For classifier free guidance, we need to do two forward passes. # Here we concatenate the unconditional and text embeddings into a single batch # to avoid doing two forward passes image_embeds = torch.cat([negative_prompt_embeds, image_embeds]) return image_embeds # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_prompt def encode_prompt( self, prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt=None, prompt_embeds: Optional[torch.FloatTensor] = None, negative_prompt_embeds: Optional[torch.FloatTensor] = None, lora_scale: Optional[float] = None, clip_skip: Optional[int] = None, ): r""" Encodes the prompt into text encoder hidden states. Args: prompt (`str` or `List[str]`, *optional*): prompt to be encoded device: (`torch.device`): torch device num_images_per_prompt (`int`): number of images that should be generated per prompt do_classifier_free_guidance (`bool`): whether to use classifier free guidance or not negative_prompt (`str` or `List[str]`, *optional*): The prompt or prompts not to guide the image generation. If not defined, one has to pass `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`). prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, text embeddings will be generated from `prompt` input argument. negative_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input argument. lora_scale (`float`, *optional*): A LoRA scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded. clip_skip (`int`, *optional*): Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that the output of the pre-final layer will be used for computing the prompt embeddings. """ # set lora scale so that monkey patched LoRA # function of text encoder can correctly access it if lora_scale is not None and isinstance(self, LoraLoaderMixin): self._lora_scale = lora_scale # dynamically adjust the LoRA scale if not USE_PEFT_BACKEND: adjust_lora_scale_text_encoder(self.text_encoder, lora_scale) else: scale_lora_layers(self.text_encoder, lora_scale) if prompt is not None and isinstance(prompt, str): batch_size = 1 elif prompt is not None and isinstance(prompt, list): batch_size = len(prompt) else: batch_size = prompt_embeds.shape[0] if prompt_embeds is None: # textual inversion: procecss multi-vector tokens if necessary if isinstance(self, TextualInversionLoaderMixin): prompt = self.maybe_convert_prompt(prompt, self.tokenizer) text_inputs = self.tokenizer( prompt, padding="max_length", max_length=self.tokenizer.model_max_length, truncation=True, return_tensors="pt", ) text_input_ids = text_inputs.input_ids untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal( text_input_ids, untruncated_ids ): removed_text = self.tokenizer.batch_decode( untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1] ) logger.warning( "The following part of your input was truncated because CLIP can only handle sequences up to" f" {self.tokenizer.model_max_length} tokens: {removed_text}" ) if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask: attention_mask = text_inputs.attention_mask.to(device) else: attention_mask = None if clip_skip is None: prompt_embeds = self.text_encoder(text_input_ids.to(device), attention_mask=attention_mask) prompt_embeds = prompt_embeds[0] else: prompt_embeds = self.text_encoder( text_input_ids.to(device), attention_mask=attention_mask, output_hidden_states=True ) # Access the `hidden_states` first, that contains a tuple of # all the hidden states from the encoder layers. Then index into # the tuple to access the hidden states from the desired layer. prompt_embeds = prompt_embeds[-1][-(clip_skip + 1)] # We also need to apply the final LayerNorm here to not mess with the # representations. The `last_hidden_states` that we typically use for # obtaining the final prompt representations passes through the LayerNorm # layer. prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds) if self.text_encoder is not None: prompt_embeds_dtype = self.text_encoder.dtype elif self.unet is not None: prompt_embeds_dtype = self.unet.dtype else: prompt_embeds_dtype = prompt_embeds.dtype prompt_embeds = prompt_embeds.to(dtype=prompt_embeds_dtype, device=device) bs_embed, seq_len, _ = prompt_embeds.shape # duplicate text embeddings for each generation per prompt, using mps friendly method prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1) prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1) # get unconditional embeddings for classifier free guidance if do_classifier_free_guidance and negative_prompt_embeds is None: uncond_tokens: List[str] if negative_prompt is None: uncond_tokens = [""] * batch_size elif prompt is not None and type(prompt) is not type(negative_prompt): raise TypeError( f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !=" f" {type(prompt)}." ) elif isinstance(negative_prompt, str): uncond_tokens = [negative_prompt] elif batch_size != len(negative_prompt): raise ValueError( f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:" f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches" " the batch size of `prompt`." ) else: uncond_tokens = negative_prompt # textual inversion: procecss multi-vector tokens if necessary if isinstance(self, TextualInversionLoaderMixin): uncond_tokens = self.maybe_convert_prompt(uncond_tokens, self.tokenizer) max_length = prompt_embeds.shape[1] uncond_input = self.tokenizer( uncond_tokens, padding="max_length", max_length=max_length, truncation=True, return_tensors="pt", ) if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask: attention_mask = uncond_input.attention_mask.to(device) else: attention_mask = None negative_prompt_embeds = self.text_encoder( uncond_input.input_ids.to(device), attention_mask=attention_mask, ) negative_prompt_embeds = negative_prompt_embeds[0] if do_classifier_free_guidance: # duplicate unconditional embeddings for each generation per prompt, using mps friendly method seq_len = negative_prompt_embeds.shape[1] negative_prompt_embeds = negative_prompt_embeds.to(dtype=prompt_embeds_dtype, device=device) negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1) negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1) if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND: # Retrieve the original scale by scaling back the LoRA layers unscale_lora_layers(self.text_encoder, lora_scale) return prompt_embeds, negative_prompt_embeds # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents def decode_latents(self, latents): deprecation_message = "The decode_latents method is deprecated and will be removed in 1.0.0. Please use VaeImageProcessor.postprocess(...) instead" deprecate("decode_latents", "1.0.0", deprecation_message, standard_warn=False) latents = 1 / self.vae.config.scaling_factor * latents image = self.vae.decode(latents, return_dict=False)[0] image = (image / 2 + 0.5).clamp(0, 1) # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16 image = image.cpu().permute(0, 2, 3, 1).float().numpy() return image # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs def prepare_extra_step_kwargs(self, generator, eta): # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers. # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502 # and should be between [0, 1] accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys()) extra_step_kwargs = {} if accepts_eta: extra_step_kwargs["eta"] = eta # check if the scheduler accepts generator accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys()) if accepts_generator: extra_step_kwargs["generator"] = generator return extra_step_kwargs def check_inputs( self, prompt, image, height, width, callback_steps, noise_level, negative_prompt=None, prompt_embeds=None, negative_prompt_embeds=None, image_embeds=None, ): if height % 8 != 0 or width % 8 != 0: raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.") if (callback_steps is None) or ( callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0) ): raise ValueError( f"`callback_steps` has to be a positive integer but is {callback_steps} of type" f" {type(callback_steps)}." ) if prompt is not None and prompt_embeds is not None: raise ValueError( "Provide either `prompt` or `prompt_embeds`. Please make sure to define only one of the two." ) if prompt is None and prompt_embeds is None: raise ValueError( "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined." ) if prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)): raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") if negative_prompt is not None and negative_prompt_embeds is not None: raise ValueError( "Provide either `negative_prompt` or `negative_prompt_embeds`. Cannot leave both `negative_prompt` and `negative_prompt_embeds` undefined." ) if prompt is not None and negative_prompt is not None: if type(prompt) is not type(negative_prompt): raise TypeError( f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !=" f" {type(prompt)}." ) if prompt_embeds is not None and negative_prompt_embeds is not None: if prompt_embeds.shape != negative_prompt_embeds.shape: raise ValueError( "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but" f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`" f" {negative_prompt_embeds.shape}." ) if noise_level < 0 or noise_level >= self.image_noising_scheduler.config.num_train_timesteps: raise ValueError( f"`noise_level` must be between 0 and {self.image_noising_scheduler.config.num_train_timesteps - 1}, inclusive." ) if image is not None and image_embeds is not None: raise ValueError( "Provide either `image` or `image_embeds`. Please make sure to define only one of the two." ) if image is None and image_embeds is None: raise ValueError( "Provide either `image` or `image_embeds`. Cannot leave both `image` and `image_embeds` undefined." ) if image is not None: if ( not isinstance(image, torch.Tensor) and not isinstance(image, PIL.Image.Image) and not isinstance(image, list) ): raise ValueError( "`image` has to be of type `torch.FloatTensor` or `PIL.Image.Image` or `List[PIL.Image.Image]` but is" f" {type(image)}" ) # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None): shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor) if isinstance(generator, list) and len(generator) != batch_size: raise ValueError( f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" f" size of {batch_size}. Make sure the batch size matches the length of the generators." ) if latents is None: latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype) else: latents = latents.to(device) # scale the initial noise by the standard deviation required by the scheduler latents = latents * self.scheduler.init_noise_sigma return latents # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_unclip.StableUnCLIPPipeline.noise_image_embeddings def noise_image_embeddings( self, image_embeds: torch.Tensor, noise_level: int, noise: Optional[torch.FloatTensor] = None, generator: Optional[torch.Generator] = None, ): """ Add noise to the image embeddings. The amount of noise is controlled by a `noise_level` input. A higher `noise_level` increases the variance in the final un-noised images. The noise is applied in two ways: 1. A noise schedule is applied directly to the embeddings. 2. A vector of sinusoidal time embeddings are appended to the output. In both cases, the amount of noise is controlled by the same `noise_level`. The embeddings are normalized before the noise is applied and un-normalized after the noise is applied. """ if noise is None: noise = randn_tensor( image_embeds.shape, generator=generator, device=image_embeds.device, dtype=image_embeds.dtype ) noise_level = torch.tensor([noise_level] * image_embeds.shape[0], device=image_embeds.device) self.image_normalizer.to(image_embeds.device) image_embeds = self.image_normalizer.scale(image_embeds) image_embeds = self.image_noising_scheduler.add_noise(image_embeds, timesteps=noise_level, noise=noise) image_embeds = self.image_normalizer.unscale(image_embeds) noise_level = get_timestep_embedding( timesteps=noise_level, embedding_dim=image_embeds.shape[-1], flip_sin_to_cos=True, downscale_freq_shift=0 ) # `get_timestep_embeddings` does not contain any weights and will always return f32 tensors, # but we might actually be running in fp16. so we need to cast here. # there might be better ways to encapsulate this. noise_level = noise_level.to(image_embeds.dtype) image_embeds = torch.cat((image_embeds, noise_level), 1) return image_embeds @torch.no_grad() @replace_example_docstring(EXAMPLE_DOC_STRING) def __call__( self, image: Union[torch.FloatTensor, PIL.Image.Image] = None, prompt: Union[str, List[str]] = None, height: Optional[int] = None, width: Optional[int] = None, num_inference_steps: int = 20, guidance_scale: float = 10, negative_prompt: Optional[Union[str, List[str]]] = None, num_images_per_prompt: Optional[int] = 1, eta: float = 0.0, generator: Optional[torch.Generator] = None, latents: Optional[torch.FloatTensor] = None, prompt_embeds: Optional[torch.FloatTensor] = None, negative_prompt_embeds: Optional[torch.FloatTensor] = None, output_type: Optional[str] = "pil", return_dict: bool = True, callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None, callback_steps: int = 1, cross_attention_kwargs: Optional[Dict[str, Any]] = None, noise_level: int = 0, image_embeds: Optional[torch.FloatTensor] = None, clip_skip: Optional[int] = None, ): r""" The call function to the pipeline for generation. Args: prompt (`str` or `List[str]`, *optional*): The prompt or prompts to guide the image generation. If not defined, either `prompt_embeds` will be used or prompt is initialized to `""`. image (`torch.FloatTensor` or `PIL.Image.Image`): `Image` or tensor representing an image batch. The image is encoded to its CLIP embedding which the `unet` is conditioned on. The image is _not_ encoded by the `vae` and then used as the latents in the denoising process like it is in the standard Stable Diffusion text-guided image variation process. height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`): The height in pixels of the generated image. width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`): The width in pixels of the generated image. num_inference_steps (`int`, *optional*, defaults to 20): The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. guidance_scale (`float`, *optional*, defaults to 10.0): A higher guidance scale value encourages the model to generate images closely linked to the text `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`. negative_prompt (`str` or `List[str]`, *optional*): The prompt or prompts to guide what to not include in image generation. If not defined, you need to pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`). num_images_per_prompt (`int`, *optional*, defaults to 1): The number of images to generate per prompt. eta (`float`, *optional*, defaults to 0.0): Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers. generator (`torch.Generator` or `List[torch.Generator]`, *optional*): A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation deterministic. latents (`torch.FloatTensor`, *optional*): Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image generation. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor is generated by sampling using the supplied random `generator`. prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from the `prompt` input argument. negative_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument. output_type (`str`, *optional*, defaults to `"pil"`): The output format of the generated image. Choose between `PIL.Image` or `np.array`. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`~pipelines.ImagePipelineOutput`] instead of a plain tuple. callback (`Callable`, *optional*): A function that calls every `callback_steps` steps during inference. The function is called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`. callback_steps (`int`, *optional*, defaults to 1): The frequency at which the `callback` function is called. If not specified, the callback is called at every step. cross_attention_kwargs (`dict`, *optional*): A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py). noise_level (`int`, *optional*, defaults to `0`): The amount of noise to add to the image embeddings. A higher `noise_level` increases the variance in the final un-noised images. See [`StableUnCLIPPipeline.noise_image_embeddings`] for more details. image_embeds (`torch.FloatTensor`, *optional*): Pre-generated CLIP embeddings to condition the `unet` on. These latents are not used in the denoising process. If you want to provide pre-generated latents, pass them to `__call__` as `latents`. clip_skip (`int`, *optional*): Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that the output of the pre-final layer will be used for computing the prompt embeddings. Examples: Returns: [`~pipelines.ImagePipelineOutput`] or `tuple`: [`~ pipeline_utils.ImagePipelineOutput`] if `return_dict` is True, otherwise a `tuple`. When returning a tuple, the first element is a list with the generated images. """ # 0. Default height and width to unet height = height or self.unet.config.sample_size * self.vae_scale_factor width = width or self.unet.config.sample_size * self.vae_scale_factor if prompt is None and prompt_embeds is None: prompt = len(image) * [""] if isinstance(image, list) else "" # 1. Check inputs. Raise error if not correct self.check_inputs( prompt=prompt, image=image, height=height, width=width, callback_steps=callback_steps, noise_level=noise_level, negative_prompt=negative_prompt, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, image_embeds=image_embeds, ) # 2. Define call parameters if prompt is not None and isinstance(prompt, str): batch_size = 1 elif prompt is not None and isinstance(prompt, list): batch_size = len(prompt) else: batch_size = prompt_embeds.shape[0] batch_size = batch_size * num_images_per_prompt device = self._execution_device # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2) # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1` # corresponds to doing no classifier free guidance. do_classifier_free_guidance = guidance_scale > 1.0 # 3. Encode input prompt text_encoder_lora_scale = ( cross_attention_kwargs.get("scale", None) if cross_attention_kwargs is not None else None ) prompt_embeds, negative_prompt_embeds = self.encode_prompt( prompt=prompt, device=device, num_images_per_prompt=num_images_per_prompt, do_classifier_free_guidance=do_classifier_free_guidance, negative_prompt=negative_prompt, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_prompt_embeds, lora_scale=text_encoder_lora_scale, ) # For classifier free guidance, we need to do two forward passes. # Here we concatenate the unconditional and text embeddings into a single batch # to avoid doing two forward passes if do_classifier_free_guidance: prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds]) # 4. Encoder input image noise_level = torch.tensor([noise_level], device=device) image_embeds = self._encode_image( image=image, device=device, batch_size=batch_size, num_images_per_prompt=num_images_per_prompt, do_classifier_free_guidance=do_classifier_free_guidance, noise_level=noise_level, generator=generator, image_embeds=image_embeds, ) # 5. Prepare timesteps self.scheduler.set_timesteps(num_inference_steps, device=device) timesteps = self.scheduler.timesteps # 6. Prepare latent variables num_channels_latents = self.unet.config.in_channels latents = self.prepare_latents( batch_size=batch_size, num_channels_latents=num_channels_latents, height=height, width=width, dtype=prompt_embeds.dtype, device=device, generator=generator, latents=latents, ) # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta) # 8. Denoising loop for i, t in enumerate(self.progress_bar(timesteps)): latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents latent_model_input = self.scheduler.scale_model_input(latent_model_input, t) # predict the noise residual noise_pred = self.unet( latent_model_input, t, encoder_hidden_states=prompt_embeds, class_labels=image_embeds, cross_attention_kwargs=cross_attention_kwargs, return_dict=False, )[0] # perform guidance if do_classifier_free_guidance: noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond) # compute the previous noisy sample x_t -> x_t-1 latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0] if callback is not None and i % callback_steps == 0: step_idx = i // getattr(self.scheduler, "order", 1) callback(step_idx, t, latents) # 9. Post-processing if not output_type == "latent": image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0] else: image = latents image = self.image_processor.postprocess(image, output_type=output_type) # Offload all models self.maybe_free_model_hooks() if not return_dict: return (image,) return ImagePipelineOutput(images=image)

diffusers/src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip_img2img.py/0

from typing import Optional import numpy as np import torch from torch import nn from transformers import GPT2Config, GPT2LMHeadModel from transformers.modeling_utils import ModuleUtilsMixin from ...configuration_utils import ConfigMixin, register_to_config from ...models import ModelMixin # Modified from ClipCaptionModel in https://github.com/thu-ml/unidiffuser/blob/main/libs/caption_decoder.py class UniDiffuserTextDecoder(ModelMixin, ConfigMixin, ModuleUtilsMixin): """ Text decoder model for a image-text [UniDiffuser](https://arxiv.org/pdf/2303.06555.pdf) model. This is used to generate text from the UniDiffuser image-text embedding. Parameters: prefix_length (`int`): Max number of prefix tokens that will be supplied to the model. prefix_inner_dim (`int`): The hidden size of the incoming prefix embeddings. For UniDiffuser, this would be the hidden dim of the CLIP text encoder. prefix_hidden_dim (`int`, *optional*): Hidden dim of the MLP if we encode the prefix. vocab_size (`int`, *optional*, defaults to 50257): Vocabulary size of the GPT-2 model. Defines the number of different tokens that can be represented by the `inputs_ids` passed when calling [`GPT2Model`] or [`TFGPT2Model`]. n_positions (`int`, *optional*, defaults to 1024): The maximum sequence length that this model might ever be used with. Typically set this to something large just in case (e.g., 512 or 1024 or 2048). n_embd (`int`, *optional*, defaults to 768): Dimensionality of the embeddings and hidden states. n_layer (`int`, *optional*, defaults to 12): Number of hidden layers in the Transformer encoder. n_head (`int`, *optional*, defaults to 12): Number of attention heads for each attention layer in the Transformer encoder. n_inner (`int`, *optional*, defaults to None): Dimensionality of the inner feed-forward layers. `None` will set it to 4 times n_embd activation_function (`str`, *optional*, defaults to `"gelu"`): Activation function, to be selected in the list `["relu", "silu", "gelu", "tanh", "gelu_new"]`. resid_pdrop (`float`, *optional*, defaults to 0.1): The dropout probability for all fully connected layers in the embeddings, encoder, and pooler. embd_pdrop (`float`, *optional*, defaults to 0.1): The dropout ratio for the embeddings. attn_pdrop (`float`, *optional*, defaults to 0.1): The dropout ratio for the attention. layer_norm_epsilon (`float`, *optional*, defaults to 1e-5): The epsilon to use in the layer normalization layers. initializer_range (`float`, *optional*, defaults to 0.02): The standard deviation of the truncated_normal_initializer for initializing all weight matrices. scale_attn_weights (`bool`, *optional*, defaults to `True`): Scale attention weights by dividing by sqrt(hidden_size).. use_cache (`bool`, *optional*, defaults to `True`): Whether or not the model should return the last key/values attentions (not used by all models). scale_attn_by_inverse_layer_idx (`bool`, *optional*, defaults to `False`): Whether to additionally scale attention weights by `1 / layer_idx + 1`. reorder_and_upcast_attn (`bool`, *optional*, defaults to `False`): Whether to scale keys (K) prior to computing attention (dot-product) and upcast attention dot-product/softmax to float() when training with mixed precision. """ _keys_to_ignore_on_load_unexpected = [r"h\.\d+\.attn\.bias", r"h\.\d+\.attn\.masked_bias"] @register_to_config def __init__( self, prefix_length: int, prefix_inner_dim: int, prefix_hidden_dim: Optional[int] = None, vocab_size: int = 50257, # Start of GPT2 config args n_positions: int = 1024, n_embd: int = 768, n_layer: int = 12, n_head: int = 12, n_inner: Optional[int] = None, activation_function: str = "gelu_new", resid_pdrop: float = 0.1, embd_pdrop: float = 0.1, attn_pdrop: float = 0.1, layer_norm_epsilon: float = 1e-5, initializer_range: float = 0.02, scale_attn_weights: bool = True, use_cache: bool = True, scale_attn_by_inverse_layer_idx: bool = False, reorder_and_upcast_attn: bool = False, ): super().__init__() self.prefix_length = prefix_length if prefix_inner_dim != n_embd and prefix_hidden_dim is None: raise ValueError( f"`prefix_hidden_dim` cannot be `None` when `prefix_inner_dim`: {prefix_hidden_dim} and" f" `n_embd`: {n_embd} are not equal." ) self.prefix_inner_dim = prefix_inner_dim self.prefix_hidden_dim = prefix_hidden_dim self.encode_prefix = ( nn.Linear(self.prefix_inner_dim, self.prefix_hidden_dim) if self.prefix_hidden_dim is not None else nn.Identity() ) self.decode_prefix = ( nn.Linear(self.prefix_hidden_dim, n_embd) if self.prefix_hidden_dim is not None else nn.Identity() ) gpt_config = GPT2Config( vocab_size=vocab_size, n_positions=n_positions, n_embd=n_embd, n_layer=n_layer, n_head=n_head, n_inner=n_inner, activation_function=activation_function, resid_pdrop=resid_pdrop, embd_pdrop=embd_pdrop, attn_pdrop=attn_pdrop, layer_norm_epsilon=layer_norm_epsilon, initializer_range=initializer_range, scale_attn_weights=scale_attn_weights, use_cache=use_cache, scale_attn_by_inverse_layer_idx=scale_attn_by_inverse_layer_idx, reorder_and_upcast_attn=reorder_and_upcast_attn, ) self.transformer = GPT2LMHeadModel(gpt_config) def forward( self, input_ids: torch.Tensor, prefix_embeds: torch.Tensor, attention_mask: Optional[torch.Tensor] = None, labels: Optional[torch.Tensor] = None, ): """ Args: input_ids (`torch.Tensor` of shape `(N, max_seq_len)`): Text tokens to use for inference. prefix_embeds (`torch.Tensor` of shape `(N, prefix_length, 768)`): Prefix embedding to preprend to the embedded tokens. attention_mask (`torch.Tensor` of shape `(N, prefix_length + max_seq_len, 768)`, *optional*): Attention mask for the prefix embedding. labels (`torch.Tensor`, *optional*): Labels to use for language modeling. """ embedding_text = self.transformer.transformer.wte(input_ids) hidden = self.encode_prefix(prefix_embeds) prefix_embeds = self.decode_prefix(hidden) embedding_cat = torch.cat((prefix_embeds, embedding_text), dim=1) if labels is not None: dummy_token = self.get_dummy_token(input_ids.shape[0], input_ids.device) labels = torch.cat((dummy_token, input_ids), dim=1) out = self.transformer(inputs_embeds=embedding_cat, labels=labels, attention_mask=attention_mask) if self.prefix_hidden_dim is not None: return out, hidden else: return out def get_dummy_token(self, batch_size: int, device: torch.device) -> torch.Tensor: return torch.zeros(batch_size, self.prefix_length, dtype=torch.int64, device=device) def encode(self, prefix): return self.encode_prefix(prefix) @torch.no_grad() def generate_captions(self, features, eos_token_id, device): """ Generate captions given text embedding features. Returns list[L]. Args: features (`torch.Tensor` of shape `(B, L, D)`): Text embedding features to generate captions from. eos_token_id (`int`): The token ID of the EOS token for the text decoder model. device: Device to perform text generation on. Returns: `List[str]`: A list of strings generated from the decoder model. """ features = torch.split(features, 1, dim=0) generated_tokens = [] generated_seq_lengths = [] for feature in features: feature = self.decode_prefix(feature.to(device)) # back to the clip feature # Only support beam search for now output_tokens, seq_lengths = self.generate_beam( input_embeds=feature, device=device, eos_token_id=eos_token_id ) generated_tokens.append(output_tokens[0]) generated_seq_lengths.append(seq_lengths[0]) generated_tokens = torch.stack(generated_tokens) generated_seq_lengths = torch.stack(generated_seq_lengths) return generated_tokens, generated_seq_lengths @torch.no_grad() def generate_beam( self, input_ids=None, input_embeds=None, device=None, beam_size: int = 5, entry_length: int = 67, temperature: float = 1.0, eos_token_id: Optional[int] = None, ): """ Generates text using the given tokenizer and text prompt or token embedding via beam search. This implementation is based on the beam search implementation from the [original UniDiffuser code](https://github.com/thu-ml/unidiffuser/blob/main/libs/caption_decoder.py#L89). Args: eos_token_id (`int`, *optional*): The token ID of the EOS token for the text decoder model. input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`, *optional*): Tokenizer indices of input sequence tokens in the vocabulary. One of `input_ids` and `input_embeds` must be supplied. input_embeds (`torch.FloatTensor` of shape `(batch_size, seq_len, hidden_size)`, *optional*): An embedded representation to directly pass to the transformer as a prefix for beam search. One of `input_ids` and `input_embeds` must be supplied. device: The device to perform beam search on. beam_size (`int`, *optional*, defaults to `5`): The number of best states to store during beam search. entry_length (`int`, *optional*, defaults to `67`): The number of iterations to run beam search. temperature (`float`, *optional*, defaults to 1.0): The temperature to use when performing the softmax over logits from the decoding model. Returns: `Tuple(torch.Tensor, torch.Tensor)`: A tuple of tensors where the first element is a tensor of generated token sequences sorted by score in descending order, and the second element is the sequence lengths corresponding to those sequences. """ # Generates text until stop_token is reached using beam search with the desired beam size. stop_token_index = eos_token_id tokens = None scores = None seq_lengths = torch.ones(beam_size, device=device, dtype=torch.int) is_stopped = torch.zeros(beam_size, device=device, dtype=torch.bool) if input_embeds is not None: generated = input_embeds else: generated = self.transformer.transformer.wte(input_ids) for i in range(entry_length): outputs = self.transformer(inputs_embeds=generated) logits = outputs.logits logits = logits[:, -1, :] / (temperature if temperature > 0 else 1.0) logits = logits.softmax(-1).log() if scores is None: scores, next_tokens = logits.topk(beam_size, -1) generated = generated.expand(beam_size, *generated.shape[1:]) next_tokens, scores = next_tokens.permute(1, 0), scores.squeeze(0) if tokens is None: tokens = next_tokens else: tokens = tokens.expand(beam_size, *tokens.shape[1:]) tokens = torch.cat((tokens, next_tokens), dim=1) else: logits[is_stopped] = -float(np.inf) logits[is_stopped, 0] = 0 scores_sum = scores[:, None] + logits seq_lengths[~is_stopped] += 1 scores_sum_average = scores_sum / seq_lengths[:, None] scores_sum_average, next_tokens = scores_sum_average.view(-1).topk(beam_size, -1) next_tokens_source = next_tokens // scores_sum.shape[1] seq_lengths = seq_lengths[next_tokens_source] next_tokens = next_tokens % scores_sum.shape[1] next_tokens = next_tokens.unsqueeze(1) tokens = tokens[next_tokens_source] tokens = torch.cat((tokens, next_tokens), dim=1) generated = generated[next_tokens_source] scores = scores_sum_average * seq_lengths is_stopped = is_stopped[next_tokens_source] next_token_embed = self.transformer.transformer.wte(next_tokens.squeeze()).view(generated.shape[0], 1, -1) generated = torch.cat((generated, next_token_embed), dim=1) is_stopped = is_stopped + next_tokens.eq(stop_token_index).squeeze() if is_stopped.all(): break scores = scores / seq_lengths order = scores.argsort(descending=True) # tokens tensors are already padded to max_seq_length output_texts = [tokens[i] for i in order] output_texts = torch.stack(output_texts, dim=0) seq_lengths = torch.tensor([seq_lengths[i] for i in order], dtype=seq_lengths.dtype) return output_texts, seq_lengths

diffusers/src/diffusers/pipelines/unidiffuser/modeling_text_decoder.py/0

# Copyright 2023 Google Brain and The HuggingFace Team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # DISCLAIMER: This file is strongly influenced by https://github.com/yang-song/score_sde_pytorch import math from typing import Union import torch from ...configuration_utils import ConfigMixin, register_to_config from ...utils.torch_utils import randn_tensor from ..scheduling_utils import SchedulerMixin class ScoreSdeVpScheduler(SchedulerMixin, ConfigMixin): """ `ScoreSdeVpScheduler` is a variance preserving stochastic differential equation (SDE) scheduler. This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. Check the superclass documentation for the generic methods the library implements for all schedulers such as loading and saving. Args: num_train_timesteps (`int`, defaults to 2000): The number of diffusion steps to train the model. beta_min (`int`, defaults to 0.1): beta_max (`int`, defaults to 20): sampling_eps (`int`, defaults to 1e-3): The end value of sampling where timesteps decrease progressively from 1 to epsilon. """ order = 1 @register_to_config def __init__(self, num_train_timesteps=2000, beta_min=0.1, beta_max=20, sampling_eps=1e-3): self.sigmas = None self.discrete_sigmas = None self.timesteps = None def set_timesteps(self, num_inference_steps, device: Union[str, torch.device] = None): """ Sets the continuous timesteps used for the diffusion chain (to be run before inference). Args: num_inference_steps (`int`): The number of diffusion steps used when generating samples with a pre-trained model. device (`str` or `torch.device`, *optional*): The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. """ self.timesteps = torch.linspace(1, self.config.sampling_eps, num_inference_steps, device=device) def step_pred(self, score, x, t, generator=None): """ Predict the sample from the previous timestep by reversing the SDE. This function propagates the diffusion process from the learned model outputs (most often the predicted noise). Args: score (): x (): t (): generator (`torch.Generator`, *optional*): A random number generator. """ if self.timesteps is None: raise ValueError( "`self.timesteps` is not set, you need to run 'set_timesteps' after creating the scheduler" ) # TODO(Patrick) better comments + non-PyTorch # postprocess model score log_mean_coeff = -0.25 * t**2 * (self.config.beta_max - self.config.beta_min) - 0.5 * t * self.config.beta_min std = torch.sqrt(1.0 - torch.exp(2.0 * log_mean_coeff)) std = std.flatten() while len(std.shape) < len(score.shape): std = std.unsqueeze(-1) score = -score / std # compute dt = -1.0 / len(self.timesteps) beta_t = self.config.beta_min + t * (self.config.beta_max - self.config.beta_min) beta_t = beta_t.flatten() while len(beta_t.shape) < len(x.shape): beta_t = beta_t.unsqueeze(-1) drift = -0.5 * beta_t * x diffusion = torch.sqrt(beta_t) drift = drift - diffusion**2 * score x_mean = x + drift * dt # add noise noise = randn_tensor(x.shape, layout=x.layout, generator=generator, device=x.device, dtype=x.dtype) x = x_mean + diffusion * math.sqrt(-dt) * noise return x, x_mean def __len__(self): return self.config.num_train_timesteps

diffusers/src/diffusers/schedulers/deprecated/scheduling_sde_vp.py/0

# Copyright 2023 Katherine Crowson, The HuggingFace Team and hlky. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import math from typing import List, Optional, Tuple, Union import numpy as np import torch import torchsde from ..configuration_utils import ConfigMixin, register_to_config from .scheduling_utils import KarrasDiffusionSchedulers, SchedulerMixin, SchedulerOutput class BatchedBrownianTree: """A wrapper around torchsde.BrownianTree that enables batches of entropy.""" def __init__(self, x, t0, t1, seed=None, **kwargs): t0, t1, self.sign = self.sort(t0, t1) w0 = kwargs.get("w0", torch.zeros_like(x)) if seed is None: seed = torch.randint(0, 2**63 - 1, []).item() self.batched = True try: assert len(seed) == x.shape[0] w0 = w0[0] except TypeError: seed = [seed] self.batched = False self.trees = [torchsde.BrownianTree(t0, w0, t1, entropy=s, **kwargs) for s in seed] @staticmethod def sort(a, b): return (a, b, 1) if a < b else (b, a, -1) def __call__(self, t0, t1): t0, t1, sign = self.sort(t0, t1) w = torch.stack([tree(t0, t1) for tree in self.trees]) * (self.sign * sign) return w if self.batched else w[0] class BrownianTreeNoiseSampler: """A noise sampler backed by a torchsde.BrownianTree. Args: x (Tensor): The tensor whose shape, device and dtype to use to generate random samples. sigma_min (float): The low end of the valid interval. sigma_max (float): The high end of the valid interval. seed (int or List[int]): The random seed. If a list of seeds is supplied instead of a single integer, then the noise sampler will use one BrownianTree per batch item, each with its own seed. transform (callable): A function that maps sigma to the sampler's internal timestep. """ def __init__(self, x, sigma_min, sigma_max, seed=None, transform=lambda x: x): self.transform = transform t0, t1 = self.transform(torch.as_tensor(sigma_min)), self.transform(torch.as_tensor(sigma_max)) self.tree = BatchedBrownianTree(x, t0, t1, seed) def __call__(self, sigma, sigma_next): t0, t1 = self.transform(torch.as_tensor(sigma)), self.transform(torch.as_tensor(sigma_next)) return self.tree(t0, t1) / (t1 - t0).abs().sqrt() # Copied from diffusers.schedulers.scheduling_ddpm.betas_for_alpha_bar def betas_for_alpha_bar( num_diffusion_timesteps, max_beta=0.999, alpha_transform_type="cosine", ): """ Create a beta schedule that discretizes the given alpha_t_bar function, which defines the cumulative product of (1-beta) over time from t = [0,1]. Contains a function alpha_bar that takes an argument t and transforms it to the cumulative product of (1-beta) up to that part of the diffusion process. Args: num_diffusion_timesteps (`int`): the number of betas to produce. max_beta (`float`): the maximum beta to use; use values lower than 1 to prevent singularities. alpha_transform_type (`str`, *optional*, default to `cosine`): the type of noise schedule for alpha_bar. Choose from `cosine` or `exp` Returns: betas (`np.ndarray`): the betas used by the scheduler to step the model outputs """ if alpha_transform_type == "cosine": def alpha_bar_fn(t): return math.cos((t + 0.008) / 1.008 * math.pi / 2) ** 2 elif alpha_transform_type == "exp": def alpha_bar_fn(t): return math.exp(t * -12.0) else: raise ValueError(f"Unsupported alpha_tranform_type: {alpha_transform_type}") betas = [] for i in range(num_diffusion_timesteps): t1 = i / num_diffusion_timesteps t2 = (i + 1) / num_diffusion_timesteps betas.append(min(1 - alpha_bar_fn(t2) / alpha_bar_fn(t1), max_beta)) return torch.tensor(betas, dtype=torch.float32) class DPMSolverSDEScheduler(SchedulerMixin, ConfigMixin): """ DPMSolverSDEScheduler implements the stochastic sampler from the [Elucidating the Design Space of Diffusion-Based Generative Models](https://huggingface.co/papers/2206.00364) paper. This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. Check the superclass documentation for the generic methods the library implements for all schedulers such as loading and saving. Args: num_train_timesteps (`int`, defaults to 1000): The number of diffusion steps to train the model. beta_start (`float`, defaults to 0.00085): The starting `beta` value of inference. beta_end (`float`, defaults to 0.012): The final `beta` value. beta_schedule (`str`, defaults to `"linear"`): The beta schedule, a mapping from a beta range to a sequence of betas for stepping the model. Choose from `linear` or `scaled_linear`. trained_betas (`np.ndarray`, *optional*): Pass an array of betas directly to the constructor to bypass `beta_start` and `beta_end`. prediction_type (`str`, defaults to `epsilon`, *optional*): Prediction type of the scheduler function; can be `epsilon` (predicts the noise of the diffusion process), `sample` (directly predicts the noisy sample`) or `v_prediction` (see section 2.4 of [Imagen Video](https://imagen.research.google/video/paper.pdf) paper). use_karras_sigmas (`bool`, *optional*, defaults to `False`): Whether to use Karras sigmas for step sizes in the noise schedule during the sampling process. If `True`, the sigmas are determined according to a sequence of noise levels {σi}. noise_sampler_seed (`int`, *optional*, defaults to `None`): The random seed to use for the noise sampler. If `None`, a random seed is generated. timestep_spacing (`str`, defaults to `"linspace"`): The way the timesteps should be scaled. Refer to Table 2 of the [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://huggingface.co/papers/2305.08891) for more information. steps_offset (`int`, defaults to 0): An offset added to the inference steps. You can use a combination of `offset=1` and `set_alpha_to_one=False` to make the last step use step 0 for the previous alpha product like in Stable Diffusion. """ _compatibles = [e.name for e in KarrasDiffusionSchedulers] order = 2 @register_to_config def __init__( self, num_train_timesteps: int = 1000, beta_start: float = 0.00085, # sensible defaults beta_end: float = 0.012, beta_schedule: str = "linear", trained_betas: Optional[Union[np.ndarray, List[float]]] = None, prediction_type: str = "epsilon", use_karras_sigmas: Optional[bool] = False, noise_sampler_seed: Optional[int] = None, timestep_spacing: str = "linspace", steps_offset: int = 0, ): if trained_betas is not None: self.betas = torch.tensor(trained_betas, dtype=torch.float32) elif beta_schedule == "linear": self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32) elif beta_schedule == "scaled_linear": # this schedule is very specific to the latent diffusion model. self.betas = torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2 elif beta_schedule == "squaredcos_cap_v2": # Glide cosine schedule self.betas = betas_for_alpha_bar(num_train_timesteps) else: raise NotImplementedError(f"{beta_schedule} does is not implemented for {self.__class__}") self.alphas = 1.0 - self.betas self.alphas_cumprod = torch.cumprod(self.alphas, dim=0) # set all values self.set_timesteps(num_train_timesteps, None, num_train_timesteps) self.use_karras_sigmas = use_karras_sigmas self.noise_sampler = None self.noise_sampler_seed = noise_sampler_seed self._step_index = None self._begin_index = None self.sigmas = self.sigmas.to("cpu") # to avoid too much CPU/GPU communication # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler.index_for_timestep def index_for_timestep(self, timestep, schedule_timesteps=None): if schedule_timesteps is None: schedule_timesteps = self.timesteps indices = (schedule_timesteps == timestep).nonzero() # The sigma index that is taken for the **very** first `step` # is always the second index (or the last index if there is only 1) # This way we can ensure we don't accidentally skip a sigma in # case we start in the middle of the denoising schedule (e.g. for image-to-image) pos = 1 if len(indices) > 1 else 0 return indices[pos].item() # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._init_step_index def _init_step_index(self, timestep): if self.begin_index is None: if isinstance(timestep, torch.Tensor): timestep = timestep.to(self.timesteps.device) self._step_index = self.index_for_timestep(timestep) else: self._step_index = self._begin_index @property def init_noise_sigma(self): # standard deviation of the initial noise distribution if self.config.timestep_spacing in ["linspace", "trailing"]: return self.sigmas.max() return (self.sigmas.max() ** 2 + 1) ** 0.5 @property def step_index(self): """ The index counter for current timestep. It will increae 1 after each scheduler step. """ return self._step_index @property def begin_index(self): """ The index for the first timestep. It should be set from pipeline with `set_begin_index` method. """ return self._begin_index # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler.set_begin_index def set_begin_index(self, begin_index: int = 0): """ Sets the begin index for the scheduler. This function should be run from pipeline before the inference. Args: begin_index (`int`): The begin index for the scheduler. """ self._begin_index = begin_index def scale_model_input( self, sample: torch.FloatTensor, timestep: Union[float, torch.FloatTensor], ) -> torch.FloatTensor: """ Ensures interchangeability with schedulers that need to scale the denoising model input depending on the current timestep. Args: sample (`torch.FloatTensor`): The input sample. timestep (`int`, *optional*): The current timestep in the diffusion chain. Returns: `torch.FloatTensor`: A scaled input sample. """ if self.step_index is None: self._init_step_index(timestep) sigma = self.sigmas[self.step_index] sigma_input = sigma if self.state_in_first_order else self.mid_point_sigma sample = sample / ((sigma_input**2 + 1) ** 0.5) return sample def set_timesteps( self, num_inference_steps: int, device: Union[str, torch.device] = None, num_train_timesteps: Optional[int] = None, ): """ Sets the discrete timesteps used for the diffusion chain (to be run before inference). Args: num_inference_steps (`int`): The number of diffusion steps used when generating samples with a pre-trained model. device (`str` or `torch.device`, *optional*): The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. """ self.num_inference_steps = num_inference_steps num_train_timesteps = num_train_timesteps or self.config.num_train_timesteps # "linspace", "leading", "trailing" corresponds to annotation of Table 2. of https://arxiv.org/abs/2305.08891 if self.config.timestep_spacing == "linspace": timesteps = np.linspace(0, num_train_timesteps - 1, num_inference_steps, dtype=float)[::-1].copy() elif self.config.timestep_spacing == "leading": step_ratio = num_train_timesteps // self.num_inference_steps # creates integer timesteps by multiplying by ratio # casting to int to avoid issues when num_inference_step is power of 3 timesteps = (np.arange(0, num_inference_steps) * step_ratio).round()[::-1].copy().astype(float) timesteps += self.config.steps_offset elif self.config.timestep_spacing == "trailing": step_ratio = num_train_timesteps / self.num_inference_steps # creates integer timesteps by multiplying by ratio # casting to int to avoid issues when num_inference_step is power of 3 timesteps = (np.arange(num_train_timesteps, 0, -step_ratio)).round().copy().astype(float) timesteps -= 1 else: raise ValueError( f"{self.config.timestep_spacing} is not supported. Please make sure to choose one of 'linspace', 'leading' or 'trailing'." ) sigmas = np.array(((1 - self.alphas_cumprod) / self.alphas_cumprod) ** 0.5) log_sigmas = np.log(sigmas) sigmas = np.interp(timesteps, np.arange(0, len(sigmas)), sigmas) if self.use_karras_sigmas: sigmas = self._convert_to_karras(in_sigmas=sigmas) timesteps = np.array([self._sigma_to_t(sigma, log_sigmas) for sigma in sigmas]) second_order_timesteps = self._second_order_timesteps(sigmas, log_sigmas) sigmas = np.concatenate([sigmas, [0.0]]).astype(np.float32) sigmas = torch.from_numpy(sigmas).to(device=device) self.sigmas = torch.cat([sigmas[:1], sigmas[1:-1].repeat_interleave(2), sigmas[-1:]]) timesteps = torch.from_numpy(timesteps) second_order_timesteps = torch.from_numpy(second_order_timesteps) timesteps = torch.cat([timesteps[:1], timesteps[1:].repeat_interleave(2)]) timesteps[1::2] = second_order_timesteps if str(device).startswith("mps"): # mps does not support float64 self.timesteps = timesteps.to(device, dtype=torch.float32) else: self.timesteps = timesteps.to(device=device) # empty first order variables self.sample = None self.mid_point_sigma = None self._step_index = None self._begin_index = None self.sigmas = self.sigmas.to("cpu") # to avoid too much CPU/GPU communication self.noise_sampler = None def _second_order_timesteps(self, sigmas, log_sigmas): def sigma_fn(_t): return np.exp(-_t) def t_fn(_sigma): return -np.log(_sigma) midpoint_ratio = 0.5 t = t_fn(sigmas) delta_time = np.diff(t) t_proposed = t[:-1] + delta_time * midpoint_ratio sig_proposed = sigma_fn(t_proposed) timesteps = np.array([self._sigma_to_t(sigma, log_sigmas) for sigma in sig_proposed]) return timesteps # copied from diffusers.schedulers.scheduling_euler_discrete._sigma_to_t def _sigma_to_t(self, sigma, log_sigmas): # get log sigma log_sigma = np.log(np.maximum(sigma, 1e-10)) # get distribution dists = log_sigma - log_sigmas[:, np.newaxis] # get sigmas range low_idx = np.cumsum((dists >= 0), axis=0).argmax(axis=0).clip(max=log_sigmas.shape[0] - 2) high_idx = low_idx + 1 low = log_sigmas[low_idx] high = log_sigmas[high_idx] # interpolate sigmas w = (low - log_sigma) / (low - high) w = np.clip(w, 0, 1) # transform interpolation to time range t = (1 - w) * low_idx + w * high_idx t = t.reshape(sigma.shape) return t # copied from diffusers.schedulers.scheduling_euler_discrete._convert_to_karras def _convert_to_karras(self, in_sigmas: torch.FloatTensor) -> torch.FloatTensor: """Constructs the noise schedule of Karras et al. (2022).""" sigma_min: float = in_sigmas[-1].item() sigma_max: float = in_sigmas[0].item() rho = 7.0 # 7.0 is the value used in the paper ramp = np.linspace(0, 1, self.num_inference_steps) min_inv_rho = sigma_min ** (1 / rho) max_inv_rho = sigma_max ** (1 / rho) sigmas = (max_inv_rho + ramp * (min_inv_rho - max_inv_rho)) ** rho return sigmas @property def state_in_first_order(self): return self.sample is None def step( self, model_output: Union[torch.FloatTensor, np.ndarray], timestep: Union[float, torch.FloatTensor], sample: Union[torch.FloatTensor, np.ndarray], return_dict: bool = True, s_noise: float = 1.0, ) -> Union[SchedulerOutput, Tuple]: """ Predict the sample from the previous timestep by reversing the SDE. This function propagates the diffusion process from the learned model outputs (most often the predicted noise). Args: model_output (`torch.FloatTensor` or `np.ndarray`): The direct output from learned diffusion model. timestep (`float` or `torch.FloatTensor`): The current discrete timestep in the diffusion chain. sample (`torch.FloatTensor` or `np.ndarray`): A current instance of a sample created by the diffusion process. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`~schedulers.scheduling_utils.SchedulerOutput`] or tuple. s_noise (`float`, *optional*, defaults to 1.0): Scaling factor for noise added to the sample. Returns: [`~schedulers.scheduling_utils.SchedulerOutput`] or `tuple`: If return_dict is `True`, [`~schedulers.scheduling_utils.SchedulerOutput`] is returned, otherwise a tuple is returned where the first element is the sample tensor. """ if self.step_index is None: self._init_step_index(timestep) # Create a noise sampler if it hasn't been created yet if self.noise_sampler is None: min_sigma, max_sigma = self.sigmas[self.sigmas > 0].min(), self.sigmas.max() self.noise_sampler = BrownianTreeNoiseSampler(sample, min_sigma, max_sigma, self.noise_sampler_seed) # Define functions to compute sigma and t from each other def sigma_fn(_t: torch.FloatTensor) -> torch.FloatTensor: return _t.neg().exp() def t_fn(_sigma: torch.FloatTensor) -> torch.FloatTensor: return _sigma.log().neg() if self.state_in_first_order: sigma = self.sigmas[self.step_index] sigma_next = self.sigmas[self.step_index + 1] else: # 2nd order sigma = self.sigmas[self.step_index - 1] sigma_next = self.sigmas[self.step_index] # Set the midpoint and step size for the current step midpoint_ratio = 0.5 t, t_next = t_fn(sigma), t_fn(sigma_next) delta_time = t_next - t t_proposed = t + delta_time * midpoint_ratio # 1. compute predicted original sample (x_0) from sigma-scaled predicted noise if self.config.prediction_type == "epsilon": sigma_input = sigma if self.state_in_first_order else sigma_fn(t_proposed) pred_original_sample = sample - sigma_input * model_output elif self.config.prediction_type == "v_prediction": sigma_input = sigma if self.state_in_first_order else sigma_fn(t_proposed) pred_original_sample = model_output * (-sigma_input / (sigma_input**2 + 1) ** 0.5) + ( sample / (sigma_input**2 + 1) ) elif self.config.prediction_type == "sample": raise NotImplementedError("prediction_type not implemented yet: sample") else: raise ValueError( f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, or `v_prediction`" ) if sigma_next == 0: derivative = (sample - pred_original_sample) / sigma dt = sigma_next - sigma prev_sample = sample + derivative * dt else: if self.state_in_first_order: t_next = t_proposed else: sample = self.sample sigma_from = sigma_fn(t) sigma_to = sigma_fn(t_next) sigma_up = min(sigma_to, (sigma_to**2 * (sigma_from**2 - sigma_to**2) / sigma_from**2) ** 0.5) sigma_down = (sigma_to**2 - sigma_up**2) ** 0.5 ancestral_t = t_fn(sigma_down) prev_sample = (sigma_fn(ancestral_t) / sigma_fn(t)) * sample - ( t - ancestral_t ).expm1() * pred_original_sample prev_sample = prev_sample + self.noise_sampler(sigma_fn(t), sigma_fn(t_next)) * s_noise * sigma_up if self.state_in_first_order: # store for 2nd order step self.sample = sample self.mid_point_sigma = sigma_fn(t_next) else: # free for "first order mode" self.sample = None self.mid_point_sigma = None # upon completion increase step index by one self._step_index += 1 if not return_dict: return (prev_sample,) return SchedulerOutput(prev_sample=prev_sample) # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler.add_noise def add_noise( self, original_samples: torch.FloatTensor, noise: torch.FloatTensor, timesteps: torch.FloatTensor, ) -> torch.FloatTensor: # Make sure sigmas and timesteps have the same device and dtype as original_samples sigmas = self.sigmas.to(device=original_samples.device, dtype=original_samples.dtype) if original_samples.device.type == "mps" and torch.is_floating_point(timesteps): # mps does not support float64 schedule_timesteps = self.timesteps.to(original_samples.device, dtype=torch.float32) timesteps = timesteps.to(original_samples.device, dtype=torch.float32) else: schedule_timesteps = self.timesteps.to(original_samples.device) timesteps = timesteps.to(original_samples.device) # self.begin_index is None when scheduler is used for training, or pipeline does not implement set_begin_index if self.begin_index is None: step_indices = [self.index_for_timestep(t, schedule_timesteps) for t in timesteps] else: step_indices = [self.begin_index] * timesteps.shape[0] sigma = sigmas[step_indices].flatten() while len(sigma.shape) < len(original_samples.shape): sigma = sigma.unsqueeze(-1) noisy_samples = original_samples + noise * sigma return noisy_samples def __len__(self): return self.config.num_train_timesteps

diffusers/src/diffusers/schedulers/scheduling_dpmsolver_sde.py/0

# Copyright 2023 Shuchen Xue, etc. in University of Chinese Academy of Sciences Team and The HuggingFace Team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # DISCLAIMER: check https://arxiv.org/abs/2309.05019 # The codebase is modified based on https://github.com/huggingface/diffusers/blob/main/src/diffusers/schedulers/scheduling_dpmsolver_multistep.py import math from typing import Callable, List, Optional, Tuple, Union import numpy as np import torch from ..configuration_utils import ConfigMixin, register_to_config from ..utils import deprecate from ..utils.torch_utils import randn_tensor from .scheduling_utils import KarrasDiffusionSchedulers, SchedulerMixin, SchedulerOutput # Copied from diffusers.schedulers.scheduling_ddpm.betas_for_alpha_bar def betas_for_alpha_bar( num_diffusion_timesteps, max_beta=0.999, alpha_transform_type="cosine", ): """ Create a beta schedule that discretizes the given alpha_t_bar function, which defines the cumulative product of (1-beta) over time from t = [0,1]. Contains a function alpha_bar that takes an argument t and transforms it to the cumulative product of (1-beta) up to that part of the diffusion process. Args: num_diffusion_timesteps (`int`): the number of betas to produce. max_beta (`float`): the maximum beta to use; use values lower than 1 to prevent singularities. alpha_transform_type (`str`, *optional*, default to `cosine`): the type of noise schedule for alpha_bar. Choose from `cosine` or `exp` Returns: betas (`np.ndarray`): the betas used by the scheduler to step the model outputs """ if alpha_transform_type == "cosine": def alpha_bar_fn(t): return math.cos((t + 0.008) / 1.008 * math.pi / 2) ** 2 elif alpha_transform_type == "exp": def alpha_bar_fn(t): return math.exp(t * -12.0) else: raise ValueError(f"Unsupported alpha_tranform_type: {alpha_transform_type}") betas = [] for i in range(num_diffusion_timesteps): t1 = i / num_diffusion_timesteps t2 = (i + 1) / num_diffusion_timesteps betas.append(min(1 - alpha_bar_fn(t2) / alpha_bar_fn(t1), max_beta)) return torch.tensor(betas, dtype=torch.float32) class SASolverScheduler(SchedulerMixin, ConfigMixin): """ `SASolverScheduler` is a fast dedicated high-order solver for diffusion SDEs. This model inherits from [`SchedulerMixin`] and [`ConfigMixin`]. Check the superclass documentation for the generic methods the library implements for all schedulers such as loading and saving. Args: num_train_timesteps (`int`, defaults to 1000): The number of diffusion steps to train the model. beta_start (`float`, defaults to 0.0001): The starting `beta` value of inference. beta_end (`float`, defaults to 0.02): The final `beta` value. beta_schedule (`str`, defaults to `"linear"`): The beta schedule, a mapping from a beta range to a sequence of betas for stepping the model. Choose from `linear`, `scaled_linear`, or `squaredcos_cap_v2`. trained_betas (`np.ndarray`, *optional*): Pass an array of betas directly to the constructor to bypass `beta_start` and `beta_end`. predictor_order (`int`, defaults to 2): The predictor order which can be `1` or `2` or `3` or '4'. It is recommended to use `predictor_order=2` for guided sampling, and `predictor_order=3` for unconditional sampling. corrector_order (`int`, defaults to 2): The corrector order which can be `1` or `2` or `3` or '4'. It is recommended to use `corrector_order=2` for guided sampling, and `corrector_order=3` for unconditional sampling. prediction_type (`str`, defaults to `epsilon`, *optional*): Prediction type of the scheduler function; can be `epsilon` (predicts the noise of the diffusion process), `sample` (directly predicts the noisy sample`) or `v_prediction` (see section 2.4 of [Imagen Video](https://imagen.research.google/video/paper.pdf) paper). tau_func (`Callable`, *optional*): Stochasticity during the sampling. Default in init is `lambda t: 1 if t >= 200 and t <= 800 else 0`. SA-Solver will sample from vanilla diffusion ODE if tau_func is set to `lambda t: 0`. SA-Solver will sample from vanilla diffusion SDE if tau_func is set to `lambda t: 1`. For more details, please check https://arxiv.org/abs/2309.05019 thresholding (`bool`, defaults to `False`): Whether to use the "dynamic thresholding" method. This is unsuitable for latent-space diffusion models such as Stable Diffusion. dynamic_thresholding_ratio (`float`, defaults to 0.995): The ratio for the dynamic thresholding method. Valid only when `thresholding=True`. sample_max_value (`float`, defaults to 1.0): The threshold value for dynamic thresholding. Valid only when `thresholding=True` and `algorithm_type="dpmsolver++"`. algorithm_type (`str`, defaults to `data_prediction`): Algorithm type for the solver; can be `data_prediction` or `noise_prediction`. It is recommended to use `data_prediction` with `solver_order=2` for guided sampling like in Stable Diffusion. lower_order_final (`bool`, defaults to `True`): Whether to use lower-order solvers in the final steps. Default = True. use_karras_sigmas (`bool`, *optional*, defaults to `False`): Whether to use Karras sigmas for step sizes in the noise schedule during the sampling process. If `True`, the sigmas are determined according to a sequence of noise levels {σi}. lambda_min_clipped (`float`, defaults to `-inf`): Clipping threshold for the minimum value of `lambda(t)` for numerical stability. This is critical for the cosine (`squaredcos_cap_v2`) noise schedule. variance_type (`str`, *optional*): Set to "learned" or "learned_range" for diffusion models that predict variance. If set, the model's output contains the predicted Gaussian variance. timestep_spacing (`str`, defaults to `"linspace"`): The way the timesteps should be scaled. Refer to Table 2 of the [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://huggingface.co/papers/2305.08891) for more information. steps_offset (`int`, defaults to 0): An offset added to the inference steps. You can use a combination of `offset=1` and `set_alpha_to_one=False` to make the last step use step 0 for the previous alpha product like in Stable Diffusion. """ _compatibles = [e.name for e in KarrasDiffusionSchedulers] order = 1 @register_to_config def __init__( self, num_train_timesteps: int = 1000, beta_start: float = 0.0001, beta_end: float = 0.02, beta_schedule: str = "linear", trained_betas: Optional[Union[np.ndarray, List[float]]] = None, predictor_order: int = 2, corrector_order: int = 2, prediction_type: str = "epsilon", tau_func: Optional[Callable] = None, thresholding: bool = False, dynamic_thresholding_ratio: float = 0.995, sample_max_value: float = 1.0, algorithm_type: str = "data_prediction", lower_order_final: bool = True, use_karras_sigmas: Optional[bool] = False, lambda_min_clipped: float = -float("inf"), variance_type: Optional[str] = None, timestep_spacing: str = "linspace", steps_offset: int = 0, ): if trained_betas is not None: self.betas = torch.tensor(trained_betas, dtype=torch.float32) elif beta_schedule == "linear": self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32) elif beta_schedule == "scaled_linear": # this schedule is very specific to the latent diffusion model. self.betas = ( torch.linspace( beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32, ) ** 2 ) elif beta_schedule == "squaredcos_cap_v2": # Glide cosine schedule self.betas = betas_for_alpha_bar(num_train_timesteps) else: raise NotImplementedError(f"{beta_schedule} does is not implemented for {self.__class__}") self.alphas = 1.0 - self.betas self.alphas_cumprod = torch.cumprod(self.alphas, dim=0) # Currently we only support VP-type noise schedule self.alpha_t = torch.sqrt(self.alphas_cumprod) self.sigma_t = torch.sqrt(1 - self.alphas_cumprod) self.lambda_t = torch.log(self.alpha_t) - torch.log(self.sigma_t) self.sigmas = ((1 - self.alphas_cumprod) / self.alphas_cumprod) ** 0.5 # standard deviation of the initial noise distribution self.init_noise_sigma = 1.0 if algorithm_type not in ["data_prediction", "noise_prediction"]: raise NotImplementedError(f"{algorithm_type} does is not implemented for {self.__class__}") # setable values self.num_inference_steps = None timesteps = np.linspace(0, num_train_timesteps - 1, num_train_timesteps, dtype=np.float32)[::-1].copy() self.timesteps = torch.from_numpy(timesteps) self.timestep_list = [None] * max(predictor_order, corrector_order - 1) self.model_outputs = [None] * max(predictor_order, corrector_order - 1) if tau_func is None: self.tau_func = lambda t: 1 if t >= 200 and t <= 800 else 0 else: self.tau_func = tau_func self.predict_x0 = algorithm_type == "data_prediction" self.lower_order_nums = 0 self.last_sample = None self._step_index = None self._begin_index = None self.sigmas = self.sigmas.to("cpu") # to avoid too much CPU/GPU communication @property def step_index(self): """ The index counter for current timestep. It will increae 1 after each scheduler step. """ return self._step_index @property def begin_index(self): """ The index for the first timestep. It should be set from pipeline with `set_begin_index` method. """ return self._begin_index # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler.set_begin_index def set_begin_index(self, begin_index: int = 0): """ Sets the begin index for the scheduler. This function should be run from pipeline before the inference. Args: begin_index (`int`): The begin index for the scheduler. """ self._begin_index = begin_index def set_timesteps(self, num_inference_steps: int = None, device: Union[str, torch.device] = None): """ Sets the discrete timesteps used for the diffusion chain (to be run before inference). Args: num_inference_steps (`int`): The number of diffusion steps used when generating samples with a pre-trained model. device (`str` or `torch.device`, *optional*): The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. """ # Clipping the minimum of all lambda(t) for numerical stability. # This is critical for cosine (squaredcos_cap_v2) noise schedule. clipped_idx = torch.searchsorted(torch.flip(self.lambda_t, [0]), self.config.lambda_min_clipped) last_timestep = ((self.config.num_train_timesteps - clipped_idx).numpy()).item() # "linspace", "leading", "trailing" corresponds to annotation of Table 2. of https://arxiv.org/abs/2305.08891 if self.config.timestep_spacing == "linspace": timesteps = ( np.linspace(0, last_timestep - 1, num_inference_steps + 1).round()[::-1][:-1].copy().astype(np.int64) ) elif self.config.timestep_spacing == "leading": step_ratio = last_timestep // (num_inference_steps + 1) # creates integer timesteps by multiplying by ratio # casting to int to avoid issues when num_inference_step is power of 3 timesteps = (np.arange(0, num_inference_steps + 1) * step_ratio).round()[::-1][:-1].copy().astype(np.int64) timesteps += self.config.steps_offset elif self.config.timestep_spacing == "trailing": step_ratio = self.config.num_train_timesteps / num_inference_steps # creates integer timesteps by multiplying by ratio # casting to int to avoid issues when num_inference_step is power of 3 timesteps = np.arange(last_timestep, 0, -step_ratio).round().copy().astype(np.int64) timesteps -= 1 else: raise ValueError( f"{self.config.timestep_spacing} is not supported. Please make sure to choose one of 'linspace', 'leading' or 'trailing'." ) sigmas = np.array(((1 - self.alphas_cumprod) / self.alphas_cumprod) ** 0.5) if self.config.use_karras_sigmas: log_sigmas = np.log(sigmas) sigmas = np.flip(sigmas).copy() sigmas = self._convert_to_karras(in_sigmas=sigmas, num_inference_steps=num_inference_steps) timesteps = np.array([self._sigma_to_t(sigma, log_sigmas) for sigma in sigmas]).round() sigmas = np.concatenate([sigmas, sigmas[-1:]]).astype(np.float32) else: sigmas = np.interp(timesteps, np.arange(0, len(sigmas)), sigmas) sigma_last = ((1 - self.alphas_cumprod[0]) / self.alphas_cumprod[0]) ** 0.5 sigmas = np.concatenate([sigmas, [sigma_last]]).astype(np.float32) self.sigmas = torch.from_numpy(sigmas) self.timesteps = torch.from_numpy(timesteps).to(device=device, dtype=torch.int64) self.num_inference_steps = len(timesteps) self.model_outputs = [ None, ] * max(self.config.predictor_order, self.config.corrector_order - 1) self.lower_order_nums = 0 self.last_sample = None # add an index counter for schedulers that allow duplicated timesteps self._step_index = None self._begin_index = None self.sigmas = self.sigmas.to("cpu") # to avoid too much CPU/GPU communication # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler._threshold_sample def _threshold_sample(self, sample: torch.FloatTensor) -> torch.FloatTensor: """ "Dynamic thresholding: At each sampling step we set s to a certain percentile absolute pixel value in xt0 (the prediction of x_0 at timestep t), and if s > 1, then we threshold xt0 to the range [-s, s] and then divide by s. Dynamic thresholding pushes saturated pixels (those near -1 and 1) inwards, thereby actively preventing pixels from saturation at each step. We find that dynamic thresholding results in significantly better photorealism as well as better image-text alignment, especially when using very large guidance weights." https://arxiv.org/abs/2205.11487 """ dtype = sample.dtype batch_size, channels, *remaining_dims = sample.shape if dtype not in (torch.float32, torch.float64): sample = sample.float() # upcast for quantile calculation, and clamp not implemented for cpu half # Flatten sample for doing quantile calculation along each image sample = sample.reshape(batch_size, channels * np.prod(remaining_dims)) abs_sample = sample.abs() # "a certain percentile absolute pixel value" s = torch.quantile(abs_sample, self.config.dynamic_thresholding_ratio, dim=1) s = torch.clamp( s, min=1, max=self.config.sample_max_value ) # When clamped to min=1, equivalent to standard clipping to [-1, 1] s = s.unsqueeze(1) # (batch_size, 1) because clamp will broadcast along dim=0 sample = torch.clamp(sample, -s, s) / s # "we threshold xt0 to the range [-s, s] and then divide by s" sample = sample.reshape(batch_size, channels, *remaining_dims) sample = sample.to(dtype) return sample # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._sigma_to_t def _sigma_to_t(self, sigma, log_sigmas): # get log sigma log_sigma = np.log(np.maximum(sigma, 1e-10)) # get distribution dists = log_sigma - log_sigmas[:, np.newaxis] # get sigmas range low_idx = np.cumsum((dists >= 0), axis=0).argmax(axis=0).clip(max=log_sigmas.shape[0] - 2) high_idx = low_idx + 1 low = log_sigmas[low_idx] high = log_sigmas[high_idx] # interpolate sigmas w = (low - log_sigma) / (low - high) w = np.clip(w, 0, 1) # transform interpolation to time range t = (1 - w) * low_idx + w * high_idx t = t.reshape(sigma.shape) return t # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler._sigma_to_alpha_sigma_t def _sigma_to_alpha_sigma_t(self, sigma): alpha_t = 1 / ((sigma**2 + 1) ** 0.5) sigma_t = sigma * alpha_t return alpha_t, sigma_t # Copied from diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler._convert_to_karras def _convert_to_karras(self, in_sigmas: torch.FloatTensor, num_inference_steps) -> torch.FloatTensor: """Constructs the noise schedule of Karras et al. (2022).""" # Hack to make sure that other schedulers which copy this function don't break # TODO: Add this logic to the other schedulers if hasattr(self.config, "sigma_min"): sigma_min = self.config.sigma_min else: sigma_min = None if hasattr(self.config, "sigma_max"): sigma_max = self.config.sigma_max else: sigma_max = None sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item() sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item() rho = 7.0 # 7.0 is the value used in the paper ramp = np.linspace(0, 1, num_inference_steps) min_inv_rho = sigma_min ** (1 / rho) max_inv_rho = sigma_max ** (1 / rho) sigmas = (max_inv_rho + ramp * (min_inv_rho - max_inv_rho)) ** rho return sigmas def convert_model_output( self, model_output: torch.FloatTensor, *args, sample: torch.FloatTensor = None, **kwargs, ) -> torch.FloatTensor: """ Convert the model output to the corresponding type the data_prediction/noise_prediction algorithm needs. Noise_prediction is designed to discretize an integral of the noise prediction model, and data_prediction is designed to discretize an integral of the data prediction model. <Tip> The algorithm and model type are decoupled. You can use either data_prediction or noise_prediction for both noise prediction and data prediction models. </Tip> Args: model_output (`torch.FloatTensor`): The direct output from the learned diffusion model. sample (`torch.FloatTensor`): A current instance of a sample created by the diffusion process. Returns: `torch.FloatTensor`: The converted model output. """ timestep = args[0] if len(args) > 0 else kwargs.pop("timestep", None) if sample is None: if len(args) > 1: sample = args[1] else: raise ValueError("missing `sample` as a required keyward argument") if timestep is not None: deprecate( "timesteps", "1.0.0", "Passing `timesteps` is deprecated and has no effect as model output conversion is now handled via an internal counter `self.step_index`", ) sigma = self.sigmas[self.step_index] alpha_t, sigma_t = self._sigma_to_alpha_sigma_t(sigma) # SA-Solver_data_prediction needs to solve an integral of the data prediction model. if self.config.algorithm_type in ["data_prediction"]: if self.config.prediction_type == "epsilon": # SA-Solver only needs the "mean" output. if self.config.variance_type in ["learned", "learned_range"]: model_output = model_output[:, :3] x0_pred = (sample - sigma_t * model_output) / alpha_t elif self.config.prediction_type == "sample": x0_pred = model_output elif self.config.prediction_type == "v_prediction": x0_pred = alpha_t * sample - sigma_t * model_output else: raise ValueError( f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, `sample`, or" " `v_prediction` for the SASolverScheduler." ) if self.config.thresholding: x0_pred = self._threshold_sample(x0_pred) return x0_pred # SA-Solver_noise_prediction needs to solve an integral of the noise prediction model. elif self.config.algorithm_type in ["noise_prediction"]: if self.config.prediction_type == "epsilon": # SA-Solver only needs the "mean" output. if self.config.variance_type in ["learned", "learned_range"]: epsilon = model_output[:, :3] else: epsilon = model_output elif self.config.prediction_type == "sample": epsilon = (sample - alpha_t * model_output) / sigma_t elif self.config.prediction_type == "v_prediction": epsilon = alpha_t * model_output + sigma_t * sample else: raise ValueError( f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, `sample`, or" " `v_prediction` for the SASolverScheduler." ) if self.config.thresholding: alpha_t, sigma_t = self.alpha_t[timestep], self.sigma_t[timestep] x0_pred = (sample - sigma_t * epsilon) / alpha_t x0_pred = self._threshold_sample(x0_pred) epsilon = (sample - alpha_t * x0_pred) / sigma_t return epsilon def get_coefficients_exponential_negative(self, order, interval_start, interval_end): """ Calculate the integral of exp(-x) * x^order dx from interval_start to interval_end """ assert order in [0, 1, 2, 3], "order is only supported for 0, 1, 2 and 3" if order == 0: return torch.exp(-interval_end) * (torch.exp(interval_end - interval_start) - 1) elif order == 1: return torch.exp(-interval_end) * ( (interval_start + 1) * torch.exp(interval_end - interval_start) - (interval_end + 1) ) elif order == 2: return torch.exp(-interval_end) * ( (interval_start**2 + 2 * interval_start + 2) * torch.exp(interval_end - interval_start) - (interval_end**2 + 2 * interval_end + 2) ) elif order == 3: return torch.exp(-interval_end) * ( (interval_start**3 + 3 * interval_start**2 + 6 * interval_start + 6) * torch.exp(interval_end - interval_start) - (interval_end**3 + 3 * interval_end**2 + 6 * interval_end + 6) ) def get_coefficients_exponential_positive(self, order, interval_start, interval_end, tau): """ Calculate the integral of exp(x(1+tau^2)) * x^order dx from interval_start to interval_end """ assert order in [0, 1, 2, 3], "order is only supported for 0, 1, 2 and 3" # after change of variable(cov) interval_end_cov = (1 + tau**2) * interval_end interval_start_cov = (1 + tau**2) * interval_start if order == 0: return ( torch.exp(interval_end_cov) * (1 - torch.exp(-(interval_end_cov - interval_start_cov))) / (1 + tau**2) ) elif order == 1: return ( torch.exp(interval_end_cov) * ( (interval_end_cov - 1) - (interval_start_cov - 1) * torch.exp(-(interval_end_cov - interval_start_cov)) ) / ((1 + tau**2) ** 2) ) elif order == 2: return ( torch.exp(interval_end_cov) * ( (interval_end_cov**2 - 2 * interval_end_cov + 2) - (interval_start_cov**2 - 2 * interval_start_cov + 2) * torch.exp(-(interval_end_cov - interval_start_cov)) ) / ((1 + tau**2) ** 3) ) elif order == 3: return ( torch.exp(interval_end_cov) * ( (interval_end_cov**3 - 3 * interval_end_cov**2 + 6 * interval_end_cov - 6) - (interval_start_cov**3 - 3 * interval_start_cov**2 + 6 * interval_start_cov - 6) * torch.exp(-(interval_end_cov - interval_start_cov)) ) / ((1 + tau**2) ** 4) ) def lagrange_polynomial_coefficient(self, order, lambda_list): """ Calculate the coefficient of lagrange polynomial """ assert order in [0, 1, 2, 3] assert order == len(lambda_list) - 1 if order == 0: return [[1]] elif order == 1: return [ [ 1 / (lambda_list[0] - lambda_list[1]), -lambda_list[1] / (lambda_list[0] - lambda_list[1]), ], [ 1 / (lambda_list[1] - lambda_list[0]), -lambda_list[0] / (lambda_list[1] - lambda_list[0]), ], ] elif order == 2: denominator1 = (lambda_list[0] - lambda_list[1]) * (lambda_list[0] - lambda_list[2]) denominator2 = (lambda_list[1] - lambda_list[0]) * (lambda_list[1] - lambda_list[2]) denominator3 = (lambda_list[2] - lambda_list[0]) * (lambda_list[2] - lambda_list[1]) return [ [ 1 / denominator1, (-lambda_list[1] - lambda_list[2]) / denominator1, lambda_list[1] * lambda_list[2] / denominator1, ], [ 1 / denominator2, (-lambda_list[0] - lambda_list[2]) / denominator2, lambda_list[0] * lambda_list[2] / denominator2, ], [ 1 / denominator3, (-lambda_list[0] - lambda_list[1]) / denominator3, lambda_list[0] * lambda_list[1] / denominator3, ], ] elif order == 3: denominator1 = ( (lambda_list[0] - lambda_list[1]) * (lambda_list[0] - lambda_list[2]) * (lambda_list[0] - lambda_list[3]) ) denominator2 = ( (lambda_list[1] - lambda_list[0]) * (lambda_list[1] - lambda_list[2]) * (lambda_list[1] - lambda_list[3]) ) denominator3 = ( (lambda_list[2] - lambda_list[0]) * (lambda_list[2] - lambda_list[1]) * (lambda_list[2] - lambda_list[3]) ) denominator4 = ( (lambda_list[3] - lambda_list[0]) * (lambda_list[3] - lambda_list[1]) * (lambda_list[3] - lambda_list[2]) ) return [ [ 1 / denominator1, (-lambda_list[1] - lambda_list[2] - lambda_list[3]) / denominator1, ( lambda_list[1] * lambda_list[2] + lambda_list[1] * lambda_list[3] + lambda_list[2] * lambda_list[3] ) / denominator1, (-lambda_list[1] * lambda_list[2] * lambda_list[3]) / denominator1, ], [ 1 / denominator2, (-lambda_list[0] - lambda_list[2] - lambda_list[3]) / denominator2, ( lambda_list[0] * lambda_list[2] + lambda_list[0] * lambda_list[3] + lambda_list[2] * lambda_list[3] ) / denominator2, (-lambda_list[0] * lambda_list[2] * lambda_list[3]) / denominator2, ], [ 1 / denominator3, (-lambda_list[0] - lambda_list[1] - lambda_list[3]) / denominator3, ( lambda_list[0] * lambda_list[1] + lambda_list[0] * lambda_list[3] + lambda_list[1] * lambda_list[3] ) / denominator3, (-lambda_list[0] * lambda_list[1] * lambda_list[3]) / denominator3, ], [ 1 / denominator4, (-lambda_list[0] - lambda_list[1] - lambda_list[2]) / denominator4, ( lambda_list[0] * lambda_list[1] + lambda_list[0] * lambda_list[2] + lambda_list[1] * lambda_list[2] ) / denominator4, (-lambda_list[0] * lambda_list[1] * lambda_list[2]) / denominator4, ], ] def get_coefficients_fn(self, order, interval_start, interval_end, lambda_list, tau): assert order in [1, 2, 3, 4] assert order == len(lambda_list), "the length of lambda list must be equal to the order" coefficients = [] lagrange_coefficient = self.lagrange_polynomial_coefficient(order - 1, lambda_list) for i in range(order): coefficient = 0 for j in range(order): if self.predict_x0: coefficient += lagrange_coefficient[i][j] * self.get_coefficients_exponential_positive( order - 1 - j, interval_start, interval_end, tau ) else: coefficient += lagrange_coefficient[i][j] * self.get_coefficients_exponential_negative( order - 1 - j, interval_start, interval_end ) coefficients.append(coefficient) assert len(coefficients) == order, "the length of coefficients does not match the order" return coefficients def stochastic_adams_bashforth_update( self, model_output: torch.FloatTensor, *args, sample: torch.FloatTensor, noise: torch.FloatTensor, order: int, tau: torch.FloatTensor, **kwargs, ) -> torch.FloatTensor: """ One step for the SA-Predictor. Args: model_output (`torch.FloatTensor`): The direct output from the learned diffusion model at the current timestep. prev_timestep (`int`): The previous discrete timestep in the diffusion chain. sample (`torch.FloatTensor`): A current instance of a sample created by the diffusion process. order (`int`): The order of SA-Predictor at this timestep. Returns: `torch.FloatTensor`: The sample tensor at the previous timestep. """ prev_timestep = args[0] if len(args) > 0 else kwargs.pop("prev_timestep", None) if sample is None: if len(args) > 1: sample = args[1] else: raise ValueError(" missing `sample` as a required keyward argument") if noise is None: if len(args) > 2: noise = args[2] else: raise ValueError(" missing `noise` as a required keyward argument") if order is None: if len(args) > 3: order = args[3] else: raise ValueError(" missing `order` as a required keyward argument") if tau is None: if len(args) > 4: tau = args[4] else: raise ValueError(" missing `tau` as a required keyward argument") if prev_timestep is not None: deprecate( "prev_timestep", "1.0.0", "Passing `prev_timestep` is deprecated and has no effect as model output conversion is now handled via an internal counter `self.step_index`", ) model_output_list = self.model_outputs sigma_t, sigma_s0 = ( self.sigmas[self.step_index + 1], self.sigmas[self.step_index], ) alpha_t, sigma_t = self._sigma_to_alpha_sigma_t(sigma_t) alpha_s0, sigma_s0 = self._sigma_to_alpha_sigma_t(sigma_s0) lambda_t = torch.log(alpha_t) - torch.log(sigma_t) lambda_s0 = torch.log(alpha_s0) - torch.log(sigma_s0) gradient_part = torch.zeros_like(sample) h = lambda_t - lambda_s0 lambda_list = [] for i in range(order): si = self.step_index - i alpha_si, sigma_si = self._sigma_to_alpha_sigma_t(self.sigmas[si]) lambda_si = torch.log(alpha_si) - torch.log(sigma_si) lambda_list.append(lambda_si) gradient_coefficients = self.get_coefficients_fn(order, lambda_s0, lambda_t, lambda_list, tau) x = sample if self.predict_x0: if ( order == 2 ): ## if order = 2 we do a modification that does not influence the convergence order similar to unipc. Note: This is used only for few steps sampling. # The added term is O(h^3). Empirically we find it will slightly improve the image quality. # ODE case # gradient_coefficients[0] += 1.0 * torch.exp(lambda_t) * (h ** 2 / 2 - (h - 1 + torch.exp(-h))) / (ns.marginal_lambda(t_prev_list[-1]) - ns.marginal_lambda(t_prev_list[-2])) # gradient_coefficients[1] -= 1.0 * torch.exp(lambda_t) * (h ** 2 / 2 - (h - 1 + torch.exp(-h))) / (ns.marginal_lambda(t_prev_list[-1]) - ns.marginal_lambda(t_prev_list[-2])) temp_sigma = self.sigmas[self.step_index - 1] temp_alpha_s, temp_sigma_s = self._sigma_to_alpha_sigma_t(temp_sigma) temp_lambda_s = torch.log(temp_alpha_s) - torch.log(temp_sigma_s) gradient_coefficients[0] += ( 1.0 * torch.exp((1 + tau**2) * lambda_t) * (h**2 / 2 - (h * (1 + tau**2) - 1 + torch.exp((1 + tau**2) * (-h))) / ((1 + tau**2) ** 2)) / (lambda_s0 - temp_lambda_s) ) gradient_coefficients[1] -= ( 1.0 * torch.exp((1 + tau**2) * lambda_t) * (h**2 / 2 - (h * (1 + tau**2) - 1 + torch.exp((1 + tau**2) * (-h))) / ((1 + tau**2) ** 2)) / (lambda_s0 - temp_lambda_s) ) for i in range(order): if self.predict_x0: gradient_part += ( (1 + tau**2) * sigma_t * torch.exp(-(tau**2) * lambda_t) * gradient_coefficients[i] * model_output_list[-(i + 1)] ) else: gradient_part += -(1 + tau**2) * alpha_t * gradient_coefficients[i] * model_output_list[-(i + 1)] if self.predict_x0: noise_part = sigma_t * torch.sqrt(1 - torch.exp(-2 * tau**2 * h)) * noise else: noise_part = tau * sigma_t * torch.sqrt(torch.exp(2 * h) - 1) * noise if self.predict_x0: x_t = torch.exp(-(tau**2) * h) * (sigma_t / sigma_s0) * x + gradient_part + noise_part else: x_t = (alpha_t / alpha_s0) * x + gradient_part + noise_part x_t = x_t.to(x.dtype) return x_t def stochastic_adams_moulton_update( self, this_model_output: torch.FloatTensor, *args, last_sample: torch.FloatTensor, last_noise: torch.FloatTensor, this_sample: torch.FloatTensor, order: int, tau: torch.FloatTensor, **kwargs, ) -> torch.FloatTensor: """ One step for the SA-Corrector. Args: this_model_output (`torch.FloatTensor`): The model outputs at `x_t`. this_timestep (`int`): The current timestep `t`. last_sample (`torch.FloatTensor`): The generated sample before the last predictor `x_{t-1}`. this_sample (`torch.FloatTensor`): The generated sample after the last predictor `x_{t}`. order (`int`): The order of SA-Corrector at this step. Returns: `torch.FloatTensor`: The corrected sample tensor at the current timestep. """ this_timestep = args[0] if len(args) > 0 else kwargs.pop("this_timestep", None) if last_sample is None: if len(args) > 1: last_sample = args[1] else: raise ValueError(" missing`last_sample` as a required keyward argument") if last_noise is None: if len(args) > 2: last_noise = args[2] else: raise ValueError(" missing`last_noise` as a required keyward argument") if this_sample is None: if len(args) > 3: this_sample = args[3] else: raise ValueError(" missing`this_sample` as a required keyward argument") if order is None: if len(args) > 4: order = args[4] else: raise ValueError(" missing`order` as a required keyward argument") if tau is None: if len(args) > 5: tau = args[5] else: raise ValueError(" missing`tau` as a required keyward argument") if this_timestep is not None: deprecate( "this_timestep", "1.0.0", "Passing `this_timestep` is deprecated and has no effect as model output conversion is now handled via an internal counter `self.step_index`", ) model_output_list = self.model_outputs sigma_t, sigma_s0 = ( self.sigmas[self.step_index], self.sigmas[self.step_index - 1], ) alpha_t, sigma_t = self._sigma_to_alpha_sigma_t(sigma_t) alpha_s0, sigma_s0 = self._sigma_to_alpha_sigma_t(sigma_s0) lambda_t = torch.log(alpha_t) - torch.log(sigma_t) lambda_s0 = torch.log(alpha_s0) - torch.log(sigma_s0) gradient_part = torch.zeros_like(this_sample) h = lambda_t - lambda_s0 lambda_list = [] for i in range(order): si = self.step_index - i alpha_si, sigma_si = self._sigma_to_alpha_sigma_t(self.sigmas[si]) lambda_si = torch.log(alpha_si) - torch.log(sigma_si) lambda_list.append(lambda_si) model_prev_list = model_output_list + [this_model_output] gradient_coefficients = self.get_coefficients_fn(order, lambda_s0, lambda_t, lambda_list, tau) x = last_sample if self.predict_x0: if ( order == 2 ): ## if order = 2 we do a modification that does not influence the convergence order similar to UniPC. Note: This is used only for few steps sampling. # The added term is O(h^3). Empirically we find it will slightly improve the image quality. # ODE case # gradient_coefficients[0] += 1.0 * torch.exp(lambda_t) * (h / 2 - (h - 1 + torch.exp(-h)) / h) # gradient_coefficients[1] -= 1.0 * torch.exp(lambda_t) * (h / 2 - (h - 1 + torch.exp(-h)) / h) gradient_coefficients[0] += ( 1.0 * torch.exp((1 + tau**2) * lambda_t) * (h / 2 - (h * (1 + tau**2) - 1 + torch.exp((1 + tau**2) * (-h))) / ((1 + tau**2) ** 2 * h)) ) gradient_coefficients[1] -= ( 1.0 * torch.exp((1 + tau**2) * lambda_t) * (h / 2 - (h * (1 + tau**2) - 1 + torch.exp((1 + tau**2) * (-h))) / ((1 + tau**2) ** 2 * h)) ) for i in range(order): if self.predict_x0: gradient_part += ( (1 + tau**2) * sigma_t * torch.exp(-(tau**2) * lambda_t) * gradient_coefficients[i] * model_prev_list[-(i + 1)] ) else: gradient_part += -(1 + tau**2) * alpha_t * gradient_coefficients[i] * model_prev_list[-(i + 1)] if self.predict_x0: noise_part = sigma_t * torch.sqrt(1 - torch.exp(-2 * tau**2 * h)) * last_noise else: noise_part = tau * sigma_t * torch.sqrt(torch.exp(2 * h) - 1) * last_noise if self.predict_x0: x_t = torch.exp(-(tau**2) * h) * (sigma_t / sigma_s0) * x + gradient_part + noise_part else: x_t = (alpha_t / alpha_s0) * x + gradient_part + noise_part x_t = x_t.to(x.dtype) return x_t # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler.index_for_timestep def index_for_timestep(self, timestep, schedule_timesteps=None): if schedule_timesteps is None: schedule_timesteps = self.timesteps index_candidates = (schedule_timesteps == timestep).nonzero() if len(index_candidates) == 0: step_index = len(self.timesteps) - 1 # The sigma index that is taken for the **very** first `step` # is always the second index (or the last index if there is only 1) # This way we can ensure we don't accidentally skip a sigma in # case we start in the middle of the denoising schedule (e.g. for image-to-image) elif len(index_candidates) > 1: step_index = index_candidates[1].item() else: step_index = index_candidates[0].item() return step_index # Copied from diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler._init_step_index def _init_step_index(self, timestep): """ Initialize the step_index counter for the scheduler. """ if self.begin_index is None: if isinstance(timestep, torch.Tensor): timestep = timestep.to(self.timesteps.device) self._step_index = self.index_for_timestep(timestep) else: self._step_index = self._begin_index def step( self, model_output: torch.FloatTensor, timestep: int, sample: torch.FloatTensor, generator=None, return_dict: bool = True, ) -> Union[SchedulerOutput, Tuple]: """ Predict the sample from the previous timestep by reversing the SDE. This function propagates the sample with the SA-Solver. Args: model_output (`torch.FloatTensor`): The direct output from learned diffusion model. timestep (`int`): The current discrete timestep in the diffusion chain. sample (`torch.FloatTensor`): A current instance of a sample created by the diffusion process. generator (`torch.Generator`, *optional*): A random number generator. return_dict (`bool`): Whether or not to return a [`~schedulers.scheduling_utils.SchedulerOutput`] or `tuple`. Returns: [`~schedulers.scheduling_utils.SchedulerOutput`] or `tuple`: If return_dict is `True`, [`~schedulers.scheduling_utils.SchedulerOutput`] is returned, otherwise a tuple is returned where the first element is the sample tensor. """ if self.num_inference_steps is None: raise ValueError( "Number of inference steps is 'None', you need to run 'set_timesteps' after creating the scheduler" ) if self.step_index is None: self._init_step_index(timestep) use_corrector = self.step_index > 0 and self.last_sample is not None model_output_convert = self.convert_model_output(model_output, sample=sample) if use_corrector: current_tau = self.tau_func(self.timestep_list[-1]) sample = self.stochastic_adams_moulton_update( this_model_output=model_output_convert, last_sample=self.last_sample, last_noise=self.last_noise, this_sample=sample, order=self.this_corrector_order, tau=current_tau, ) for i in range(max(self.config.predictor_order, self.config.corrector_order - 1) - 1): self.model_outputs[i] = self.model_outputs[i + 1] self.timestep_list[i] = self.timestep_list[i + 1] self.model_outputs[-1] = model_output_convert self.timestep_list[-1] = timestep noise = randn_tensor( model_output.shape, generator=generator, device=model_output.device, dtype=model_output.dtype, ) if self.config.lower_order_final: this_predictor_order = min(self.config.predictor_order, len(self.timesteps) - self.step_index) this_corrector_order = min(self.config.corrector_order, len(self.timesteps) - self.step_index + 1) else: this_predictor_order = self.config.predictor_order this_corrector_order = self.config.corrector_order self.this_predictor_order = min(this_predictor_order, self.lower_order_nums + 1) # warmup for multistep self.this_corrector_order = min(this_corrector_order, self.lower_order_nums + 2) # warmup for multistep assert self.this_predictor_order > 0 assert self.this_corrector_order > 0 self.last_sample = sample self.last_noise = noise current_tau = self.tau_func(self.timestep_list[-1]) prev_sample = self.stochastic_adams_bashforth_update( model_output=model_output_convert, sample=sample, noise=noise, order=self.this_predictor_order, tau=current_tau, ) if self.lower_order_nums < max(self.config.predictor_order, self.config.corrector_order - 1): self.lower_order_nums += 1 # upon completion increase step index by one self._step_index += 1 if not return_dict: return (prev_sample,) return SchedulerOutput(prev_sample=prev_sample) def scale_model_input(self, sample: torch.FloatTensor, *args, **kwargs) -> torch.FloatTensor: """ Ensures interchangeability with schedulers that need to scale the denoising model input depending on the current timestep. Args: sample (`torch.FloatTensor`): The input sample. Returns: `torch.FloatTensor`: A scaled input sample. """ return sample # Copied from diffusers.schedulers.scheduling_ddpm.DDPMScheduler.add_noise def add_noise( self, original_samples: torch.FloatTensor, noise: torch.FloatTensor, timesteps: torch.IntTensor, ) -> torch.FloatTensor: # Make sure alphas_cumprod and timestep have same device and dtype as original_samples # Move the self.alphas_cumprod to device to avoid redundant CPU to GPU data movement # for the subsequent add_noise calls self.alphas_cumprod = self.alphas_cumprod.to(device=original_samples.device) alphas_cumprod = self.alphas_cumprod.to(dtype=original_samples.dtype) timesteps = timesteps.to(original_samples.device) sqrt_alpha_prod = alphas_cumprod[timesteps] ** 0.5 sqrt_alpha_prod = sqrt_alpha_prod.flatten() while len(sqrt_alpha_prod.shape) < len(original_samples.shape): sqrt_alpha_prod = sqrt_alpha_prod.unsqueeze(-1) sqrt_one_minus_alpha_prod = (1 - alphas_cumprod[timesteps]) ** 0.5 sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.flatten() while len(sqrt_one_minus_alpha_prod.shape) < len(original_samples.shape): sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.unsqueeze(-1) noisy_samples = sqrt_alpha_prod * original_samples + sqrt_one_minus_alpha_prod * noise return noisy_samples def __len__(self): return self.config.num_train_timesteps

diffusers/src/diffusers/schedulers/scheduling_sasolver.py/0

# This file is autogenerated by the command `make fix-copies`, do not edit. from ..utils import DummyObject, requires_backends class MidiProcessor(metaclass=DummyObject): _backends = ["note_seq"] def __init__(self, *args, **kwargs): requires_backends(self, ["note_seq"]) @classmethod def from_config(cls, *args, **kwargs): requires_backends(cls, ["note_seq"]) @classmethod def from_pretrained(cls, *args, **kwargs): requires_backends(cls, ["note_seq"])

diffusers/src/diffusers/utils/dummy_note_seq_objects.py/0

# Copyright 2023 The HuggingFace Team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """ Generic utilities """ from collections import OrderedDict from dataclasses import fields, is_dataclass from typing import Any, Tuple import numpy as np from .import_utils import is_torch_available def is_tensor(x) -> bool: """ Tests if `x` is a `torch.Tensor` or `np.ndarray`. """ if is_torch_available(): import torch if isinstance(x, torch.Tensor): return True return isinstance(x, np.ndarray) class BaseOutput(OrderedDict): """ Base class for all model outputs as dataclass. Has a `__getitem__` that allows indexing by integer or slice (like a tuple) or strings (like a dictionary) that will ignore the `None` attributes. Otherwise behaves like a regular Python dictionary. <Tip warning={true}> You can't unpack a [`BaseOutput`] directly. Use the [`~utils.BaseOutput.to_tuple`] method to convert it to a tuple first. </Tip> """ def __init_subclass__(cls) -> None: """Register subclasses as pytree nodes. This is necessary to synchronize gradients when using `torch.nn.parallel.DistributedDataParallel` with `static_graph=True` with modules that output `ModelOutput` subclasses. """ if is_torch_available(): import torch.utils._pytree torch.utils._pytree._register_pytree_node( cls, torch.utils._pytree._dict_flatten, lambda values, context: cls(**torch.utils._pytree._dict_unflatten(values, context)), ) def __post_init__(self) -> None: class_fields = fields(self) # Safety and consistency checks if not len(class_fields): raise ValueError(f"{self.__class__.__name__} has no fields.") first_field = getattr(self, class_fields[0].name) other_fields_are_none = all(getattr(self, field.name) is None for field in class_fields[1:]) if other_fields_are_none and isinstance(first_field, dict): for key, value in first_field.items(): self[key] = value else: for field in class_fields: v = getattr(self, field.name) if v is not None: self[field.name] = v def __delitem__(self, *args, **kwargs): raise Exception(f"You cannot use ``__delitem__`` on a {self.__class__.__name__} instance.") def setdefault(self, *args, **kwargs): raise Exception(f"You cannot use ``setdefault`` on a {self.__class__.__name__} instance.") def pop(self, *args, **kwargs): raise Exception(f"You cannot use ``pop`` on a {self.__class__.__name__} instance.") def update(self, *args, **kwargs): raise Exception(f"You cannot use ``update`` on a {self.__class__.__name__} instance.") def __getitem__(self, k: Any) -> Any: if isinstance(k, str): inner_dict = dict(self.items()) return inner_dict[k] else: return self.to_tuple()[k] def __setattr__(self, name: Any, value: Any) -> None: if name in self.keys() and value is not None: # Don't call self.__setitem__ to avoid recursion errors super().__setitem__(name, value) super().__setattr__(name, value) def __setitem__(self, key, value): # Will raise a KeyException if needed super().__setitem__(key, value) # Don't call self.__setattr__ to avoid recursion errors super().__setattr__(key, value) def __reduce__(self): if not is_dataclass(self): return super().__reduce__() callable, _args, *remaining = super().__reduce__() args = tuple(getattr(self, field.name) for field in fields(self)) return callable, args, *remaining def to_tuple(self) -> Tuple[Any, ...]: """ Convert self to a tuple containing all the attributes/keys that are not `None`. """ return tuple(self[k] for k in self.keys())

diffusers/src/diffusers/utils/outputs.py/0

# coding=utf-8 # Copyright 2023 HuggingFace Inc. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import unittest import numpy as np import torch from diffusers.models import ModelMixin, UNet3DConditionModel from diffusers.utils import logging from diffusers.utils.import_utils import is_xformers_available from diffusers.utils.testing_utils import enable_full_determinism, floats_tensor, skip_mps, torch_device from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin enable_full_determinism() logger = logging.get_logger(__name__) @skip_mps class UNet3DConditionModelTests(ModelTesterMixin, UNetTesterMixin, unittest.TestCase): model_class = UNet3DConditionModel main_input_name = "sample" @property def dummy_input(self): batch_size = 4 num_channels = 4 num_frames = 4 sizes = (32, 32) noise = floats_tensor((batch_size, num_channels, num_frames) + sizes).to(torch_device) time_step = torch.tensor([10]).to(torch_device) encoder_hidden_states = floats_tensor((batch_size, 4, 32)).to(torch_device) return {"sample": noise, "timestep": time_step, "encoder_hidden_states": encoder_hidden_states} @property def input_shape(self): return (4, 4, 32, 32) @property def output_shape(self): return (4, 4, 32, 32) def prepare_init_args_and_inputs_for_common(self): init_dict = { "block_out_channels": (32, 64), "down_block_types": ( "CrossAttnDownBlock3D", "DownBlock3D", ), "up_block_types": ("UpBlock3D", "CrossAttnUpBlock3D"), "cross_attention_dim": 32, "attention_head_dim": 8, "out_channels": 4, "in_channels": 4, "layers_per_block": 1, "sample_size": 32, } inputs_dict = self.dummy_input return init_dict, inputs_dict @unittest.skipIf( torch_device != "cuda" or not is_xformers_available(), reason="XFormers attention is only available with CUDA and `xformers` installed", ) def test_xformers_enable_works(self): init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common() model = self.model_class(**init_dict) model.enable_xformers_memory_efficient_attention() assert ( model.mid_block.attentions[0].transformer_blocks[0].attn1.processor.__class__.__name__ == "XFormersAttnProcessor" ), "xformers is not enabled" # Overriding to set `norm_num_groups` needs to be different for this model. def test_forward_with_norm_groups(self): init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common() init_dict["norm_num_groups"] = 32 model = self.model_class(**init_dict) model.to(torch_device) model.eval() with torch.no_grad(): output = model(**inputs_dict) if isinstance(output, dict): output = output.sample self.assertIsNotNone(output) expected_shape = inputs_dict["sample"].shape self.assertEqual(output.shape, expected_shape, "Input and output shapes do not match") # Overriding since the UNet3D outputs a different structure. def test_determinism(self): init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common() model = self.model_class(**init_dict) model.to(torch_device) model.eval() with torch.no_grad(): # Warmup pass when using mps (see #372) if torch_device == "mps" and isinstance(model, ModelMixin): model(**self.dummy_input) first = model(**inputs_dict) if isinstance(first, dict): first = first.sample second = model(**inputs_dict) if isinstance(second, dict): second = second.sample out_1 = first.cpu().numpy() out_2 = second.cpu().numpy() out_1 = out_1[~np.isnan(out_1)] out_2 = out_2[~np.isnan(out_2)] max_diff = np.amax(np.abs(out_1 - out_2)) self.assertLessEqual(max_diff, 1e-5) def test_model_attention_slicing(self): init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common() init_dict["attention_head_dim"] = 8 model = self.model_class(**init_dict) model.to(torch_device) model.eval() model.set_attention_slice("auto") with torch.no_grad(): output = model(**inputs_dict) assert output is not None model.set_attention_slice("max") with torch.no_grad(): output = model(**inputs_dict) assert output is not None model.set_attention_slice(2) with torch.no_grad(): output = model(**inputs_dict) assert output is not None def test_feed_forward_chunking(self): init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common() init_dict["norm_num_groups"] = 32 model = self.model_class(**init_dict) model.to(torch_device) model.eval() with torch.no_grad(): output = model(**inputs_dict)[0] model.enable_forward_chunking() with torch.no_grad(): output_2 = model(**inputs_dict)[0] self.assertEqual(output.shape, output_2.shape, "Shape doesn't match") assert np.abs(output.cpu() - output_2.cpu()).max() < 1e-2

diffusers/tests/models/unets/test_models_unet_3d_condition.py/0

# coding=utf-8 # Copyright 2023 HuggingFace Inc. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import gc import tempfile import traceback import unittest import numpy as np import torch from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer from diffusers import ( AutoencoderKL, ControlNetModel, DDIMScheduler, EulerDiscreteScheduler, LCMScheduler, StableDiffusionControlNetPipeline, UNet2DConditionModel, ) from diffusers.pipelines.controlnet.pipeline_controlnet import MultiControlNetModel from diffusers.utils.import_utils import is_xformers_available from diffusers.utils.testing_utils import ( enable_full_determinism, load_image, load_numpy, numpy_cosine_similarity_distance, require_python39_or_higher, require_torch_2, require_torch_gpu, run_test_in_subprocess, slow, torch_device, ) from diffusers.utils.torch_utils import randn_tensor from ..pipeline_params import ( IMAGE_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_PARAMS, ) from ..test_pipelines_common import ( PipelineKarrasSchedulerTesterMixin, PipelineLatentTesterMixin, PipelineTesterMixin, ) enable_full_determinism() # Will be run via run_test_in_subprocess def _test_stable_diffusion_compile(in_queue, out_queue, timeout): error = None try: _ = in_queue.get(timeout=timeout) controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.to("cuda") pipe.set_progress_bar_config(disable=None) pipe.unet.to(memory_format=torch.channels_last) pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True) pipe.controlnet.to(memory_format=torch.channels_last) pipe.controlnet = torch.compile(pipe.controlnet, mode="reduce-overhead", fullgraph=True) generator = torch.Generator(device="cpu").manual_seed(0) prompt = "bird" image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/bird_canny.png" ).resize((512, 512)) output = pipe(prompt, image, num_inference_steps=10, generator=generator, output_type="np") image = output.images[0] assert image.shape == (512, 512, 3) expected_image = load_numpy( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/bird_canny_out_full.npy" ) expected_image = np.resize(expected_image, (512, 512, 3)) assert np.abs(expected_image - image).max() < 1.0 except Exception: error = f"{traceback.format_exc()}" results = {"error": error} out_queue.put(results, timeout=timeout) out_queue.join() class ControlNetPipelineFastTests( PipelineLatentTesterMixin, PipelineKarrasSchedulerTesterMixin, PipelineTesterMixin, unittest.TestCase ): pipeline_class = StableDiffusionControlNetPipeline params = TEXT_TO_IMAGE_PARAMS batch_params = TEXT_TO_IMAGE_BATCH_PARAMS image_params = IMAGE_TO_IMAGE_IMAGE_PARAMS image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS def get_dummy_components(self, time_cond_proj_dim=None): torch.manual_seed(0) unet = UNet2DConditionModel( block_out_channels=(4, 8), layers_per_block=2, sample_size=32, in_channels=4, out_channels=4, down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"), up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"), cross_attention_dim=32, norm_num_groups=1, time_cond_proj_dim=time_cond_proj_dim, ) torch.manual_seed(0) controlnet = ControlNetModel( block_out_channels=(4, 8), layers_per_block=2, in_channels=4, down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"), cross_attention_dim=32, conditioning_embedding_out_channels=(16, 32), norm_num_groups=1, ) torch.manual_seed(0) scheduler = DDIMScheduler( beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", clip_sample=False, set_alpha_to_one=False, ) torch.manual_seed(0) vae = AutoencoderKL( block_out_channels=[4, 8], in_channels=3, out_channels=3, down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"], up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"], latent_channels=4, norm_num_groups=2, ) torch.manual_seed(0) text_encoder_config = CLIPTextConfig( bos_token_id=0, eos_token_id=2, hidden_size=32, intermediate_size=37, layer_norm_eps=1e-05, num_attention_heads=4, num_hidden_layers=5, pad_token_id=1, vocab_size=1000, ) text_encoder = CLIPTextModel(text_encoder_config) tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip") components = { "unet": unet, "controlnet": controlnet, "scheduler": scheduler, "vae": vae, "text_encoder": text_encoder, "tokenizer": tokenizer, "safety_checker": None, "feature_extractor": None, "image_encoder": None, } return components def get_dummy_inputs(self, device, seed=0): if str(device).startswith("mps"): generator = torch.manual_seed(seed) else: generator = torch.Generator(device=device).manual_seed(seed) controlnet_embedder_scale_factor = 2 image = randn_tensor( (1, 3, 32 * controlnet_embedder_scale_factor, 32 * controlnet_embedder_scale_factor), generator=generator, device=torch.device(device), ) inputs = { "prompt": "A painting of a squirrel eating a burger", "generator": generator, "num_inference_steps": 2, "guidance_scale": 6.0, "output_type": "numpy", "image": image, } return inputs def test_attention_slicing_forward_pass(self): return self._test_attention_slicing_forward_pass(expected_max_diff=2e-3) @unittest.skipIf( torch_device != "cuda" or not is_xformers_available(), reason="XFormers attention is only available with CUDA and `xformers` installed", ) def test_xformers_attention_forwardGenerator_pass(self): self._test_xformers_attention_forwardGenerator_pass(expected_max_diff=2e-3) def test_inference_batch_single_identical(self): self._test_inference_batch_single_identical(expected_max_diff=2e-3) def test_controlnet_lcm(self): device = "cpu" # ensure determinism for the device-dependent torch.Generator components = self.get_dummy_components(time_cond_proj_dim=256) sd_pipe = StableDiffusionControlNetPipeline(**components) sd_pipe.scheduler = LCMScheduler.from_config(sd_pipe.scheduler.config) sd_pipe = sd_pipe.to(torch_device) sd_pipe.set_progress_bar_config(disable=None) inputs = self.get_dummy_inputs(device) output = sd_pipe(**inputs) image = output.images image_slice = image[0, -3:, -3:, -1] assert image.shape == (1, 64, 64, 3) expected_slice = np.array( [0.52700454, 0.3930534, 0.25509018, 0.7132304, 0.53696585, 0.46568912, 0.7095368, 0.7059624, 0.4744786] ) assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2 def test_controlnet_lcm_custom_timesteps(self): device = "cpu" # ensure determinism for the device-dependent torch.Generator components = self.get_dummy_components(time_cond_proj_dim=256) sd_pipe = StableDiffusionControlNetPipeline(**components) sd_pipe.scheduler = LCMScheduler.from_config(sd_pipe.scheduler.config) sd_pipe = sd_pipe.to(torch_device) sd_pipe.set_progress_bar_config(disable=None) inputs = self.get_dummy_inputs(device) del inputs["num_inference_steps"] inputs["timesteps"] = [999, 499] output = sd_pipe(**inputs) image = output.images image_slice = image[0, -3:, -3:, -1] assert image.shape == (1, 64, 64, 3) expected_slice = np.array( [0.52700454, 0.3930534, 0.25509018, 0.7132304, 0.53696585, 0.46568912, 0.7095368, 0.7059624, 0.4744786] ) assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2 class StableDiffusionMultiControlNetPipelineFastTests( PipelineTesterMixin, PipelineKarrasSchedulerTesterMixin, unittest.TestCase ): pipeline_class = StableDiffusionControlNetPipeline params = TEXT_TO_IMAGE_PARAMS batch_params = TEXT_TO_IMAGE_BATCH_PARAMS image_params = frozenset([]) # TO_DO: add image_params once refactored VaeImageProcessor.preprocess def get_dummy_components(self): torch.manual_seed(0) unet = UNet2DConditionModel( block_out_channels=(4, 8), layers_per_block=2, sample_size=32, in_channels=4, out_channels=4, down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"), up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"), cross_attention_dim=32, norm_num_groups=1, ) torch.manual_seed(0) def init_weights(m): if isinstance(m, torch.nn.Conv2d): torch.nn.init.normal(m.weight) m.bias.data.fill_(1.0) controlnet1 = ControlNetModel( block_out_channels=(4, 8), layers_per_block=2, in_channels=4, down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"), cross_attention_dim=32, conditioning_embedding_out_channels=(16, 32), norm_num_groups=1, ) controlnet1.controlnet_down_blocks.apply(init_weights) torch.manual_seed(0) controlnet2 = ControlNetModel( block_out_channels=(4, 8), layers_per_block=2, in_channels=4, down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"), cross_attention_dim=32, conditioning_embedding_out_channels=(16, 32), norm_num_groups=1, ) controlnet2.controlnet_down_blocks.apply(init_weights) torch.manual_seed(0) scheduler = DDIMScheduler( beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", clip_sample=False, set_alpha_to_one=False, ) torch.manual_seed(0) vae = AutoencoderKL( block_out_channels=[4, 8], in_channels=3, out_channels=3, down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"], up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"], latent_channels=4, norm_num_groups=2, ) torch.manual_seed(0) text_encoder_config = CLIPTextConfig( bos_token_id=0, eos_token_id=2, hidden_size=32, intermediate_size=37, layer_norm_eps=1e-05, num_attention_heads=4, num_hidden_layers=5, pad_token_id=1, vocab_size=1000, ) text_encoder = CLIPTextModel(text_encoder_config) tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip") controlnet = MultiControlNetModel([controlnet1, controlnet2]) components = { "unet": unet, "controlnet": controlnet, "scheduler": scheduler, "vae": vae, "text_encoder": text_encoder, "tokenizer": tokenizer, "safety_checker": None, "feature_extractor": None, "image_encoder": None, } return components def get_dummy_inputs(self, device, seed=0): if str(device).startswith("mps"): generator = torch.manual_seed(seed) else: generator = torch.Generator(device=device).manual_seed(seed) controlnet_embedder_scale_factor = 2 images = [ randn_tensor( (1, 3, 32 * controlnet_embedder_scale_factor, 32 * controlnet_embedder_scale_factor), generator=generator, device=torch.device(device), ), randn_tensor( (1, 3, 32 * controlnet_embedder_scale_factor, 32 * controlnet_embedder_scale_factor), generator=generator, device=torch.device(device), ), ] inputs = { "prompt": "A painting of a squirrel eating a burger", "generator": generator, "num_inference_steps": 2, "guidance_scale": 6.0, "output_type": "numpy", "image": images, } return inputs def test_control_guidance_switch(self): components = self.get_dummy_components() pipe = self.pipeline_class(**components) pipe.to(torch_device) scale = 10.0 steps = 4 inputs = self.get_dummy_inputs(torch_device) inputs["num_inference_steps"] = steps inputs["controlnet_conditioning_scale"] = scale output_1 = pipe(**inputs)[0] inputs = self.get_dummy_inputs(torch_device) inputs["num_inference_steps"] = steps inputs["controlnet_conditioning_scale"] = scale output_2 = pipe(**inputs, control_guidance_start=0.1, control_guidance_end=0.2)[0] inputs = self.get_dummy_inputs(torch_device) inputs["num_inference_steps"] = steps inputs["controlnet_conditioning_scale"] = scale output_3 = pipe(**inputs, control_guidance_start=[0.1, 0.3], control_guidance_end=[0.2, 0.7])[0] inputs = self.get_dummy_inputs(torch_device) inputs["num_inference_steps"] = steps inputs["controlnet_conditioning_scale"] = scale output_4 = pipe(**inputs, control_guidance_start=0.4, control_guidance_end=[0.5, 0.8])[0] # make sure that all outputs are different assert np.sum(np.abs(output_1 - output_2)) > 1e-3 assert np.sum(np.abs(output_1 - output_3)) > 1e-3 assert np.sum(np.abs(output_1 - output_4)) > 1e-3 def test_attention_slicing_forward_pass(self): return self._test_attention_slicing_forward_pass(expected_max_diff=2e-3) @unittest.skipIf( torch_device != "cuda" or not is_xformers_available(), reason="XFormers attention is only available with CUDA and `xformers` installed", ) def test_xformers_attention_forwardGenerator_pass(self): self._test_xformers_attention_forwardGenerator_pass(expected_max_diff=2e-3) def test_inference_batch_single_identical(self): self._test_inference_batch_single_identical(expected_max_diff=2e-3) def test_save_pretrained_raise_not_implemented_exception(self): components = self.get_dummy_components() pipe = self.pipeline_class(**components) pipe.to(torch_device) pipe.set_progress_bar_config(disable=None) with tempfile.TemporaryDirectory() as tmpdir: try: # save_pretrained is not implemented for Multi-ControlNet pipe.save_pretrained(tmpdir) except NotImplementedError: pass def test_inference_multiple_prompt_input(self): device = "cpu" components = self.get_dummy_components() sd_pipe = StableDiffusionControlNetPipeline(**components) sd_pipe = sd_pipe.to(torch_device) sd_pipe.set_progress_bar_config(disable=None) inputs = self.get_dummy_inputs(device) inputs["prompt"] = [inputs["prompt"], inputs["prompt"]] inputs["image"] = [inputs["image"], inputs["image"]] output = sd_pipe(**inputs) image = output.images assert image.shape == (2, 64, 64, 3) image_1, image_2 = image # make sure that the outputs are different assert np.sum(np.abs(image_1 - image_2)) > 1e-3 # multiple prompts, single image conditioning inputs = self.get_dummy_inputs(device) inputs["prompt"] = [inputs["prompt"], inputs["prompt"]] output_1 = sd_pipe(**inputs) assert np.abs(image - output_1.images).max() < 1e-3 class StableDiffusionMultiControlNetOneModelPipelineFastTests( PipelineTesterMixin, PipelineKarrasSchedulerTesterMixin, unittest.TestCase ): pipeline_class = StableDiffusionControlNetPipeline params = TEXT_TO_IMAGE_PARAMS batch_params = TEXT_TO_IMAGE_BATCH_PARAMS image_params = frozenset([]) # TO_DO: add image_params once refactored VaeImageProcessor.preprocess def get_dummy_components(self): torch.manual_seed(0) unet = UNet2DConditionModel( block_out_channels=(4, 8), layers_per_block=2, sample_size=32, in_channels=4, out_channels=4, down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"), up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"), cross_attention_dim=32, norm_num_groups=1, ) torch.manual_seed(0) def init_weights(m): if isinstance(m, torch.nn.Conv2d): torch.nn.init.normal(m.weight) m.bias.data.fill_(1.0) controlnet = ControlNetModel( block_out_channels=(4, 8), layers_per_block=2, in_channels=4, down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"), cross_attention_dim=32, conditioning_embedding_out_channels=(16, 32), norm_num_groups=1, ) controlnet.controlnet_down_blocks.apply(init_weights) torch.manual_seed(0) scheduler = DDIMScheduler( beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", clip_sample=False, set_alpha_to_one=False, ) torch.manual_seed(0) vae = AutoencoderKL( block_out_channels=[4, 8], in_channels=3, out_channels=3, down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"], up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"], latent_channels=4, norm_num_groups=2, ) torch.manual_seed(0) text_encoder_config = CLIPTextConfig( bos_token_id=0, eos_token_id=2, hidden_size=32, intermediate_size=37, layer_norm_eps=1e-05, num_attention_heads=4, num_hidden_layers=5, pad_token_id=1, vocab_size=1000, ) text_encoder = CLIPTextModel(text_encoder_config) tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip") controlnet = MultiControlNetModel([controlnet]) components = { "unet": unet, "controlnet": controlnet, "scheduler": scheduler, "vae": vae, "text_encoder": text_encoder, "tokenizer": tokenizer, "safety_checker": None, "feature_extractor": None, "image_encoder": None, } return components def get_dummy_inputs(self, device, seed=0): if str(device).startswith("mps"): generator = torch.manual_seed(seed) else: generator = torch.Generator(device=device).manual_seed(seed) controlnet_embedder_scale_factor = 2 images = [ randn_tensor( (1, 3, 32 * controlnet_embedder_scale_factor, 32 * controlnet_embedder_scale_factor), generator=generator, device=torch.device(device), ), ] inputs = { "prompt": "A painting of a squirrel eating a burger", "generator": generator, "num_inference_steps": 2, "guidance_scale": 6.0, "output_type": "numpy", "image": images, } return inputs def test_control_guidance_switch(self): components = self.get_dummy_components() pipe = self.pipeline_class(**components) pipe.to(torch_device) scale = 10.0 steps = 4 inputs = self.get_dummy_inputs(torch_device) inputs["num_inference_steps"] = steps inputs["controlnet_conditioning_scale"] = scale output_1 = pipe(**inputs)[0] inputs = self.get_dummy_inputs(torch_device) inputs["num_inference_steps"] = steps inputs["controlnet_conditioning_scale"] = scale output_2 = pipe(**inputs, control_guidance_start=0.1, control_guidance_end=0.2)[0] inputs = self.get_dummy_inputs(torch_device) inputs["num_inference_steps"] = steps inputs["controlnet_conditioning_scale"] = scale output_3 = pipe( **inputs, control_guidance_start=[0.1], control_guidance_end=[0.2], )[0] inputs = self.get_dummy_inputs(torch_device) inputs["num_inference_steps"] = steps inputs["controlnet_conditioning_scale"] = scale output_4 = pipe(**inputs, control_guidance_start=0.4, control_guidance_end=[0.5])[0] # make sure that all outputs are different assert np.sum(np.abs(output_1 - output_2)) > 1e-3 assert np.sum(np.abs(output_1 - output_3)) > 1e-3 assert np.sum(np.abs(output_1 - output_4)) > 1e-3 def test_attention_slicing_forward_pass(self): return self._test_attention_slicing_forward_pass(expected_max_diff=2e-3) @unittest.skipIf( torch_device != "cuda" or not is_xformers_available(), reason="XFormers attention is only available with CUDA and `xformers` installed", ) def test_xformers_attention_forwardGenerator_pass(self): self._test_xformers_attention_forwardGenerator_pass(expected_max_diff=2e-3) def test_inference_batch_single_identical(self): self._test_inference_batch_single_identical(expected_max_diff=2e-3) def test_save_pretrained_raise_not_implemented_exception(self): components = self.get_dummy_components() pipe = self.pipeline_class(**components) pipe.to(torch_device) pipe.set_progress_bar_config(disable=None) with tempfile.TemporaryDirectory() as tmpdir: try: # save_pretrained is not implemented for Multi-ControlNet pipe.save_pretrained(tmpdir) except NotImplementedError: pass @slow @require_torch_gpu class ControlNetPipelineSlowTests(unittest.TestCase): def tearDown(self): super().tearDown() gc.collect() torch.cuda.empty_cache() def test_canny(self): controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.enable_model_cpu_offload() pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device="cpu").manual_seed(0) prompt = "bird" image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/bird_canny.png" ) output = pipe(prompt, image, generator=generator, output_type="np", num_inference_steps=3) image = output.images[0] assert image.shape == (768, 512, 3) expected_image = load_numpy( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/bird_canny_out.npy" ) assert np.abs(expected_image - image).max() < 9e-2 def test_depth(self): controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-depth") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.enable_model_cpu_offload() pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device="cpu").manual_seed(0) prompt = "Stormtrooper's lecture" image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/stormtrooper_depth.png" ) output = pipe(prompt, image, generator=generator, output_type="np", num_inference_steps=3) image = output.images[0] assert image.shape == (512, 512, 3) expected_image = load_numpy( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/stormtrooper_depth_out.npy" ) assert np.abs(expected_image - image).max() < 8e-1 def test_hed(self): controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-hed") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.enable_model_cpu_offload() pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device="cpu").manual_seed(0) prompt = "oil painting of handsome old man, masterpiece" image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/man_hed.png" ) output = pipe(prompt, image, generator=generator, output_type="np", num_inference_steps=3) image = output.images[0] assert image.shape == (704, 512, 3) expected_image = load_numpy( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/man_hed_out.npy" ) assert np.abs(expected_image - image).max() < 8e-2 def test_mlsd(self): controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-mlsd") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.enable_model_cpu_offload() pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device="cpu").manual_seed(0) prompt = "room" image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/room_mlsd.png" ) output = pipe(prompt, image, generator=generator, output_type="np", num_inference_steps=3) image = output.images[0] assert image.shape == (704, 512, 3) expected_image = load_numpy( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/room_mlsd_out.npy" ) assert np.abs(expected_image - image).max() < 5e-2 def test_normal(self): controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-normal") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.enable_model_cpu_offload() pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device="cpu").manual_seed(0) prompt = "cute toy" image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/cute_toy_normal.png" ) output = pipe(prompt, image, generator=generator, output_type="np", num_inference_steps=3) image = output.images[0] assert image.shape == (512, 512, 3) expected_image = load_numpy( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/cute_toy_normal_out.npy" ) assert np.abs(expected_image - image).max() < 5e-2 def test_openpose(self): controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-openpose") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.enable_model_cpu_offload() pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device="cpu").manual_seed(0) prompt = "Chef in the kitchen" image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/pose.png" ) output = pipe(prompt, image, generator=generator, output_type="np", num_inference_steps=3) image = output.images[0] assert image.shape == (768, 512, 3) expected_image = load_numpy( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/chef_pose_out.npy" ) assert np.abs(expected_image - image).max() < 8e-2 def test_scribble(self): controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-scribble") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.enable_model_cpu_offload() pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device="cpu").manual_seed(5) prompt = "bag" image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/bag_scribble.png" ) output = pipe(prompt, image, generator=generator, output_type="np", num_inference_steps=3) image = output.images[0] assert image.shape == (640, 512, 3) expected_image = load_numpy( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/bag_scribble_out.npy" ) assert np.abs(expected_image - image).max() < 8e-2 def test_seg(self): controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-seg") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.enable_model_cpu_offload() pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device="cpu").manual_seed(5) prompt = "house" image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/house_seg.png" ) output = pipe(prompt, image, generator=generator, output_type="np", num_inference_steps=3) image = output.images[0] assert image.shape == (512, 512, 3) expected_image = load_numpy( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/house_seg_out.npy" ) assert np.abs(expected_image - image).max() < 8e-2 def test_sequential_cpu_offloading(self): torch.cuda.empty_cache() torch.cuda.reset_max_memory_allocated() torch.cuda.reset_peak_memory_stats() controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-seg") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.set_progress_bar_config(disable=None) pipe.enable_attention_slicing() pipe.enable_sequential_cpu_offload() prompt = "house" image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/house_seg.png" ) _ = pipe( prompt, image, num_inference_steps=2, output_type="np", ) mem_bytes = torch.cuda.max_memory_allocated() # make sure that less than 7 GB is allocated assert mem_bytes < 4 * 10**9 def test_canny_guess_mode(self): controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.enable_model_cpu_offload() pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device="cpu").manual_seed(0) prompt = "" image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/bird_canny.png" ) output = pipe( prompt, image, generator=generator, output_type="np", num_inference_steps=3, guidance_scale=3.0, guess_mode=True, ) image = output.images[0] assert image.shape == (768, 512, 3) image_slice = image[-3:, -3:, -1] expected_slice = np.array([0.2724, 0.2846, 0.2724, 0.3843, 0.3682, 0.2736, 0.4675, 0.3862, 0.2887]) assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2 def test_canny_guess_mode_euler(self): controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config) pipe.enable_model_cpu_offload() pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device="cpu").manual_seed(0) prompt = "" image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/bird_canny.png" ) output = pipe( prompt, image, generator=generator, output_type="np", num_inference_steps=3, guidance_scale=3.0, guess_mode=True, ) image = output.images[0] assert image.shape == (768, 512, 3) image_slice = image[-3:, -3:, -1] expected_slice = np.array([0.1655, 0.1721, 0.1623, 0.1685, 0.1711, 0.1646, 0.1651, 0.1631, 0.1494]) assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2 @require_python39_or_higher @require_torch_2 def test_stable_diffusion_compile(self): run_test_in_subprocess(test_case=self, target_func=_test_stable_diffusion_compile, inputs=None) def test_v11_shuffle_global_pool_conditions(self): controlnet = ControlNetModel.from_pretrained("lllyasviel/control_v11e_sd15_shuffle") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.enable_model_cpu_offload() pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device="cpu").manual_seed(0) prompt = "New York" image = load_image( "https://huggingface.co/lllyasviel/control_v11e_sd15_shuffle/resolve/main/images/control.png" ) output = pipe( prompt, image, generator=generator, output_type="np", num_inference_steps=3, guidance_scale=7.0, ) image = output.images[0] assert image.shape == (512, 640, 3) image_slice = image[-3:, -3:, -1] expected_slice = np.array([0.1338, 0.1597, 0.1202, 0.1687, 0.1377, 0.1017, 0.2070, 0.1574, 0.1348]) assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2 def test_load_local(self): controlnet = ControlNetModel.from_pretrained("lllyasviel/control_v11p_sd15_canny") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=controlnet ) pipe.unet.set_default_attn_processor() pipe.enable_model_cpu_offload() controlnet = ControlNetModel.from_single_file( "https://huggingface.co/lllyasviel/ControlNet-v1-1/blob/main/control_v11p_sd15_canny.pth" ) pipe_sf = StableDiffusionControlNetPipeline.from_single_file( "https://huggingface.co/runwayml/stable-diffusion-v1-5/blob/main/v1-5-pruned-emaonly.safetensors", safety_checker=None, controlnet=controlnet, scheduler_type="pndm", ) pipe_sf.unet.set_default_attn_processor() pipe_sf.enable_model_cpu_offload() control_image = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/bird_canny.png" ).resize((512, 512)) prompt = "bird" generator = torch.Generator(device="cpu").manual_seed(0) output = pipe( prompt, image=control_image, generator=generator, output_type="np", num_inference_steps=3, ).images[0] generator = torch.Generator(device="cpu").manual_seed(0) output_sf = pipe_sf( prompt, image=control_image, generator=generator, output_type="np", num_inference_steps=3, ).images[0] max_diff = numpy_cosine_similarity_distance(output_sf.flatten(), output.flatten()) assert max_diff < 1e-3 @slow @require_torch_gpu class StableDiffusionMultiControlNetPipelineSlowTests(unittest.TestCase): def tearDown(self): super().tearDown() gc.collect() torch.cuda.empty_cache() def test_pose_and_canny(self): controlnet_canny = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny") controlnet_pose = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-openpose") pipe = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", safety_checker=None, controlnet=[controlnet_pose, controlnet_canny] ) pipe.enable_model_cpu_offload() pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device="cpu").manual_seed(0) prompt = "bird and Chef" image_canny = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/bird_canny.png" ) image_pose = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/pose.png" ) output = pipe(prompt, [image_pose, image_canny], generator=generator, output_type="np", num_inference_steps=3) image = output.images[0] assert image.shape == (768, 512, 3) expected_image = load_numpy( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/pose_canny_out.npy" ) assert np.abs(expected_image - image).max() < 5e-2

diffusers/tests/pipelines/controlnet/test_controlnet.py/0

# coding=utf-8 # Copyright 2023 HuggingFace Inc. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import gc import unittest import numpy as np import torch from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer from diffusers import AutoencoderKL, DDIMScheduler, LDMTextToImagePipeline, UNet2DConditionModel from diffusers.utils.testing_utils import ( enable_full_determinism, load_numpy, nightly, require_torch_gpu, torch_device, ) from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_PARAMS from ..test_pipelines_common import PipelineTesterMixin enable_full_determinism() class LDMTextToImagePipelineFastTests(PipelineTesterMixin, unittest.TestCase): pipeline_class = LDMTextToImagePipeline params = TEXT_TO_IMAGE_PARAMS - { "negative_prompt", "negative_prompt_embeds", "cross_attention_kwargs", "prompt_embeds", } required_optional_params = PipelineTesterMixin.required_optional_params - { "num_images_per_prompt", "callback", "callback_steps", } batch_params = TEXT_TO_IMAGE_BATCH_PARAMS def get_dummy_components(self): torch.manual_seed(0) unet = UNet2DConditionModel( block_out_channels=(32, 64), layers_per_block=2, sample_size=32, in_channels=4, out_channels=4, down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"), up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"), cross_attention_dim=32, ) scheduler = DDIMScheduler( beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", clip_sample=False, set_alpha_to_one=False, ) torch.manual_seed(0) vae = AutoencoderKL( block_out_channels=(32, 64), in_channels=3, out_channels=3, down_block_types=("DownEncoderBlock2D", "DownEncoderBlock2D"), up_block_types=("UpDecoderBlock2D", "UpDecoderBlock2D"), latent_channels=4, ) torch.manual_seed(0) text_encoder_config = CLIPTextConfig( bos_token_id=0, eos_token_id=2, hidden_size=32, intermediate_size=37, layer_norm_eps=1e-05, num_attention_heads=4, num_hidden_layers=5, pad_token_id=1, vocab_size=1000, ) text_encoder = CLIPTextModel(text_encoder_config) tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip") components = { "unet": unet, "scheduler": scheduler, "vqvae": vae, "bert": text_encoder, "tokenizer": tokenizer, } return components def get_dummy_inputs(self, device, seed=0): if str(device).startswith("mps"): generator = torch.manual_seed(seed) else: generator = torch.Generator(device=device).manual_seed(seed) inputs = { "prompt": "A painting of a squirrel eating a burger", "generator": generator, "num_inference_steps": 2, "guidance_scale": 6.0, "output_type": "numpy", } return inputs def test_inference_text2img(self): device = "cpu" # ensure determinism for the device-dependent torch.Generator components = self.get_dummy_components() pipe = LDMTextToImagePipeline(**components) pipe.to(device) pipe.set_progress_bar_config(disable=None) inputs = self.get_dummy_inputs(device) image = pipe(**inputs).images image_slice = image[0, -3:, -3:, -1] assert image.shape == (1, 16, 16, 3) expected_slice = np.array([0.6101, 0.6156, 0.5622, 0.4895, 0.6661, 0.3804, 0.5748, 0.6136, 0.5014]) assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3 @nightly @require_torch_gpu class LDMTextToImagePipelineSlowTests(unittest.TestCase): def tearDown(self): super().tearDown() gc.collect() torch.cuda.empty_cache() def get_inputs(self, device, dtype=torch.float32, seed=0): generator = torch.manual_seed(seed) latents = np.random.RandomState(seed).standard_normal((1, 4, 32, 32)) latents = torch.from_numpy(latents).to(device=device, dtype=dtype) inputs = { "prompt": "A painting of a squirrel eating a burger", "latents": latents, "generator": generator, "num_inference_steps": 3, "guidance_scale": 6.0, "output_type": "numpy", } return inputs def test_ldm_default_ddim(self): pipe = LDMTextToImagePipeline.from_pretrained("CompVis/ldm-text2im-large-256").to(torch_device) pipe.set_progress_bar_config(disable=None) inputs = self.get_inputs(torch_device) image = pipe(**inputs).images image_slice = image[0, -3:, -3:, -1].flatten() assert image.shape == (1, 256, 256, 3) expected_slice = np.array([0.51825, 0.52850, 0.52543, 0.54258, 0.52304, 0.52569, 0.54363, 0.55276, 0.56878]) max_diff = np.abs(expected_slice - image_slice).max() assert max_diff < 1e-3 @nightly @require_torch_gpu class LDMTextToImagePipelineNightlyTests(unittest.TestCase): def tearDown(self): super().tearDown() gc.collect() torch.cuda.empty_cache() def get_inputs(self, device, dtype=torch.float32, seed=0): generator = torch.manual_seed(seed) latents = np.random.RandomState(seed).standard_normal((1, 4, 32, 32)) latents = torch.from_numpy(latents).to(device=device, dtype=dtype) inputs = { "prompt": "A painting of a squirrel eating a burger", "latents": latents, "generator": generator, "num_inference_steps": 50, "guidance_scale": 6.0, "output_type": "numpy", } return inputs def test_ldm_default_ddim(self): pipe = LDMTextToImagePipeline.from_pretrained("CompVis/ldm-text2im-large-256").to(torch_device) pipe.set_progress_bar_config(disable=None) inputs = self.get_inputs(torch_device) image = pipe(**inputs).images[0] expected_image = load_numpy( "https://huggingface.co/datasets/diffusers/test-arrays/resolve/main/ldm_text2img/ldm_large_256_ddim.npy" ) max_diff = np.abs(expected_image - image).max() assert max_diff < 1e-3

diffusers/tests/pipelines/latent_diffusion/test_latent_diffusion.py/0

# Copyright 2023 HuggingFace Inc. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import gc import unittest import numpy as np import torch from transformers import CLIPTextConfig, CLIPTextModelWithProjection, CLIPTokenizer from diffusers import HeunDiscreteScheduler, PriorTransformer, ShapEPipeline from diffusers.pipelines.shap_e import ShapERenderer from diffusers.utils.testing_utils import load_numpy, nightly, require_torch_gpu, torch_device from ..test_pipelines_common import PipelineTesterMixin, assert_mean_pixel_difference class ShapEPipelineFastTests(PipelineTesterMixin, unittest.TestCase): pipeline_class = ShapEPipeline params = ["prompt"] batch_params = ["prompt"] required_optional_params = [ "num_images_per_prompt", "num_inference_steps", "generator", "latents", "guidance_scale", "frame_size", "output_type", "return_dict", ] test_xformers_attention = False @property def text_embedder_hidden_size(self): return 16 @property def time_input_dim(self): return 16 @property def time_embed_dim(self): return self.time_input_dim * 4 @property def renderer_dim(self): return 8 @property def dummy_tokenizer(self): tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip") return tokenizer @property def dummy_text_encoder(self): torch.manual_seed(0) config = CLIPTextConfig( bos_token_id=0, eos_token_id=2, hidden_size=self.text_embedder_hidden_size, projection_dim=self.text_embedder_hidden_size, intermediate_size=37, layer_norm_eps=1e-05, num_attention_heads=4, num_hidden_layers=5, pad_token_id=1, vocab_size=1000, ) return CLIPTextModelWithProjection(config) @property def dummy_prior(self): torch.manual_seed(0) model_kwargs = { "num_attention_heads": 2, "attention_head_dim": 16, "embedding_dim": self.time_input_dim, "num_embeddings": 32, "embedding_proj_dim": self.text_embedder_hidden_size, "time_embed_dim": self.time_embed_dim, "num_layers": 1, "clip_embed_dim": self.time_input_dim * 2, "additional_embeddings": 0, "time_embed_act_fn": "gelu", "norm_in_type": "layer", "encoder_hid_proj_type": None, "added_emb_type": None, } model = PriorTransformer(**model_kwargs) return model @property def dummy_renderer(self): torch.manual_seed(0) model_kwargs = { "param_shapes": ( (self.renderer_dim, 93), (self.renderer_dim, 8), (self.renderer_dim, 8), (self.renderer_dim, 8), ), "d_latent": self.time_input_dim, "d_hidden": self.renderer_dim, "n_output": 12, "background": ( 0.1, 0.1, 0.1, ), } model = ShapERenderer(**model_kwargs) return model def get_dummy_components(self): prior = self.dummy_prior text_encoder = self.dummy_text_encoder tokenizer = self.dummy_tokenizer shap_e_renderer = self.dummy_renderer scheduler = HeunDiscreteScheduler( beta_schedule="exp", num_train_timesteps=1024, prediction_type="sample", use_karras_sigmas=True, clip_sample=True, clip_sample_range=1.0, ) components = { "prior": prior, "text_encoder": text_encoder, "tokenizer": tokenizer, "shap_e_renderer": shap_e_renderer, "scheduler": scheduler, } return components def get_dummy_inputs(self, device, seed=0): if str(device).startswith("mps"): generator = torch.manual_seed(seed) else: generator = torch.Generator(device=device).manual_seed(seed) inputs = { "prompt": "horse", "generator": generator, "num_inference_steps": 1, "frame_size": 32, "output_type": "latent", } return inputs def test_shap_e(self): device = "cpu" components = self.get_dummy_components() pipe = self.pipeline_class(**components) pipe = pipe.to(device) pipe.set_progress_bar_config(disable=None) output = pipe(**self.get_dummy_inputs(device)) image = output.images[0] image = image.cpu().numpy() image_slice = image[-3:, -3:] assert image.shape == (32, 16) expected_slice = np.array([-1.0000, -0.6241, 1.0000, -0.8978, -0.6866, 0.7876, -0.7473, -0.2874, 0.6103]) assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2 def test_inference_batch_consistent(self): # NOTE: Larger batch sizes cause this test to timeout, only test on smaller batches self._test_inference_batch_consistent(batch_sizes=[1, 2]) def test_inference_batch_single_identical(self): self._test_inference_batch_single_identical(batch_size=2, expected_max_diff=6e-3) def test_num_images_per_prompt(self): components = self.get_dummy_components() pipe = self.pipeline_class(**components) pipe = pipe.to(torch_device) pipe.set_progress_bar_config(disable=None) batch_size = 1 num_images_per_prompt = 2 inputs = self.get_dummy_inputs(torch_device) for key in inputs.keys(): if key in self.batch_params: inputs[key] = batch_size * [inputs[key]] images = pipe(**inputs, num_images_per_prompt=num_images_per_prompt)[0] assert images.shape[0] == batch_size * num_images_per_prompt def test_float16_inference(self): super().test_float16_inference(expected_max_diff=5e-1) def test_save_load_local(self): super().test_save_load_local(expected_max_difference=5e-3) @unittest.skip("Key error is raised with accelerate") def test_sequential_cpu_offload_forward_pass(self): pass @nightly @require_torch_gpu class ShapEPipelineIntegrationTests(unittest.TestCase): def tearDown(self): # clean up the VRAM after each test super().tearDown() gc.collect() torch.cuda.empty_cache() def test_shap_e(self): expected_image = load_numpy( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main" "/shap_e/test_shap_e_np_out.npy" ) pipe = ShapEPipeline.from_pretrained("openai/shap-e") pipe = pipe.to(torch_device) pipe.set_progress_bar_config(disable=None) generator = torch.Generator(device=torch_device).manual_seed(0) images = pipe( "a shark", generator=generator, guidance_scale=15.0, num_inference_steps=64, frame_size=64, output_type="np", ).images[0] assert images.shape == (20, 64, 64, 3) assert_mean_pixel_difference(images, expected_image)

diffusers/tests/pipelines/shap_e/test_shap_e.py/0

# coding=utf-8 # Copyright 2023 HuggingFace Inc. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import gc import unittest from diffusers import FlaxDPMSolverMultistepScheduler, FlaxStableDiffusionPipeline from diffusers.utils import is_flax_available from diffusers.utils.testing_utils import nightly, require_flax if is_flax_available(): import jax import jax.numpy as jnp from flax.jax_utils import replicate from flax.training.common_utils import shard @nightly @require_flax class FlaxStableDiffusion2PipelineIntegrationTests(unittest.TestCase): def tearDown(self): # clean up the VRAM after each test super().tearDown() gc.collect() def test_stable_diffusion_flax(self): sd_pipe, params = FlaxStableDiffusionPipeline.from_pretrained( "stabilityai/stable-diffusion-2", revision="bf16", dtype=jnp.bfloat16, ) prompt = "A painting of a squirrel eating a burger" num_samples = jax.device_count() prompt = num_samples * [prompt] prompt_ids = sd_pipe.prepare_inputs(prompt) params = replicate(params) prompt_ids = shard(prompt_ids) prng_seed = jax.random.PRNGKey(0) prng_seed = jax.random.split(prng_seed, jax.device_count()) images = sd_pipe(prompt_ids, params, prng_seed, num_inference_steps=25, jit=True)[0] assert images.shape == (jax.device_count(), 1, 768, 768, 3) images = images.reshape((images.shape[0] * images.shape[1],) + images.shape[-3:]) image_slice = images[0, 253:256, 253:256, -1] output_slice = jnp.asarray(jax.device_get(image_slice.flatten())) expected_slice = jnp.array([0.4238, 0.4414, 0.4395, 0.4453, 0.4629, 0.4590, 0.4531, 0.45508, 0.4512]) print(f"output_slice: {output_slice}") assert jnp.abs(output_slice - expected_slice).max() < 1e-2 @nightly @require_flax class FlaxStableDiffusion2PipelineNightlyTests(unittest.TestCase): def tearDown(self): # clean up the VRAM after each test super().tearDown() gc.collect() def test_stable_diffusion_dpm_flax(self): model_id = "stabilityai/stable-diffusion-2" scheduler, scheduler_params = FlaxDPMSolverMultistepScheduler.from_pretrained(model_id, subfolder="scheduler") sd_pipe, params = FlaxStableDiffusionPipeline.from_pretrained( model_id, scheduler=scheduler, revision="bf16", dtype=jnp.bfloat16, ) params["scheduler"] = scheduler_params prompt = "A painting of a squirrel eating a burger" num_samples = jax.device_count() prompt = num_samples * [prompt] prompt_ids = sd_pipe.prepare_inputs(prompt) params = replicate(params) prompt_ids = shard(prompt_ids) prng_seed = jax.random.PRNGKey(0) prng_seed = jax.random.split(prng_seed, jax.device_count()) images = sd_pipe(prompt_ids, params, prng_seed, num_inference_steps=25, jit=True)[0] assert images.shape == (jax.device_count(), 1, 768, 768, 3) images = images.reshape((images.shape[0] * images.shape[1],) + images.shape[-3:]) image_slice = images[0, 253:256, 253:256, -1] output_slice = jnp.asarray(jax.device_get(image_slice.flatten())) expected_slice = jnp.array([0.4336, 0.42969, 0.4453, 0.4199, 0.4297, 0.4531, 0.4434, 0.4434, 0.4297]) print(f"output_slice: {output_slice}") assert jnp.abs(output_slice - expected_slice).max() < 1e-2

diffusers/tests/pipelines/stable_diffusion_2/test_stable_diffusion_flax.py/0

import gc import unittest import torch from transformers import CLIPTextConfig, CLIPTextModel, CLIPTextModelWithProjection, CLIPTokenizer from diffusers import ( AutoencoderKL, DDIMScheduler, DDPMScheduler, PriorTransformer, StableUnCLIPPipeline, UNet2DConditionModel, ) from diffusers.pipelines.stable_diffusion.stable_unclip_image_normalizer import StableUnCLIPImageNormalizer from diffusers.utils.testing_utils import enable_full_determinism, load_numpy, nightly, require_torch_gpu, torch_device from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_PARAMS from ..test_pipelines_common import ( PipelineKarrasSchedulerTesterMixin, PipelineLatentTesterMixin, PipelineTesterMixin, assert_mean_pixel_difference, ) enable_full_determinism() class StableUnCLIPPipelineFastTests( PipelineLatentTesterMixin, PipelineKarrasSchedulerTesterMixin, PipelineTesterMixin, unittest.TestCase ): pipeline_class = StableUnCLIPPipeline params = TEXT_TO_IMAGE_PARAMS batch_params = TEXT_TO_IMAGE_BATCH_PARAMS image_params = TEXT_TO_IMAGE_IMAGE_PARAMS image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS # TODO(will) Expected attn_bias.stride(1) == 0 to be true, but got false test_xformers_attention = False def get_dummy_components(self): embedder_hidden_size = 32 embedder_projection_dim = embedder_hidden_size # prior components torch.manual_seed(0) prior_tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip") torch.manual_seed(0) prior_text_encoder = CLIPTextModelWithProjection( CLIPTextConfig( bos_token_id=0, eos_token_id=2, hidden_size=embedder_hidden_size, projection_dim=embedder_projection_dim, intermediate_size=37, layer_norm_eps=1e-05, num_attention_heads=4, num_hidden_layers=5, pad_token_id=1, vocab_size=1000, ) ) torch.manual_seed(0) prior = PriorTransformer( num_attention_heads=2, attention_head_dim=12, embedding_dim=embedder_projection_dim, num_layers=1, ) torch.manual_seed(0) prior_scheduler = DDPMScheduler( variance_type="fixed_small_log", prediction_type="sample", num_train_timesteps=1000, clip_sample=True, clip_sample_range=5.0, beta_schedule="squaredcos_cap_v2", ) # regular denoising components torch.manual_seed(0) image_normalizer = StableUnCLIPImageNormalizer(embedding_dim=embedder_hidden_size) image_noising_scheduler = DDPMScheduler(beta_schedule="squaredcos_cap_v2") torch.manual_seed(0) tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip") torch.manual_seed(0) text_encoder = CLIPTextModel( CLIPTextConfig( bos_token_id=0, eos_token_id=2, hidden_size=embedder_hidden_size, projection_dim=32, intermediate_size=37, layer_norm_eps=1e-05, num_attention_heads=4, num_hidden_layers=5, pad_token_id=1, vocab_size=1000, ) ) torch.manual_seed(0) unet = UNet2DConditionModel( sample_size=32, in_channels=4, out_channels=4, down_block_types=("CrossAttnDownBlock2D", "DownBlock2D"), up_block_types=("UpBlock2D", "CrossAttnUpBlock2D"), block_out_channels=(32, 64), attention_head_dim=(2, 4), class_embed_type="projection", # The class embeddings are the noise augmented image embeddings. # I.e. the image embeddings concated with the noised embeddings of the same dimension projection_class_embeddings_input_dim=embedder_projection_dim * 2, cross_attention_dim=embedder_hidden_size, layers_per_block=1, upcast_attention=True, use_linear_projection=True, ) torch.manual_seed(0) scheduler = DDIMScheduler( beta_schedule="scaled_linear", beta_start=0.00085, beta_end=0.012, prediction_type="v_prediction", set_alpha_to_one=False, steps_offset=1, ) torch.manual_seed(0) vae = AutoencoderKL() components = { # prior components "prior_tokenizer": prior_tokenizer, "prior_text_encoder": prior_text_encoder, "prior": prior, "prior_scheduler": prior_scheduler, # image noising components "image_normalizer": image_normalizer, "image_noising_scheduler": image_noising_scheduler, # regular denoising components "tokenizer": tokenizer, "text_encoder": text_encoder, "unet": unet, "scheduler": scheduler, "vae": vae, } return components def get_dummy_inputs(self, device, seed=0): if str(device).startswith("mps"): generator = torch.manual_seed(seed) else: generator = torch.Generator(device=device).manual_seed(seed) inputs = { "prompt": "A painting of a squirrel eating a burger", "generator": generator, "num_inference_steps": 2, "prior_num_inference_steps": 2, "output_type": "numpy", } return inputs # Overriding PipelineTesterMixin::test_attention_slicing_forward_pass # because UnCLIP GPU undeterminism requires a looser check. def test_attention_slicing_forward_pass(self): test_max_difference = torch_device == "cpu" self._test_attention_slicing_forward_pass(test_max_difference=test_max_difference) # Overriding PipelineTesterMixin::test_inference_batch_single_identical # because UnCLIP undeterminism requires a looser check. def test_inference_batch_single_identical(self): self._test_inference_batch_single_identical(expected_max_diff=1e-3) @nightly @require_torch_gpu class StableUnCLIPPipelineIntegrationTests(unittest.TestCase): def tearDown(self): # clean up the VRAM after each test super().tearDown() gc.collect() torch.cuda.empty_cache() def test_stable_unclip(self): expected_image = load_numpy( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/stable_unclip/stable_unclip_2_1_l_anime_turtle_fp16.npy" ) pipe = StableUnCLIPPipeline.from_pretrained("fusing/stable-unclip-2-1-l", torch_dtype=torch.float16) pipe.to(torch_device) pipe.set_progress_bar_config(disable=None) # stable unclip will oom when integration tests are run on a V100, # so turn on memory savings pipe.enable_attention_slicing() pipe.enable_sequential_cpu_offload() generator = torch.Generator(device="cpu").manual_seed(0) output = pipe("anime turle", generator=generator, output_type="np") image = output.images[0] assert image.shape == (768, 768, 3) assert_mean_pixel_difference(image, expected_image) def test_stable_unclip_pipeline_with_sequential_cpu_offloading(self): torch.cuda.empty_cache() torch.cuda.reset_max_memory_allocated() torch.cuda.reset_peak_memory_stats() pipe = StableUnCLIPPipeline.from_pretrained("fusing/stable-unclip-2-1-l", torch_dtype=torch.float16) pipe = pipe.to(torch_device) pipe.set_progress_bar_config(disable=None) pipe.enable_attention_slicing() pipe.enable_sequential_cpu_offload() _ = pipe( "anime turtle", prior_num_inference_steps=2, num_inference_steps=2, output_type="np", ) mem_bytes = torch.cuda.max_memory_allocated() # make sure that less than 7 GB is allocated assert mem_bytes < 7 * 10**9

diffusers/tests/pipelines/stable_unclip/test_stable_unclip.py/0

import tempfile import unittest import numpy as np import torch from diffusers import ScoreSdeVeScheduler class ScoreSdeVeSchedulerTest(unittest.TestCase): # TODO adapt with class SchedulerCommonTest (scheduler needs Numpy Integration) scheduler_classes = (ScoreSdeVeScheduler,) forward_default_kwargs = () @property def dummy_sample(self): batch_size = 4 num_channels = 3 height = 8 width = 8 sample = torch.rand((batch_size, num_channels, height, width)) return sample @property def dummy_sample_deter(self): batch_size = 4 num_channels = 3 height = 8 width = 8 num_elems = batch_size * num_channels * height * width sample = torch.arange(num_elems) sample = sample.reshape(num_channels, height, width, batch_size) sample = sample / num_elems sample = sample.permute(3, 0, 1, 2) return sample def dummy_model(self): def model(sample, t, *args): return sample * t / (t + 1) return model def get_scheduler_config(self, **kwargs): config = { "num_train_timesteps": 2000, "snr": 0.15, "sigma_min": 0.01, "sigma_max": 1348, "sampling_eps": 1e-5, } config.update(**kwargs) return config def check_over_configs(self, time_step=0, **config): kwargs = dict(self.forward_default_kwargs) for scheduler_class in self.scheduler_classes: sample = self.dummy_sample residual = 0.1 * sample scheduler_config = self.get_scheduler_config(**config) scheduler = scheduler_class(**scheduler_config) with tempfile.TemporaryDirectory() as tmpdirname: scheduler.save_config(tmpdirname) new_scheduler = scheduler_class.from_pretrained(tmpdirname) output = scheduler.step_pred( residual, time_step, sample, generator=torch.manual_seed(0), **kwargs ).prev_sample new_output = new_scheduler.step_pred( residual, time_step, sample, generator=torch.manual_seed(0), **kwargs ).prev_sample assert torch.sum(torch.abs(output - new_output)) < 1e-5, "Scheduler outputs are not identical" output = scheduler.step_correct(residual, sample, generator=torch.manual_seed(0), **kwargs).prev_sample new_output = new_scheduler.step_correct( residual, sample, generator=torch.manual_seed(0), **kwargs ).prev_sample assert torch.sum(torch.abs(output - new_output)) < 1e-5, "Scheduler correction are not identical" def check_over_forward(self, time_step=0, **forward_kwargs): kwargs = dict(self.forward_default_kwargs) kwargs.update(forward_kwargs) for scheduler_class in self.scheduler_classes: sample = self.dummy_sample residual = 0.1 * sample scheduler_config = self.get_scheduler_config() scheduler = scheduler_class(**scheduler_config) with tempfile.TemporaryDirectory() as tmpdirname: scheduler.save_config(tmpdirname) new_scheduler = scheduler_class.from_pretrained(tmpdirname) output = scheduler.step_pred( residual, time_step, sample, generator=torch.manual_seed(0), **kwargs ).prev_sample new_output = new_scheduler.step_pred( residual, time_step, sample, generator=torch.manual_seed(0), **kwargs ).prev_sample assert torch.sum(torch.abs(output - new_output)) < 1e-5, "Scheduler outputs are not identical" output = scheduler.step_correct(residual, sample, generator=torch.manual_seed(0), **kwargs).prev_sample new_output = new_scheduler.step_correct( residual, sample, generator=torch.manual_seed(0), **kwargs ).prev_sample assert torch.sum(torch.abs(output - new_output)) < 1e-5, "Scheduler correction are not identical" def test_timesteps(self): for timesteps in [10, 100, 1000]: self.check_over_configs(num_train_timesteps=timesteps) def test_sigmas(self): for sigma_min, sigma_max in zip([0.0001, 0.001, 0.01], [1, 100, 1000]): self.check_over_configs(sigma_min=sigma_min, sigma_max=sigma_max) def test_time_indices(self): for t in [0.1, 0.5, 0.75]: self.check_over_forward(time_step=t) def test_full_loop_no_noise(self): kwargs = dict(self.forward_default_kwargs) scheduler_class = self.scheduler_classes[0] scheduler_config = self.get_scheduler_config() scheduler = scheduler_class(**scheduler_config) num_inference_steps = 3 model = self.dummy_model() sample = self.dummy_sample_deter scheduler.set_sigmas(num_inference_steps) scheduler.set_timesteps(num_inference_steps) generator = torch.manual_seed(0) for i, t in enumerate(scheduler.timesteps): sigma_t = scheduler.sigmas[i] for _ in range(scheduler.config.correct_steps): with torch.no_grad(): model_output = model(sample, sigma_t) sample = scheduler.step_correct(model_output, sample, generator=generator, **kwargs).prev_sample with torch.no_grad(): model_output = model(sample, sigma_t) output = scheduler.step_pred(model_output, t, sample, generator=generator, **kwargs) sample, _ = output.prev_sample, output.prev_sample_mean result_sum = torch.sum(torch.abs(sample)) result_mean = torch.mean(torch.abs(sample)) assert np.isclose(result_sum.item(), 14372758528.0) assert np.isclose(result_mean.item(), 18714530.0) def test_step_shape(self): kwargs = dict(self.forward_default_kwargs) num_inference_steps = kwargs.pop("num_inference_steps", None) for scheduler_class in self.scheduler_classes: scheduler_config = self.get_scheduler_config() scheduler = scheduler_class(**scheduler_config) sample = self.dummy_sample residual = 0.1 * sample if num_inference_steps is not None and hasattr(scheduler, "set_timesteps"): scheduler.set_timesteps(num_inference_steps) elif num_inference_steps is not None and not hasattr(scheduler, "set_timesteps"): kwargs["num_inference_steps"] = num_inference_steps output_0 = scheduler.step_pred(residual, 0, sample, generator=torch.manual_seed(0), **kwargs).prev_sample output_1 = scheduler.step_pred(residual, 1, sample, generator=torch.manual_seed(0), **kwargs).prev_sample self.assertEqual(output_0.shape, sample.shape) self.assertEqual(output_0.shape, output_1.shape)

diffusers/tests/schedulers/test_scheduler_score_sde_ve.py/0

<jupyter_start><jupyter_text>Introduction to 🤗 Diffusers In this notebook, you'll train your first diffusion model to **generate images of cute butterflies 🦋.** Along the way, you'll learn about the core components of the 🤗 Diffusers library, which will provide a good foundation for the more advanced applications that we'll cover later in the course.Let's dive in! What You Will LearnIn this notebook you will:- See a powerful custom diffusion model pipeline in action (with information on how to make your own version)- Create your own mini pipeline by: - Recapping the core ideas behind diffusion models - Loading in data from the Hub for training - Exploring how we add noise to this data with a scheduler - Creating and training the UNet model - Putting the pieces together into a working pipeline- Edit and run a script for initializing longer training runs, that will handle - Multi-GPU training via 🤗 Accelerate - Experiment logging to track critical stats - Uploading the final model to the Hugging Face Hub❓If you have any questions, please post them on the `diffusion-models-class` channel on the Hugging Face Discord server. If you haven't signed up yet, you can do so here: https://huggingface.co/join/discord PrerequisitesBefore diving into the notebook, you should:* 📖 Read the Unit 1 materials* 🤗 Create an account on the Hugging Face Hub. If you haven't done so yet, you can do so here: https://huggingface.co/join Step 1: Setup Run the following cell to install the diffusers library as well as a few other requirements:<jupyter_code>%pip install -qq -U diffusers datasets transformers accelerate ftfy pyarrow==9.0.0<jupyter_output><empty_output><jupyter_text>Next, head over to https://huggingface.co/settings/tokens and create an access token with write permission if you don't already have one: You can login with this token using the command line (`huggingface-cli login`) or by running the following cell:<jupyter_code>from huggingface_hub import notebook_login notebook_login()<jupyter_output>Login successful Your token has been saved to /root/.huggingface/token<jupyter_text>Then you need to install Git-LFS to upload your model checkpoints:<jupyter_code>%%capture !sudo apt -qq install git-lfs !git config --global credential.helper store<jupyter_output><empty_output><jupyter_text>Finally, let's import the libraries we'll be using and define a few convenience functions which we'll use later in the notebook:<jupyter_code>import numpy as np import torch import torch.nn.functional as F from matplotlib import pyplot as plt from PIL import Image def show_images(x): """Given a batch of images x, make a grid and convert to PIL""" x = x * 0.5 + 0.5 # Map from (-1, 1) back to (0, 1) grid = torchvision.utils.make_grid(x) grid_im = grid.detach().cpu().permute(1, 2, 0).clip(0, 1) * 255 grid_im = Image.fromarray(np.array(grid_im).astype(np.uint8)) return grid_im def make_grid(images, size=64): """Given a list of PIL images, stack them together into a line for easy viewing""" output_im = Image.new("RGB", (size * len(images), size)) for i, im in enumerate(images): output_im.paste(im.resize((size, size)), (i * size, 0)) return output_im # Mac users may need device = 'mps' (untested) device = torch.device("cuda" if torch.cuda.is_available() else "cpu")<jupyter_output><empty_output><jupyter_text>OK, we're all set! Dreambooth: A Sneak Peak at What's to Come If you've looked at AI-related social media at all in the past few months, you've heard about Stable Diffusion. It's a powerful text-conditioned latent diffusion model (don't worry, we'll learn what all that means). But it has a flaw: it doesn't know what you or I look like unless we're famous enough to have our images plastered around the internet. Dreambooth let's us create our own model variant with some extra knowledge of a specific face, object or style. The Corridor Crew made an excellent video using this to tell stories with consistent characters, which is a great example of what this technique can do:<jupyter_code>from IPython.display import YouTubeVideo YouTubeVideo("W4Mcuh38wyM")<jupyter_output><empty_output><jupyter_text>Here's an example using [a model](https://huggingface.co/sd-dreambooth-library/mr-potato-head) trained on 5 photos of a popular children's toy called "Mr Potato Head".First, we load the pipeline. This will download model weights etc. from the Hub. Since this will download several gigabytes of data for a one-line demo, you are welcome to skip this cell and simply admire the example output!<jupyter_code>from diffusers import StableDiffusionPipeline # Check out https://huggingface.co/sd-dreambooth-library for loads of models from the community model_id = "sd-dreambooth-library/mr-potato-head" # Load the pipeline pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16).to( device )<jupyter_output><empty_output><jupyter_text>Once the pipeline has finished loading, we can generate images with:<jupyter_code>prompt = "an abstract oil painting of sks mr potato head by picasso" image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0] image<jupyter_output><empty_output><jupyter_text>**Exercise:** Try it yourself with different prompts. The `sks` token represents a unique identifier for the novel concept in this case - what happens if you leave that out? You can also experiment with changing the number of sampling steps (how low can you go?) and the `guidance_scale`, which determines how much the model will try to match the prompt. There's a lot going on in that magical pipeline! By the end of the course you'll know how it all works. For now, let's take a look at how we can train a diffusion model from scratch. MVP (Minimum Viable Pipeline)The core API of 🤗 Diffusers is divided into three main components:1. **Pipelines**: high-level classes designed to rapidly generate samples from popular trained diffusion models in a user-friendly fashion.2. **Models**: popular architectures for training new diffusion models, *e.g.* [UNet](https://arxiv.org/abs/1505.04597).3. **Schedulers**: various techniques for generating images from noise during *inference* as well as to generate noisy images for *training*.Pipelines are great for end-users, but if you're here for this course we assume you want to know what is going on under the hood! So, over the rest of this notebook we're going to build our own pipeline capable of generating small butterfly pictures. Here's the final result in action:<jupyter_code>from diffusers import DDPMPipeline # Load the butterfly pipeline butterfly_pipeline = DDPMPipeline.from_pretrained( "johnowhitaker/ddpm-butterflies-32px" ).to(device) # Create 8 images images = butterfly_pipeline(batch_size=8).images # View the result make_grid(images)<jupyter_output><empty_output><jupyter_text>Not as impressive as the DreamBooth example perhaps, but then we're training from scratch with ~0.0001% of the data used to train Stable Diffusion. Speaking of training, recall from the introduction to this unit that training a diffusion model looks something like this:1. Load in some images from the training data2. Add noise, in different amounts. 3. Feed the noisy versions of the inputs into the model4. Evaluate how well the model does at denoising these inputs5. Use this information to update the model weights, and repeatWe'll explore these steps one by one in the next few sections until we have a complete training loop working, and then we'll explore how to sample from the trained model and how to package everything up into a pipeline for easy sharing. Let's begin with the data... Step 2: Download a training datasetFor this example, we'll use a dataset of images from the Hugging Face Hub. Specifically, [this collection of 1000 butterfly pictures](https://huggingface.co/datasets/huggan/smithsonian_butterflies_subset). This is a very small dataset, so we've also included commented out lines for a few larger options. If you'd prefer to use your own collection of images, you can also use the commented-out code example to load in pictures from a folder instead.<jupyter_code>import torchvision from datasets import load_dataset from torchvision import transforms dataset = load_dataset("huggan/smithsonian_butterflies_subset", split="train") # Or load images from a local folder # dataset = load_dataset("imagefolder", data_dir="path/to/folder") # We'll train on 32-pixel square images, but you can try larger sizes too image_size = 32 # You can lower your batch size if you're running out of GPU memory batch_size = 64 # Define data augmentations preprocess = transforms.Compose( [ transforms.Resize((image_size, image_size)), # Resize transforms.RandomHorizontalFlip(), # Randomly flip (data augmentation) transforms.ToTensor(), # Convert to tensor (0, 1) transforms.Normalize([0.5], [0.5]), # Map to (-1, 1) ] ) def transform(examples): images = [preprocess(image.convert("RGB")) for image in examples["image"]] return {"images": images} dataset.set_transform(transform) # Create a dataloader from the dataset to serve up the transformed images in batches train_dataloader = torch.utils.data.DataLoader( dataset, batch_size=batch_size, shuffle=True )<jupyter_output>Using custom data configuration huggan--smithsonian_butterflies_subset-7665b1021a37404c Found cached dataset parquet (/home/lewis_huggingface_co/.cache/huggingface/datasets/huggan___parquet/huggan--smithsonian_butterflies_subset-7665b1021a37404c/0.0.0/2a3b91fbd88a2c90d1dbbb32b460cf621d31bd5b05b934492fdef7d8d6f236ec)<jupyter_text>We can grab a batch of images and view some of them like so:<jupyter_code>xb = next(iter(train_dataloader))["images"].to(device)[:8] print("X shape:", xb.shape) show_images(xb).resize((8 * 64, 64), resample=Image.NEAREST)<jupyter_output>X shape: torch.Size([8, 3, 32, 32])<jupyter_text>We're sticking to a small dataset with 32 pixel images to keep training times manageable in this notebook. Step 3: Define the SchedulerOur plan for training is to take these input images and add noise to them, then feed the noisy images to the model. And during inference, we will use the model predictions to iteratively remove noise. In `diffusers`, these processes are both handled by the **scheduler**. The noise schedule determines how much noise is added at different timesteps. Here's how we might create a scheduler using the default settings for 'DDPM' training and sampling (based on the paper ["Denoising Diffusion Probabalistic Models"](https://arxiv.org/abs/2006.11239):<jupyter_code>from diffusers import DDPMScheduler noise_scheduler = DDPMScheduler(num_train_timesteps=1000)<jupyter_output><empty_output><jupyter_text>The DDPM paper describes a corruption process that adds a small amount of noise for every 'timestep'. Given $x_{t-1}$ for some timestep, we can get the next (slightly more noisy) version $x_t$ with:$q(\mathbf{x}_t \vert \mathbf{x}_{t-1}) = \mathcal{N}(\mathbf{x}_t; \sqrt{1 - \beta_t} \mathbf{x}_{t-1}, \beta_t\mathbf{I}) \quadq(\mathbf{x}_{1:T} \vert \mathbf{x}_0) = \prod^T_{t=1} q(\mathbf{x}_t \vert \mathbf{x}_{t-1})$That is, we take $x_{t-1}$, scale it by $\sqrt{1 - \beta_t}$ and add noise scaled by $\beta_t$. This $\beta$ is defined for every t according to some schedule, and determines how much noise is added per timestep. Now, we don't necessarily want to do this operation 500 times to get $x_{500}$ so we have another formula to get $x_t$ for any t given $x_0$: $\begin{aligned}q(\mathbf{x}_t \vert \mathbf{x}_0) &= \mathcal{N}(\mathbf{x}_t; \sqrt{\bar{\alpha}_t} \mathbf{x}_0, {(1 - \bar{\alpha}_t)} \mathbf{I})\end{aligned}$ where $\bar{\alpha}_t = \prod_{i=1}^T \alpha_i$ and $\alpha_i = 1-\beta_i$The maths notation always looks scary! Luckily the scheduler handles all that for us. We can plot $\sqrt{\bar{\alpha}_t}$ (labelled as `sqrt_alpha_prod`) and $\sqrt{(1 - \bar{\alpha}_t)}$ (labelled as `sqrt_one_minus_alpha_prod`) to view how the input (x) and the noise are scaled and mixed across different timesteps:<jupyter_code>plt.plot(noise_scheduler.alphas_cumprod.cpu() ** 0.5, label=r"${\sqrt{\bar{\alpha}_t}}$") plt.plot((1 - noise_scheduler.alphas_cumprod.cpu()) ** 0.5, label=r"$\sqrt{(1 - \bar{\alpha}_t)}$") plt.legend(fontsize="x-large");<jupyter_output><empty_output><jupyter_text>**Exercise:** You can explore how this plot changes with different settings for beta_start, beta_end and beta_schedule by swapping in one of the commented-out options here:<jupyter_code># One with too little noise added: # noise_scheduler = DDPMScheduler(num_train_timesteps=1000, beta_start=0.001, beta_end=0.004) # The 'cosine' schedule, which may be better for small image sizes: # noise_scheduler = DDPMScheduler(num_train_timesteps=1000, beta_schedule='squaredcos_cap_v2')<jupyter_output><empty_output><jupyter_text>Whichever scheduler you've chosen, we can now use it to add noise in different amounts using the `noise_scheduler.add_noise` function like so:<jupyter_code>timesteps = torch.linspace(0, 999, 8).long().to(device) noise = torch.randn_like(xb) noisy_xb = noise_scheduler.add_noise(xb, noise, timesteps) print("Noisy X shape", noisy_xb.shape) show_images(noisy_xb).resize((8 * 64, 64), resample=Image.NEAREST)<jupyter_output>Noisy X shape torch.Size([8, 3, 32, 32])<jupyter_text>Again, explore the effect of using different noise schedules and parameters here. [This video](https://www.youtube.com/watch?v=fbLgFrlTnGU) does a great job explaining some of the maths above in more detail, and is a great introduction to some of these concepts. Step 4: Define the Model Now we come to the core component: the model itself. Most diffusion models use architectures that are some variant of a [U-net](https://arxiv.org/abs/1505.04597) and that's what we'll use here.In a nutshell:- the model has the input image go through several blocks of ResNet layers, each of which halves the image size by 2- then through the same number of blocks that upsample it again.- there are skip connections linking the features on the downsample path to the corresponding layers in the upsample path.A key feature of this model is that it predicts images of the same size as the input, which is exactly what we need here.Diffusers provides us a handy `UNet2DModel` class which creates the desired architecture in PyTorch.Let's create a U-net for our desired image size. Note that `down_block_types` correspond to the downsampling blocks (green on the diagram above), and `up_block_types` are the upsampling blocks (red on the diagram):<jupyter_code>from diffusers import UNet2DModel # Create a model model = UNet2DModel( sample_size=image_size, # the target image resolution in_channels=3, # the number of input channels, 3 for RGB images out_channels=3, # the number of output channels layers_per_block=2, # how many ResNet layers to use per UNet block block_out_channels=(64, 128, 128, 256), # More channels -> more parameters down_block_types=( "DownBlock2D", # a regular ResNet downsampling block "DownBlock2D", "AttnDownBlock2D", # a ResNet downsampling block with spatial self-attention "AttnDownBlock2D", ), up_block_types=( "AttnUpBlock2D", "AttnUpBlock2D", # a ResNet upsampling block with spatial self-attention "UpBlock2D", "UpBlock2D", # a regular ResNet upsampling block ), ) model.to(device);<jupyter_output><empty_output><jupyter_text>When dealing with higher-resolution inputs you may want to use more down and up-blocks, and keep the attention layers only at the lowest resolution (bottom) layers to reduce memory usage. We'll talk later about how you might experiment to find the best settings for your use-case. We can check that passing in a batch of data and some random timesteps produces an output the same shape as the input data:<jupyter_code>with torch.no_grad(): model_prediction = model(noisy_xb, timesteps).sample model_prediction.shape<jupyter_output><empty_output><jupyter_text>In the next section we'll see how to train this model. Step 5: Create a Training Loop Time to train! Below is a typical optimization loop in PyTorch, where we run through the data batch by batch and update the parameters of our model each step using an optimizer - in this case the AdamW optimizer with a learning rate of 0.0004. For each batch of data, we- Sample some random timesteps- Noise the data accordingly- Feed the noisy data through the model- Compare the model predictions with the target (i.e. the noise in this case) using mean squared error as our loss function- Update the model parameters via `loss.backward()` and `optimizer.step()`During this process we also log the losses over time for later plotting.NB: This code takes nearly 10 minutes to run - feel free to skip these two cells and use the pretrained model if you are in a hurry. Alternatively, you can explore how reducing the number of channels in each layer via the model definition above can speed things up. The [official diffusers training example](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/training_example.ipynb) trains a larger model on this dataset at higher resolution, and is a good reference for what a less minimal training loop looks like:<jupyter_code># Set the noise scheduler noise_scheduler = DDPMScheduler( num_train_timesteps=1000, beta_schedule="squaredcos_cap_v2" ) # Training loop optimizer = torch.optim.AdamW(model.parameters(), lr=4e-4) losses = [] for epoch in range(30): for step, batch in enumerate(train_dataloader): clean_images = batch["images"].to(device) # Sample noise to add to the images noise = torch.randn(clean_images.shape).to(clean_images.device) bs = clean_images.shape[0] # Sample a random timestep for each image timesteps = torch.randint( 0, noise_scheduler.num_train_timesteps, (bs,), device=clean_images.device ).long() # Add noise to the clean images according to the noise magnitude at each timestep noisy_images = noise_scheduler.add_noise(clean_images, noise, timesteps) # Get the model prediction noise_pred = model(noisy_images, timesteps, return_dict=False)[0] # Calculate the loss loss = F.mse_loss(noise_pred, noise) loss.backward(loss) losses.append(loss.item()) # Update the model parameters with the optimizer optimizer.step() optimizer.zero_grad() if (epoch + 1) % 5 == 0: loss_last_epoch = sum(losses[-len(train_dataloader) :]) / len(train_dataloader) print(f"Epoch:{epoch+1}, loss: {loss_last_epoch}")<jupyter_output>Epoch:5, loss: 0.16273280512541533 Epoch:10, loss: 0.11161588924005628 Epoch:15, loss: 0.10206522420048714 Epoch:20, loss: 0.08302505919709802 Epoch:25, loss: 0.07805309211835265 Epoch:30, loss: 0.07474562455900013<jupyter_text>Plotting the loss, we see that the model rapidly improves initially and then continues to get better at a slower rate (which is more obvious if we use a log scale as shown on the right):<jupyter_code>fig, axs = plt.subplots(1, 2, figsize=(12, 4)) axs[0].plot(losses) axs[1].plot(np.log(losses)) plt.show()<jupyter_output><empty_output><jupyter_text>As an alternative to running the training code above, you can use the model from the pipeline like so:<jupyter_code># Uncomment to instead load the model I trained earlier: # model = butterfly_pipeline.unet<jupyter_output><empty_output><jupyter_text>Step 6: Generate ImagesHow do we get images with this model? Option 1: Creating a pipeline:<jupyter_code>from diffusers import DDPMPipeline image_pipe = DDPMPipeline(unet=model, scheduler=noise_scheduler) pipeline_output = image_pipe() pipeline_output.images[0]<jupyter_output><empty_output><jupyter_text>We can save a pipeline to a local folder like so:<jupyter_code>image_pipe.save_pretrained("my_pipeline")<jupyter_output><empty_output><jupyter_text>Inspecting the folder contents:<jupyter_code>!ls my_pipeline/<jupyter_output>model_index.json scheduler unet<jupyter_text>The `scheduler` and `unet` subfolders contain everything needed to re-create those components. For example, inside the `unet` folder you'll find the model weights (`diffusion_pytorch_model.bin`) alongside a config file which specifies the UNet architecture.<jupyter_code>!ls my_pipeline/unet/<jupyter_output>config.json diffusion_pytorch_model.bin<jupyter_text>Together, these files contain everything needed to recreate the pipeline. You can manually upload them to the hub to share the pipeline with others, or check out the code to do this via the API in the next section. Option 2: Writing a Sampling LoopIf you inspect the forward method of the pipeline you'll be able to see what is happening when we run `image_pipe()`:<jupyter_code># ??image_pipe.forward<jupyter_output><empty_output><jupyter_text>We begin with random noise, and run through the scheduler timesteps from most to least noisy, removing a small amount of noise each step based on the model prediction:<jupyter_code># Random starting point (8 random images): sample = torch.randn(8, 3, 32, 32).to(device) for i, t in enumerate(noise_scheduler.timesteps): # Get model pred with torch.no_grad(): residual = model(sample, t).sample # Update sample with step sample = noise_scheduler.step(residual, t, sample).prev_sample show_images(sample)<jupyter_output><empty_output><jupyter_text>The `noise_scheduler.step()` function does the maths required to update `sample` appropriately. There are a number of sampling methods - in the next unit we'll see how we can swap in a different sampler to speed up image generation with existing models, and talk more about the theory behind sampling from diffusion models. Step 7: Push your model to the HubIn the example above we saved our pipeline to a local folder. To push our model to the Hub, we will need to model repository to push our files to. We'll determine the repository name from the model ID we want to give our model (feel free to replace the `model_name` with your own choice; it just needs to contain your username, which is what the function `get_full_repo_name()` does):<jupyter_code>from huggingface_hub import get_full_repo_name model_name = "sd-class-butterflies-32" hub_model_id = get_full_repo_name(model_name) hub_model_id<jupyter_output><empty_output><jupyter_text>Next, create a model repository on the 🤗 Hub and push our model:<jupyter_code>from huggingface_hub import HfApi, create_repo create_repo(hub_model_id) api = HfApi() api.upload_folder( folder_path="my_pipeline/scheduler", path_in_repo="", repo_id=hub_model_id ) api.upload_folder(folder_path="my_pipeline/unet", path_in_repo="", repo_id=hub_model_id) api.upload_file( path_or_fileobj="my_pipeline/model_index.json", path_in_repo="model_index.json", repo_id=hub_model_id, )<jupyter_output><empty_output><jupyter_text>The last thing to do is create a nice model card so that our butterfly generator can easily be found on the Hub (feel free to expand and edit the description!):<jupyter_code>from huggingface_hub import ModelCard content = f""" --- license: mit tags: - pytorch - diffusers - unconditional-image-generation - diffusion-models-class --- # Model Card for Unit 1 of the [Diffusion Models Class 🧨](https://github.com/huggingface/diffusion-models-class) This model is a diffusion model for unconditional image generation of cute 🦋. ## Usage ```python from diffusers import DDPMPipeline pipeline = DDPMPipeline.from_pretrained('{hub_model_id}') image = pipeline().images[0] image ``` """ card = ModelCard(content) card.push_to_hub(hub_model_id)<jupyter_output><empty_output><jupyter_text>Now that the model is on the Hub, you can download it from anywhere by using the `from_pretrained()` method of the `DDPMPipeline` as follows"<jupyter_code>from diffusers import DDPMPipeline image_pipe = DDPMPipeline.from_pretrained(hub_model_id) pipeline_output = image_pipe() pipeline_output.images[0]<jupyter_output><empty_output><jupyter_text>Great it works! Scaling up with 🤗 AccelerateThis notebook was made for learning purposes, and as such I tried to keep the code as minimal and clean as possible. Because of this, we omitted some of the things you might want if you were to try training a larger model on much more data, such as multi-GPU support, logging of progress and example images, gradient checkpointing to support larger batch sizes, automatic uploading of models and so on. Thankfully most of these features are available in the example training script [here](https://github.com/huggingface/diffusers/raw/main/examples/unconditional_image_generation/train_unconditional.py).You can download the file like so:<jupyter_code>!wget https://github.com/huggingface/diffusers/raw/main/examples/unconditional_image_generation/train_unconditional.py<jupyter_output><empty_output><jupyter_text>Open up the file and you'll see where the model is defined and what settings are available. I ran the script with the following command:<jupyter_code># Let's give our new model a name for the Hub model_name = "sd-class-butterflies-64" hub_model_id = get_full_repo_name(model_name) hub_model_id !accelerate launch train_unconditional.py \ --dataset_name="huggan/smithsonian_butterflies_subset" \ --resolution=64 \ --output_dir={model_name} \ --train_batch_size=32 \ --num_epochs=50 \ --gradient_accumulation_steps=1 \ --learning_rate=1e-4 \ --lr_warmup_steps=500 \ --mixed_precision="no"<jupyter_output><empty_output><jupyter_text>As before, let's push the model to the Hub and create a nice model card (and feel free to edit it as you wish!):<jupyter_code>create_repo(hub_model_id) api = HfApi() api.upload_folder( folder_path=f"{model_name}/scheduler", path_in_repo="", repo_id=hub_model_id ) api.upload_folder( folder_path=f"{model_name}/unet", path_in_repo="", repo_id=hub_model_id ) api.upload_file( path_or_fileobj=f"{model_name}/model_index.json", path_in_repo="model_index.json", repo_id=hub_model_id, ) content = f""" --- license: mit tags: - pytorch - diffusers - unconditional-image-generation - diffusion-models-class --- # Model Card for Unit 1 of the [Diffusion Models Class 🧨](https://github.com/huggingface/diffusion-models-class) This model is a diffusion model for unconditional image generation of cute 🦋. ## Usage ```python from diffusers import DDPMPipeline pipeline = DDPMPipeline.from_pretrained('{hub_model_id}') image = pipeline().images[0] image ``` """ card = ModelCard(content) card.push_to_hub(hub_model_id)<jupyter_output><empty_output><jupyter_text>About 45 minutes later, this is the result:<jupyter_code>pipeline = DDPMPipeline.from_pretrained(hub_model_id).to(device) images = pipeline(batch_size=8).images make_grid(images)<jupyter_output><empty_output>

diffusion-models-class/units/en/unit1/introduction_to_diffusers.ipynb/0

<jupyter_start><jupyter_text>Stable Diffusion Deep DiveStable Diffusion is a powerful text-to-image model. There are various websites and tools to make using it as easy as possible. It is also [integrated into the Huggingface diffusers library](https://huggingface.co/blog/stable_diffusion) where generating images can be as simple as:```pythonfrom diffusers import StableDiffusionPipelinepipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", revision="fp16", torch_dtype=torch.float16, use_auth_token=True).to("cuda")image = pipe("An astronaught scuba diving").images[0]```In this notebook we're going to dig into the code behind these easy-to-use interfaces, to see what is going on under the hood. We'll begin by re-creating the functionality above as a scary chunk of code, and then one by one we'll inspect the different components and figure out what they do. By the end of this notebook that same sampling loop should feel like something you can tweak and modify as you like. Setup & ImportsYou'll need to log into huggingface and accept the terms of the licence for this model - see the [model card](https://huggingface.co/CompVis/stable-diffusion-v1-4) for details. And when you first run this notebook you need to uncomment the following two cells to install the requirements and log in to huggingface with an access token.<jupyter_code># !pip install -q --upgrade transformers diffusers ftfy from base64 import b64encode import numpy import torch from diffusers import AutoencoderKL, LMSDiscreteScheduler, UNet2DConditionModel from huggingface_hub import notebook_login # For video display: from IPython.display import HTML from matplotlib import pyplot as plt from pathlib import Path from PIL import Image from torch import autocast from torchvision import transforms as tfms from tqdm.auto import tqdm from transformers import CLIPTextModel, CLIPTokenizer, logging torch.manual_seed(1) if not (Path.home()/'.huggingface'/'token').exists(): notebook_login() # Supress some unnecessary warnings when loading the CLIPTextModel logging.set_verbosity_error() # Set device torch_device = "cuda" if torch.cuda.is_available() else "cpu"<jupyter_output><empty_output><jupyter_text>Loading the modelsThis code (and that in the next section) comes from the [Huggingface example notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/stable_diffusion.ipynb).This will download and set up the relevant models and components we'll be using. Let's just run this for now and move on to the next section to check that it all works before diving deeper.If you've loaded a pipeline, you can also access these components using `pipe.unet`, `pipe.vae` and so on.In this notebook we aren't doing any memory-saving tricks - if you find yourself running out of GPU RAM, look at the pipeline code for inspiration with things like attention slicing, switching to half precision (fp16), keeping the VAE on the CPU and other modifications.<jupyter_code># Load the autoencoder model which will be used to decode the latents into image space. vae = AutoencoderKL.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="vae") # Load the tokenizer and text encoder to tokenize and encode the text. tokenizer = CLIPTokenizer.from_pretrained("openai/clip-vit-large-patch14") text_encoder = CLIPTextModel.from_pretrained("openai/clip-vit-large-patch14") # The UNet model for generating the latents. unet = UNet2DConditionModel.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="unet") # The noise scheduler scheduler = LMSDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000) # To the GPU we go! vae = vae.to(torch_device) text_encoder = text_encoder.to(torch_device) unet = unet.to(torch_device);<jupyter_output><empty_output><jupyter_text>A diffusion loopIf all you want is to make a picture with some text, you could ignore this notebook and use one of the existing tools (such as [DreamStudio](https://beta.dreamstudio.ai/)) or use the simplified pipeline from huggingface, as documented [here](https://huggingface.co/blog/stable_diffusion).What we want to do in this notebook is dig a little deeper into how this works, so we'll start by checking that the example code runs. Again, this is adapted from the [HF notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/stable_diffusion.ipynb) and looks very similar to what you'll find if you inspect [the `__call__()` method of the stable diffusion pipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.pyL200).<jupyter_code># Some settings prompt = ["A watercolor painting of an otter"] height = 512 # default height of Stable Diffusion width = 512 # default width of Stable Diffusion num_inference_steps = 30 # Number of denoising steps guidance_scale = 7.5 # Scale for classifier-free guidance generator = torch.manual_seed(32) # Seed generator to create the inital latent noise batch_size = 1 # Prep text text_input = tokenizer(prompt, padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt") with torch.no_grad(): text_embeddings = text_encoder(text_input.input_ids.to(torch_device))[0] max_length = text_input.input_ids.shape[-1] uncond_input = tokenizer( [""] * batch_size, padding="max_length", max_length=max_length, return_tensors="pt" ) with torch.no_grad(): uncond_embeddings = text_encoder(uncond_input.input_ids.to(torch_device))[0] text_embeddings = torch.cat([uncond_embeddings, text_embeddings]) # Prep Scheduler scheduler.set_timesteps(num_inference_steps) # Prep latents latents = torch.randn( (batch_size, unet.in_channels, height // 8, width // 8), generator=generator, ) latents = latents.to(torch_device) latents = latents * scheduler.init_noise_sigma # Scaling (previous versions did latents = latents * self.scheduler.sigmas[0] # Loop with autocast("cuda"): for i, t in tqdm(enumerate(scheduler.timesteps)): # expand the latents if we are doing classifier-free guidance to avoid doing two forward passes. latent_model_input = torch.cat([latents] * 2) sigma = scheduler.sigmas[i] # Scale the latents (preconditioning): # latent_model_input = latent_model_input / ((sigma**2 + 1) ** 0.5) # Diffusers 0.3 and below latent_model_input = scheduler.scale_model_input(latent_model_input, t) # predict the noise residual with torch.no_grad(): noise_pred = unet(latent_model_input, t, encoder_hidden_states=text_embeddings).sample # perform guidance noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond) # compute the previous noisy sample x_t -> x_t-1 # latents = scheduler.step(noise_pred, i, latents)["prev_sample"] # Diffusers 0.3 and below latents = scheduler.step(noise_pred, t, latents).prev_sample # scale and decode the image latents with vae latents = 1 / 0.18215 * latents with torch.no_grad(): image = vae.decode(latents).sample # Display image = (image / 2 + 0.5).clamp(0, 1) image = image.detach().cpu().permute(0, 2, 3, 1).numpy() images = (image * 255).round().astype("uint8") pil_images = [Image.fromarray(image) for image in images] pil_images[0]<jupyter_output><empty_output><jupyter_text>It's working, but that's quite a bit of code! Let's look at the components one by one. The Autoencoder (AE)The AE can 'encode' an image into some sort of latent representation, and decode this back into an image. I've wrapped the code for this into a couple of functions here so we can see what this looks like in action:<jupyter_code>def pil_to_latent(input_im): # Single image -> single latent in a batch (so size 1, 4, 64, 64) with torch.no_grad(): latent = vae.encode(tfms.ToTensor()(input_im).unsqueeze(0).to(torch_device)*2-1) # Note scaling return 0.18215 * latent.latent_dist.sample() def latents_to_pil(latents): # bath of latents -> list of images latents = (1 / 0.18215) * latents with torch.no_grad(): image = vae.decode(latents).sample image = (image / 2 + 0.5).clamp(0, 1) image = image.detach().cpu().permute(0, 2, 3, 1).numpy() images = (image * 255).round().astype("uint8") pil_images = [Image.fromarray(image) for image in images] return pil_images<jupyter_output><empty_output><jupyter_text>We'll use a pic from the web here, but you can load your own instead by uploading it and editing the filename in the next cell.<jupyter_code># Download a demo Image !curl --output macaw.jpg 'https://lafeber.com/pet-birds/wp-content/uploads/2018/06/Scarlet-Macaw-2.jpg' # Load the image with PIL input_image = Image.open('macaw.jpg').resize((512, 512)) input_image<jupyter_output><empty_output><jupyter_text>Encoding this into the latent space of the AE with the function defined above looks like this:<jupyter_code># Encode to the latent space encoded = pil_to_latent(input_image) encoded.shape # Let's visualize the four channels of this latent representation: fig, axs = plt.subplots(1, 4, figsize=(16, 4)) for c in range(4): axs[c].imshow(encoded[0][c].cpu(), cmap='Greys')<jupyter_output><empty_output><jupyter_text>This 4x64x64 tensor captures lots of information about the image, hopefully enough that when we feed it through the decoder we get back something very close to our input image:<jupyter_code># Decode this latent representation back into an image decoded = latents_to_pil(encoded)[0] decoded<jupyter_output><empty_output><jupyter_text>You'll see some small differences if you squint! Forcus on the eye if you can't see anything obvious. This is pretty impressive - that 4x64x64 latent seems to hold a lot more information that a 64px image...This autoencoder has been trained to squish down an image to a smaller representation and then re-create the image back from this compressed version again.In this particular case the compression factor is 48, we start with a 3x512x512(chxhtxwd) image and it get compressed to a latent vector 4x64x64. Each 3x8x8 pixel volume in the input image gets compressed down to just 4 numbers(4x1x1). You can find AEs with a higher compression ratio (eg f16 like some popular VQGAN models) but at some point they begin to introduce artifacts that we don't want.Why do we even use an autoencoder? We can do diffusion in pixel space - where the model gets all the image data as inputs and produces an output prediction of the same shape. But this means processing a LOT of data, and make high-resolution generation very computationally expensive. Some solutions to this involve doing diffusion at low resolution (64px for eg) and then training a separate model to upscale repeatedly (as with D2/Imagen). But latent diffusion instead does the diffusion process in this 'latent space', using the compressed representations from our AE rather than raw images. These representations are information rich, and can be small enough to handle manageably on consumer hardware. Once we've generated a new 'image' as a latent representation, the autoencoder can take those final latent outputs and turn them into actual pixels. The SchedulerNow we need to talk about adding noise...During training, we add some noise to an image an then have the model try to predict the noise. If we always added a ton of noise, the model might not have much to work with. If we only add a tiny amount, the model won't be able to do much with the random starting points we use for sampling. So during training the amount is varied, according to some distribution.During sampling, we want to 'denoise' over a number of steps. How many steps and how much noise we should aim for at each step are going to affect the final result.The scheduler is in charge of handling all of these details. For example: `scheduler = LMSDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000)` sets up a scheduler that matches the one used to train this model. When we want to sample over a smaller number of steps, we set this up with `scheduler.set_timesteps`:<jupyter_code># Setting the number of sampling steps: scheduler.set_timesteps(15)<jupyter_output><empty_output><jupyter_text>You can see how our new set of steps corresponds to those used in training:<jupyter_code># See these in terms of the original 1000 steps used for training: print(scheduler.timesteps)<jupyter_output>tensor([999.0000, 927.6429, 856.2857, 784.9286, 713.5714, 642.2143, 570.8571, 499.5000, 428.1429, 356.7857, 285.4286, 214.0714, 142.7143, 71.3571, 0.0000], dtype=torch.float64)<jupyter_text>And how much noise is present at each:<jupyter_code># Look at the equivalent noise levels: print(scheduler.sigmas)<jupyter_output>tensor([14.6146, 9.6826, 6.6780, 4.7746, 3.5221, 2.6666, 2.0606, 1.6156, 1.2768, 1.0097, 0.7913, 0.6056, 0.4397, 0.2780, 0.0292, 0.0000])<jupyter_text>During sampling, we'll start at a high noise level (in fact, our input will be pure noise) and gradually 'denoise' down to an image, according to this schedule.<jupyter_code># Plotting this noise schedule: plt.plot(scheduler.sigmas) plt.title('Noise Schedule') plt.xlabel('Sampling step') plt.ylabel('sigma') plt.show() # TODO maybe show timestep as well<jupyter_output><empty_output><jupyter_text>This 'sigma' is the amount of noise added to the latent representation. Let's visualize what this looks like by adding a bit of noise to our encoded image and then decoding this noised version:<jupyter_code>noise = torch.randn_like(encoded) # Random noise sampling_step = 10 # Equivalent to step 10 out of 15 in the schedule above # encoded_and_noised = scheduler.add_noise(encoded, noise, timestep) # Diffusers 0.3 and below encoded_and_noised = scheduler.add_noise(encoded, noise, timesteps=torch.tensor([scheduler.timesteps[sampling_step]])) latents_to_pil(encoded_and_noised.float())[0] # Display<jupyter_output><empty_output><jupyter_text>What does this look like at different timesteps? Experiment and see for yourself!If you uncomment the cell below you'll see that in this case the `scheduler.add_noise` function literally just adds noise scaled by sigma: `noisy_samples = original_samples + noise * sigmas`<jupyter_code># ??scheduler.add_noise<jupyter_output><empty_output><jupyter_text>Other diffusion models may be trained with different noising and scheduling approaches, some of which keep the variance fairly constant across noise levels ('variance preserving') with different scaling and mixing tricks instead of having noisy latents with higher and higher variance as more noise is added ('variance exploding').If we want to start from random noise instead of a noised image, we need to scale it by the largest sigma value used during training, ~14 in this case. And before these noisy latents are fed to the model they are scaled again in the so-called pre-conditioning step:`latent_model_input = latent_model_input / ((sigma**2 + 1) ** 0.5)` (now handled by `latent_model_input = scheduler.scale_model_input(latent_model_input, t)`).Again, this scaling/pre-conditioning differs between papers and implementations, so keep an eye out for this if you work with a different type of diffusion model. Loop starting from noised version of input (AKA image2image)Let's see what happens when we use our image as a starting point, adding some noise and then doing the final few denoising steps in the loop with a new prompt.We'll use a similar loop to the first demo, but we'll skip the first `start_step` steps.To noise our image we'll use code like that shown above, using the scheduler to noise it to a level equivalent to step 10 (`start_step`).<jupyter_code># Settings (same as before except for the new prompt) prompt = ["A colorful dancer, nat geo photo"] height = 512 # default height of Stable Diffusion width = 512 # default width of Stable Diffusion num_inference_steps = 50 # Number of denoising steps guidance_scale = 8 # Scale for classifier-free guidance generator = torch.manual_seed(32) # Seed generator to create the inital latent noise batch_size = 1 # Prep text (same as before) text_input = tokenizer(prompt, padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt") with torch.no_grad(): text_embeddings = text_encoder(text_input.input_ids.to(torch_device))[0] max_length = text_input.input_ids.shape[-1] uncond_input = tokenizer( [""] * batch_size, padding="max_length", max_length=max_length, return_tensors="pt" ) with torch.no_grad(): uncond_embeddings = text_encoder(uncond_input.input_ids.to(torch_device))[0] text_embeddings = torch.cat([uncond_embeddings, text_embeddings]) # Prep Scheduler (setting the number of inference steps) scheduler.set_timesteps(num_inference_steps) # Prep latents (noising appropriately for start_step) start_step = 10 start_sigma = scheduler.sigmas[start_step] noise = torch.randn_like(encoded) latents = scheduler.add_noise(encoded, noise, timesteps=torch.tensor([scheduler.timesteps[start_step]])) latents = latents.to(torch_device).float() # Loop for i, t in tqdm(enumerate(scheduler.timesteps)): if i >= start_step: # << This is the only modification to the loop we do # expand the latents if we are doing classifier-free guidance to avoid doing two forward passes. latent_model_input = torch.cat([latents] * 2) sigma = scheduler.sigmas[i] latent_model_input = scheduler.scale_model_input(latent_model_input, t) # predict the noise residual with torch.no_grad(): noise_pred = unet(latent_model_input, t, encoder_hidden_states=text_embeddings)["sample"] # perform guidance noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond) # compute the previous noisy sample x_t -> x_t-1 latents = scheduler.step(noise_pred, t, latents).prev_sample latents_to_pil(latents)[0]<jupyter_output><empty_output><jupyter_text>You can see that some colours and structure from the image are kept, but we now have a new picture! The more noise you add and the more steps you do, the further away it gets from the input image.This is how the popular img2img pipeline works. Again, if this is your end goal there are tools to make this easy!But you can see that under the hood this is the same as the generation loop just skipping the first few steps and starting from a noised image rather than pure noise.Explore changing how many steps are skipped and see how this affects the amount the image changes from the input. Exploring the text -> embedding pipelineWe use a text encoder model to turn our text into a set of 'embeddings' which are fed to the diffusion model as conditioning. Let's follow a piece of text through this process and see how it works.<jupyter_code># Our text prompt prompt = 'A picture of a puppy'<jupyter_output><empty_output><jupyter_text>We begin with tokenization:<jupyter_code># Turn the text into a sequnce of tokens: text_input = tokenizer(prompt, padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt") text_input['input_ids'][0] # View the tokens # See the individual tokens for t in text_input['input_ids'][0][:8]: # We'll just look at the first 7 to save you from a wall of '<|endoftext|>' print(t, tokenizer.decoder.get(int(t))) # TODO call out that 6829 is puppy<jupyter_output><empty_output><jupyter_text>We can jump straight to the final (output) embeddings like so:<jupyter_code># Grab the output embeddings output_embeddings = text_encoder(text_input.input_ids.to(torch_device))[0] print('Shape:', output_embeddings.shape) output_embeddings<jupyter_output>Shape: torch.Size([1, 77, 768])<jupyter_text>We pass our tokens through the text_encoder and we magically get some numbers we can feed to the model.How are these generated? The tokens are transformed into a set of input embeddings, which are then fed through the transformer model to get the final output embeddings.To get these input embeddings, there are actually two steps - as revealed by inspecting `text_encoder.text_model.embeddings`:<jupyter_code>text_encoder.text_model.embeddings<jupyter_output><empty_output><jupyter_text>Token embeddingsThe token is fed to the `token_embedding` to transform it into a vector. The function name `get_input_embeddings` here is misleading since these token embeddings need to be combined with the position embeddings before they are actually used as inputs to the model! Anyway, let's look at just the token embedding part firstWe can look at the embedding layer:<jupyter_code># Access the embedding layer token_emb_layer = text_encoder.text_model.embeddings.token_embedding token_emb_layer # Vocab size 49408, emb_dim 768<jupyter_output><empty_output><jupyter_text>And embed a token like so:<jupyter_code># Embed a token - in this case the one for 'puppy' embedding = token_emb_layer(torch.tensor(6829, device=torch_device)) embedding.shape # 768-dim representation<jupyter_output><empty_output><jupyter_text>This single token has been mapped to a 768-dimensional vector - the token embedding.We can do the same with all of the tokens in the prompt to get all the token embeddings:<jupyter_code>token_embeddings = token_emb_layer(text_input.input_ids.to(torch_device)) print(token_embeddings.shape) # batch size 1, 77 tokens, 768 values for each token_embeddings<jupyter_output>torch.Size([1, 77, 768])<jupyter_text>Positional EmbeddingsPositional embeddings tell the model where in a sequence a token is. Much like the token embedding, this is a set of (optionally learnable) parameters. But now instead of dealing with ~50k tokens we just need one for each position (77 total):<jupyter_code>pos_emb_layer = text_encoder.text_model.embeddings.position_embedding pos_emb_layer<jupyter_output><empty_output><jupyter_text>We can get the positional embedding for each position:<jupyter_code>position_ids = text_encoder.text_model.embeddings.position_ids[:, :77] position_embeddings = pos_emb_layer(position_ids) print(position_embeddings.shape) position_embeddings<jupyter_output>torch.Size([1, 77, 768])<jupyter_text>Combining token and position embeddingsTime to combine the two. How do we do this? Just add them! Other approaches are possible but for this model this is how it is done.Combining them in this way gives us the final input embeddings ready to feed through the transformer model:<jupyter_code># And combining them we get the final input embeddings input_embeddings = token_embeddings + position_embeddings print(input_embeddings.shape) input_embeddings<jupyter_output>torch.Size([1, 77, 768])<jupyter_text>We can check that these are the same as the result we'd get from `text_encoder.text_model.embeddings`:<jupyter_code># The following combines all the above steps (but doesn't let us fiddle with them!) text_encoder.text_model.embeddings(text_input.input_ids.to(torch_device))<jupyter_output><empty_output><jupyter_text>Feeding these through the transformer model We want to mess with these input embeddings (specifically the token embeddings) before we send them through the rest of the model, but first we should check that we know how to do that. I read the code of the text_encoders `forward` method, and based on that the code for the `forward` method of the text_model that the text_encoder wraps. To inspect it yourself, type `??text_encoder.text_model.forward` and you'll get the function info and source code - a useful debugging trick!Anyway, based on that we can copy in the bits we need to get the so-called 'last hidden state' and thus generate our final embeddings:<jupyter_code>def get_output_embeds(input_embeddings): # CLIP's text model uses causal mask, so we prepare it here: bsz, seq_len = input_embeddings.shape[:2] causal_attention_mask = text_encoder.text_model._build_causal_attention_mask(bsz, seq_len, dtype=input_embeddings.dtype) # Getting the output embeddings involves calling the model with passing output_hidden_states=True # so that it doesn't just return the pooled final predictions: encoder_outputs = text_encoder.text_model.encoder( inputs_embeds=input_embeddings, attention_mask=None, # We aren't using an attention mask so that can be None causal_attention_mask=causal_attention_mask.to(torch_device), output_attentions=None, output_hidden_states=True, # We want the output embs not the final output return_dict=None, ) # We're interested in the output hidden state only output = encoder_outputs[0] # There is a final layer norm we need to pass these through output = text_encoder.text_model.final_layer_norm(output) # And now they're ready! return output out_embs_test = get_output_embeds(input_embeddings) # Feed through the model with our new function print(out_embs_test.shape) # Check the output shape out_embs_test # Inspect the output<jupyter_output>torch.Size([1, 77, 768])<jupyter_text>Note that these match the `output_embeddings` we saw near the start - we've figured out how to split up that one step ("get the text embeddings") into multiple sub-steps ready for us to modify.Now that we have this process in place, we can replace the input embedding of a token with a new one of our choice - which in our final use-case will be something we learn. To demonstrate the concept though, let's replace the input embedding for 'puppy' in the prompt we've been playing with with the embedding for token 2368, get a new set of output embeddings based on this, and use these to generate an image to see what we get:<jupyter_code>prompt = 'A picture of a puppy' # Tokenize text_input = tokenizer(prompt, padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt") input_ids = text_input.input_ids.to(torch_device) # Get token embeddings token_embeddings = token_emb_layer(input_ids) # The new embedding. In this case just the input embedding of token 2368... replacement_token_embedding = text_encoder.get_input_embeddings()(torch.tensor(2368, device=torch_device)) # Insert this into the token embeddings ( token_embeddings[0, torch.where(input_ids[0]==6829)] = replacement_token_embedding.to(torch_device) # Combine with pos embs input_embeddings = token_embeddings + position_embeddings # Feed through to get final output embs modified_output_embeddings = get_output_embeds(input_embeddings) print(modified_output_embeddings.shape) modified_output_embeddings<jupyter_output>torch.Size([1, 77, 768])<jupyter_text>The first few are the same, the last aren't. Everything at and after the position of the token we're replacing will be affected.If all went well, we should see something other than a puppy when we use these to generate an image. And sure enough, we do!<jupyter_code>#Generating an image with these modified embeddings def generate_with_embs(text_embeddings): height = 512 # default height of Stable Diffusion width = 512 # default width of Stable Diffusion num_inference_steps = 30 # Number of denoising steps guidance_scale = 7.5 # Scale for classifier-free guidance generator = torch.manual_seed(32) # Seed generator to create the inital latent noise batch_size = 1 max_length = text_input.input_ids.shape[-1] uncond_input = tokenizer( [""] * batch_size, padding="max_length", max_length=max_length, return_tensors="pt" ) with torch.no_grad(): uncond_embeddings = text_encoder(uncond_input.input_ids.to(torch_device))[0] text_embeddings = torch.cat([uncond_embeddings, text_embeddings]) # Prep Scheduler scheduler.set_timesteps(num_inference_steps) # Prep latents latents = torch.randn( (batch_size, unet.in_channels, height // 8, width // 8), generator=generator, ) latents = latents.to(torch_device) latents = latents * scheduler.init_noise_sigma # Loop for i, t in tqdm(enumerate(scheduler.timesteps)): # expand the latents if we are doing classifier-free guidance to avoid doing two forward passes. latent_model_input = torch.cat([latents] * 2) sigma = scheduler.sigmas[i] latent_model_input = scheduler.scale_model_input(latent_model_input, t) # predict the noise residual with torch.no_grad(): noise_pred = unet(latent_model_input, t, encoder_hidden_states=text_embeddings)["sample"] # perform guidance noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond) # compute the previous noisy sample x_t -> x_t-1 latents = scheduler.step(noise_pred, t, latents).prev_sample return latents_to_pil(latents)[0] generate_with_embs(modified_output_embeddings)<jupyter_output><empty_output><jupyter_text>Suprise! Now you know what token 2368 means ;) **What can we do with this?** Why did we go to all of this trouble? Well, we'll see a more compelling use-case shortly but the tl;dr is that once we can access and modify the token embeddings we can do tricks like replacing them with something else. In the example we just did, that was just another token embedding from the model's vocabulary, equivalent to just editing the prompt. But we can also mix tokens - for example, here's a half-puppy-half-skunk:<jupyter_code># In case you're wondering how to get the token for a word, or the embedding for a token: prompt = 'skunk' print('tokenizer(prompt):', tokenizer(prompt)) print('token_emb_layer([token_id]) shape:', token_emb_layer(torch.tensor([8797], device=torch_device)).shape) prompt = 'A picture of a puppy' # Tokenize text_input = tokenizer(prompt, padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt") input_ids = text_input.input_ids.to(torch_device) # Get token embeddings token_embeddings = token_emb_layer(input_ids) # The new embedding. Which is now a mixture of the token embeddings for 'puppy' and 'skunk' puppy_token_embedding = token_emb_layer(torch.tensor(6829, device=torch_device)) skunk_token_embedding = token_emb_layer(torch.tensor(42194, device=torch_device)) replacement_token_embedding = 0.5*puppy_token_embedding + 0.5*skunk_token_embedding # Insert this into the token embeddings ( token_embeddings[0, torch.where(input_ids[0]==6829)] = replacement_token_embedding.to(torch_device) # Combine with pos embs input_embeddings = token_embeddings + position_embeddings # Feed through to get final output embs modified_output_embeddings = get_output_embeds(input_embeddings) # Generate an image with these generate_with_embs(modified_output_embeddings)<jupyter_output><empty_output><jupyter_text>Textual InversionOK, so we can slip in a modified token embedding, and use this to generate an image. We used the token embedding for 'cat' in the above example, but what if instead could 'learn' a new token embedding for a specific concept? This is the idea behind 'Textual Inversion', in which a few example images are used to create a new token embedding:_Diagram from the [textual inversion blog post](https://textual-inversion.github.io/static/images/training/training.JPG) - note it doesn't show the positional embeddings step for simplicity_We won't cover how this training works, but we can try loading one of these new 'concepts' from the [community-created SD concepts library](https://huggingface.co/sd-concepts-library) and see how it fits in with our example above. I'll use https://huggingface.co/sd-concepts-library/birb-style since it was the first one I made :) Download the learned_embeds.bin file from there and upload the file to wherever this notebook is before running this next cell:<jupyter_code>birb_embed = torch.load('learned_embeds.bin') birb_embed.keys(), birb_embed['<birb-style>'].shape<jupyter_output><empty_output><jupyter_text>We get a dictionary with a key (the special placeholder I used, ) and the corresponding token embedding. As in the previous example, let's replace the 'puppy' token embedding with this and see what happens:<jupyter_code>prompt = 'A mouse in the style of puppy' # Tokenize text_input = tokenizer(prompt, padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt") input_ids = text_input.input_ids.to(torch_device) # Get token embeddings token_embeddings = token_emb_layer(input_ids) # The new embedding - our special birb word replacement_token_embedding = birb_embed['<birb-style>'].to(torch_device) # Insert this into the token embeddings token_embeddings[0, torch.where(input_ids[0]==6829)] = replacement_token_embedding.to(torch_device) # Combine with pos embs input_embeddings = token_embeddings + position_embeddings # Feed through to get final output embs modified_output_embeddings = get_output_embeds(input_embeddings) # And generate an image with this: generate_with_embs(modified_output_embeddings)<jupyter_output><empty_output><jupyter_text>The token for 'puppy' was replaced with one that captures a particular style of painting, but it could just as easily represent a specific object or class of objects. Again, there is [a nice inference notebook ](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/stable_conceptualizer_inference.ipynb) from hf to make it easy to use the different concepts, that properly handles using the names in prompts ("A \ in the style of \") without worrying about all this manual stuff. The goal of this notebook is to pull back the curtain a bit so you know what is going on behind the scenes :) Messing with EmbeddingsBesides just replacing the token embedding of a single word, there are various other tricks we can try. For example, what if we create a 'chimera' by averaging the embeddings of two different prompts?<jupyter_code># Embed two prompts text_input1 = tokenizer(["A mouse"], padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt") text_input2 = tokenizer(["A leopard"], padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt") with torch.no_grad(): text_embeddings1 = text_encoder(text_input1.input_ids.to(torch_device))[0] text_embeddings2 = text_encoder(text_input2.input_ids.to(torch_device))[0] # Mix them together mix_factor = 0.35 mixed_embeddings = (text_embeddings1*mix_factor + \ text_embeddings2*(1-mix_factor)) # Generate! generate_with_embs(mixed_embeddings)<jupyter_output><empty_output><jupyter_text>The UNET and CFGNow it's time we looked at the actual diffusion model. This is typically a Unet that takes in the noisy latents (x) and predicts the noise. We use a conditional model that also takes in the timestep (t) and our text embedding (aka encoder_hidden_states) as conditioning. Feeding all of these into the model looks like this:`noise_pred = unet(latents, t, encoder_hidden_states=text_embeddings)["sample"]`We can try it out and see what the output looks like:<jupyter_code># Prep Scheduler scheduler.set_timesteps(num_inference_steps) # What is our timestep t = scheduler.timesteps[0] sigma = scheduler.sigmas[0] # A noisy latent latents = torch.randn( (batch_size, unet.in_channels, height // 8, width // 8), generator=generator, ) latents = latents.to(torch_device) latents = latents * scheduler.init_noise_sigma # Text embedding text_input = tokenizer(['A macaw'], padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt") with torch.no_grad(): text_embeddings = text_encoder(text_input.input_ids.to(torch_device))[0] # Run this through the unet to predict the noise residual with torch.no_grad(): noise_pred = unet(latents, t, encoder_hidden_states=text_embeddings)["sample"] latents.shape, noise_pred.shape # We get preds in the same shape as the input<jupyter_output><empty_output><jupyter_text>Given a set of noisy latents, the model predicts the noise component. We can remove this noise from the noisy latents to see what the output image looks like (`latents_x0 = latents - sigma * noise_pred`). And we can add most of the noise back to this predicted output to get the (slightly less noisy hopefully) input for the next diffusion step. To visualize this let's generate another image, saving both the predicted output (x0) and the next step (xt-1) after every step:<jupyter_code>prompt = 'Oil painting of an otter in a top hat' height = 512 width = 512 num_inference_steps = 50 guidance_scale = 8 generator = torch.manual_seed(32) batch_size = 1 # Make a folder to store results !rm -rf steps/ !mkdir -p steps/ # Prep text text_input = tokenizer([prompt], padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt") with torch.no_grad(): text_embeddings = text_encoder(text_input.input_ids.to(torch_device))[0] max_length = text_input.input_ids.shape[-1] uncond_input = tokenizer( [""] * batch_size, padding="max_length", max_length=max_length, return_tensors="pt" ) with torch.no_grad(): uncond_embeddings = text_encoder(uncond_input.input_ids.to(torch_device))[0] text_embeddings = torch.cat([uncond_embeddings, text_embeddings]) # Prep Scheduler scheduler.set_timesteps(num_inference_steps) # Prep latents latents = torch.randn( (batch_size, unet.in_channels, height // 8, width // 8), generator=generator, ) latents = latents.to(torch_device) latents = latents * scheduler.init_noise_sigma # Loop for i, t in tqdm(enumerate(scheduler.timesteps)): # expand the latents if we are doing classifier-free guidance to avoid doing two forward passes. latent_model_input = torch.cat([latents] * 2) sigma = scheduler.sigmas[i] latent_model_input = scheduler.scale_model_input(latent_model_input, t) # predict the noise residual with torch.no_grad(): noise_pred = unet(latent_model_input, t, encoder_hidden_states=text_embeddings)["sample"] # perform guidance noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond) # Get the predicted x0: # latents_x0 = latents - sigma * noise_pred # Calculating ourselves latents_x0 = scheduler.step(noise_pred, t, latents).pred_original_sample # Using the scheduler (Diffusers 0.4 and above) # compute the previous noisy sample x_t -> x_t-1 latents = scheduler.step(noise_pred, t, latents).prev_sample # To PIL Images im_t0 = latents_to_pil(latents_x0)[0] im_next = latents_to_pil(latents)[0] # Combine the two images and save for later viewing im = Image.new('RGB', (1024, 512)) im.paste(im_next, (0, 0)) im.paste(im_t0, (512, 0)) im.save(f'steps/{i:04}.jpeg') # Make and show the progress video (change width to 1024 for full res) !ffmpeg -v 1 -y -f image2 -framerate 12 -i steps/%04d.jpeg -c:v libx264 -preset slow -qp 18 -pix_fmt yuv420p out.mp4 mp4 = open('out.mp4','rb').read() data_url = "data:video/mp4;base64," + b64encode(mp4).decode() HTML(""" <video width=600 controls> <source src="%s" type="video/mp4"> </video> """ % data_url)<jupyter_output><empty_output><jupyter_text>The version on the right shows the predicted 'final output' (x0) at each step, and this is what is usually used for progress videos etc. The version on the left is the 'next step'. I found it interesteing to compare the two - watching the progress videos only you'd think drastic changes are happening expecially at early stages, but since the changes made per-step are relatively small the actual process is much more gradual. Classifier Free GuidanceBy default, the model doesn't often do what we ask. If we want it to follow the prompt better, we use a hack called CFG. There's a good explanation in this video (AI coffee break GLIDE).In the code, this comes down to us doing:`noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)`This works suprisingly well :) Explore changing the guidance_scale in the code above and see how this affects the results. How high can you push it before the results get worse? SamplingThere is still some complexity hidden from us inside `latents = scheduler.step(noise_pred, i, latents)["prev_sample"]`. How exactly does the sampler go from the current noisy latents to a slightly less noisy version? Why don't we just use the model in a single step? Are there other ways to view this?The model tries to predict the noise in an image. For low noise values, we assume it does a pretty good job. For higher noise levels, it has a hard task! So instead of producing a perfect image, the results tend to look like a blurry mess - see the start of the video above for a visual! So, samplers use the model predictions to move a small amount towards the model prediction (removing some of the noise) and then get another prediction based on this marginally-less-rubbish input, and hope that this iteratively improves the result.Different samplers do this in different ways. You can try to inspect the code for the default LMS sampler with:<jupyter_code># ??scheduler.step<jupyter_output><empty_output><jupyter_text>**Time to draw some diagrams!** (Whiteboard/paper interlude) GuidanceOK, final trick! How can we add some extra control to this generation process?At each step, we're going to use our model as before to predict the noise component of x. Then we'll use this to produce a predicted output image, and apply some loss function to this image.This function can be anything, but let's demo with a super simple example. If we want images that have a lot of blue, we can craft a loss function that gives a high loss if pixels have a low blue component:<jupyter_code>def blue_loss(images): # How far are the blue channel values to 0.9: error = torch.abs(images[:,2] - 0.9).mean() # [:,2] -> all images in batch, only the blue channel return error<jupyter_output><empty_output><jupyter_text>During each update step, we find the gradient of the loss with respect to the current noisy latents, and tweak them in the direction that reduces this loss as well as performing the normal update step:<jupyter_code>prompt = 'A campfire (oil on canvas)' #@param height = 512 # default height of Stable Diffusion width = 512 # default width of Stable Diffusion num_inference_steps = 50 #@param # Number of denoising steps guidance_scale = 8 #@param # Scale for classifier-free guidance generator = torch.manual_seed(32) # Seed generator to create the inital latent noise batch_size = 1 blue_loss_scale = 200 #@param # Prep text text_input = tokenizer([prompt], padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt") with torch.no_grad(): text_embeddings = text_encoder(text_input.input_ids.to(torch_device))[0] # And the uncond. input as before: max_length = text_input.input_ids.shape[-1] uncond_input = tokenizer( [""] * batch_size, padding="max_length", max_length=max_length, return_tensors="pt" ) with torch.no_grad(): uncond_embeddings = text_encoder(uncond_input.input_ids.to(torch_device))[0] text_embeddings = torch.cat([uncond_embeddings, text_embeddings]) # Prep Scheduler scheduler.set_timesteps(num_inference_steps) # Prep latents latents = torch.randn( (batch_size, unet.in_channels, height // 8, width // 8), generator=generator, ) latents = latents.to(torch_device) latents = latents * scheduler.init_noise_sigma # Loop for i, t in tqdm(enumerate(scheduler.timesteps)): # expand the latents if we are doing classifier-free guidance to avoid doing two forward passes. latent_model_input = torch.cat([latents] * 2) sigma = scheduler.sigmas[i] latent_model_input = scheduler.scale_model_input(latent_model_input, t) # predict the noise residual with torch.no_grad(): noise_pred = unet(latent_model_input, t, encoder_hidden_states=text_embeddings)["sample"] # perform CFG noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond) #### ADDITIONAL GUIDANCE ### if i%5 == 0: # Requires grad on the latents latents = latents.detach().requires_grad_() # Get the predicted x0: # latents_x0 = latents - sigma * noise_pred latents_x0 = scheduler.step(noise_pred, t, latents).pred_original_sample # Decode to image space denoised_images = vae.decode((1 / 0.18215) * latents_x0).sample / 2 + 0.5 # range (0, 1) # Calculate loss loss = blue_loss(denoised_images) * blue_loss_scale # Occasionally print it out if i%10==0: print(i, 'loss:', loss.item()) # Get gradient cond_grad = torch.autograd.grad(loss, latents)[0] # Modify the latents based on this gradient latents = latents.detach() - cond_grad * sigma**2 # Now step with scheduler latents = scheduler.step(noise_pred, t, latents).prev_sample latents_to_pil(latents)[0]<jupyter_output><empty_output>

diffusion-models-class/units/en/unit3/stable_diffusion_deep_dive.ipynb/0

# Introduction à 🤗 Diffusers <CourseFloatingBanner unit={1} classNames="absolute z-10 right-0 top-0" notebooks={[ {label: "Introduction to Diffusers", value: "https://colab.research.google.com/github/huggingface/diffusion-models-class/blob/main/units/fr/unit1/introduction_to_diffusers.ipynb"}, {label: "Introduction to Diffusers", value: "https://studiolab.sagemaker.aws/import/github/huggingface/diffusion-models-class/blob/main/units/fr/unit1/introduction_to_diffusers.ipynb"}, ]} /> Dans ce *notebook*, vous allez entraîner votre premier modèle de diffusion pour générer des images de mignons papillons 🦋. En cours de route, vous apprendrez les composants de base de la bibliothèque 🤗 *Diffusers*, qui fournira une bonne assise pour les applications plus avancées que nous couvrirons plus tard dans le cours. Débutons par une vue d'ensemble de ce qu'on va faire dans ce *notebook*. Nous allons : - Voir un puissant pipeline de modèles de diffusion personnalisé en action (avec des informations sur la façon de créer votre propre version). - Créer votre propre mini-pipeline en : - Récapitulant les idées principales derrière les modèles de diffusion - Chargement de données à partir du Hub pour l'entraînement - Explorer comment ajouter du bruit à ces données à l'aide d'un planificateur - Créer et entraîner le modèle UNet - Rassembler les pièces du puzzle pour en faire un pipeline fonctionnel - Éditer et exécuter un script pour initialiser des séries d'entraînement plus longues, qui gèrera - Entraînement multi-GPU via 🤗 *Accelerate* - Journalisation de l'expérience pour suivre les statistiques critiques - Téléchargement du modèle final sur le *Hub* d'*Hugging Face* ## Installation des bibliothèques Exécutez la cellule suivante pour installer la bibliothèque 🤗 *Diffusers* ainsi que quelques autres prérequis : ```py %pip install -qq -U diffusers datasets transformers accelerate ftfy pyarrow==9.0.0 ``` Ensuite, rendez-vous sur https://huggingface.co/settings/tokens et créez un *tokens* d'accès avec autorisation d'écriture si vous n'en avez pas déjà un : <div class="flex justify-center"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary.svg" alt="Bref aperçu du contenu du cours."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary-dark.svg" alt="Bref aperçu des différents chapitres du cours."> </div> Vous pouvez vous connecter avec ce token en utilisant la ligne de commande (`huggingface-cli login`) ou en exécutant la cellule suivante : ```py from huggingface_hub import notebook_login notebook_login() ``` Vous devez ensuite installer Git-LFS pour télécharger les *checkpoints* de votre modèle : ```py %%capture !sudo apt -qq install git-lfs !git config --global credential.helper store ``` Enfin, importons les bibliothèques que nous utiliserons et définissons quelques fonctions de confort que nous utiliserons plus tard dans le *notebook* : ```py import numpy as np import torch import torch.nn.functional as F from matplotlib import pyplot as plt from PIL import Image def show_images(x): """Étant donné un lot d'images x, faire une grille et convertir en PIL""" x = x * 0.5 + 0.5 # On va de (-1, 1) et revenons (0, 1) grid = torchvision.utils.make_grid(x) grid_im = grid.detach().cpu().permute(1, 2, 0).clip(0, 1) * 255 grid_im = Image.fromarray(np.array(grid_im).astype(np.uint8)) return grid_im def make_grid(images, size=64): """Étant donné une liste d'images PIL, les empiler en une ligne pour faciliter la visualisation.""" output_im = Image.new("RGB", (size * len(images), size)) for i, im in enumerate(images): output_im.paste(im.resize((size, size)), (i * size, 0)) return output_im # Les utilisateurs de Mac peuvent avoir besoin de device = 'mps' (non testé) device = torch.device("cuda" if torch.cuda.is_available() else "cpu") ``` OK, nous sommes prêts ! ## Exemple générique d'inférence avec Dreambooth, un avant-goût de ce qui nous attend Si vous avez un tant soit peu consulté les médias sociaux au cours des derniers mois, vous avez certainement entendu parler de *Stable Diffusion*. Il s'agit d'un puissant modèle de diffusion latent conditionné par le texte (ne vous inquiétez pas, nous allons apprendre ce que cela signifie). Mais il a un défaut : il ne sait pas à quoi vous ou moi ressemblons, à moins que nous soyons suffisamment célèbres pour que nos images soient répandues sur internet. Dreambooth nous permet de créer notre propre variante de modèle avec une connaissance supplémentaire d'un visage, d'un objet ou d'un style spécifique. Le Corridor Crew a réalisé une excellente vidéo (en anglais) en utilisant cette technique pour raconter des histoires avec des personnages cohérents, ce qui est un excellent exemple de ce que cette technique peut faire : ```py from IPython.display import YouTubeVideo YouTubeVideo("W4Mcuh38wyM") ``` Voici un exemple d'une sortie d'un [modèle](https://huggingface.co/sd-dreambooth-library/mr-potato-head) entraîné sur 5 photos du jouet Monsieur Patate. Tout d'abord, nous chargeons le pipeline. Ceci télécharge les poids du modèle depuis le Hub. Étant donné que plusieurs gigaoctets de données sont téléchargés pour une démonstration d'une ligne, vous pouvez sauter cette cellule et simplement admirer la sortie de l'exemple ! ```py from diffusers import StableDiffusionPipeline # Consultez https://huggingface.co/sd-dreambooth-library pour découvrir de nombreux modèles provenant de la communauté model_id = "sd-dreambooth-library/mr-potato-head" # Chargement du pipeline pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16).to( device ) ``` Une fois le chargement du pipeline terminé, nous pouvons générer des images avec : ```py prompt = "an abstract oil painting of sks mr potato head by picasso" image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0] image ``` <div class="flex justify-center"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary.svg" alt="Bref aperçu du contenu du cours."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary-dark.svg" alt="Bref aperçu des différents chapitres du cours."> </div> <Tip> ✏️ *A votre tour !* essayez vous-même avec des prompts différents. Le *token* `sks` représente un identifiant unique pour le nouveau concept, que se passe-t-il si vous l'omettez ? Vous pouvez aussi expérimenter en changeant le nombre de pas d'échantillonnage (jusqu'où pouvez-vous descendre ?) et le paramètre `guidance_scale`, qui détermine jusqu'à quel point le modèle va essayer de correspondre au prompt. Il se passe beaucoup de choses dans ce pipeline ! À la fin du cours, vous saurez comment tout cela fonctionne. Pour l'instant, voyons comment nous pouvons entraîner un modèle de diffusion à partir de zéro. </Tip> ## MVP (Minimum Viable Pipeline) ### Exemple d'inférence sur les papillons L'API de base de 🤗 *Diffusers* est divisée en trois composants principaux : - **Pipelines** : classes de haut niveau conçues pour générer rapidement des échantillons à partir de modèles de diffusion populaires entraînés de manière conviviale. - **Models** : architectures populaires pour entraîner de nouveaux modèles de diffusion, par exemple [UNet](https://arxiv.org/abs/1505.04597). - **Schedulers** : diverses techniques pour générer des images à partir du bruit pendant l'*inférence* ainsi que pour générer des images bruitées pour l'*entraînement*. Les pipelines sont parfaits pour les utilisateurs finaux, mais si vous êtes ici pour ce cours, nous supposons que vous voulez savoir ce qui se passe sous le capot ! Dans le reste de ce *notebook*, nous allons donc construire notre propre pipeline capable de générer de petites images de papillons. Voici le résultat final en action : ```py from diffusers import DDPMPipeline # Chargement du pipeline de papillons butterfly_pipeline = DDPMPipeline.from_pretrained( "johnowhitaker/ddpm-butterflies-32px" ).to(device) # Création de 8 images images = butterfly_pipeline(batch_size=8).images # Visualisation du résultat make_grid(images) ``` <div class="flex justify-center"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary.svg" alt="Bref aperçu du contenu du cours."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary-dark.svg" alt="Bref aperçu des différents chapitres du cours."> </div> Ce n'est peut-être pas aussi impressionnant que l'exemple de DreamBooth, mais nous entraînons notre modèle à partir de zéro avec ~0,0001% des données utilisées pour entraîner Stable Diffusion. En parlant d'entraînement, rappelez-vous que l'entraînement d'un modèle de diffusion ressemble à ceci : - Chargement de quelques images à partir des données entraînées. - Ajout de bruit, en différentes quantités. - Introduction des versions bruitées des données d'entrée dans le modèle. - Évaluation de la capacité du modèle à débruiter ces données d'entrée - Utilisation de ces informations pour mettre à jour les poids du modèle, et répétition. Nous allons explorer ces étapes une par une dans les prochaines parties jusqu'à ce que nous ayons une boucle d'entraînement complète, puis nous verrons comment échantillonner à partir du modèle entraîné et comment regrouper le tout dans un pipeline pour faciliter le partage. Commençons par les données. ### Télécharger le jeu de données d'entraînement Pour cet exemple, nous utilisons un jeu de données d'images provenant du *Hub* d'*Hugging Face*. Plus précisément, cette collection de [1000 images de papillons](https://huggingface.co/datasets/huggan/smithsonian_butterflies_subset). Il s'agit d'un très petit jeu de données, c'est pourquoi nous avons aussi inclus des lignes en commentaires pour quelques options plus importantes. Si vous préférez utiliser votre propre collection d'images, vous pouvez également utiliser l'exemple de code commenté pour charger des images à partir d'un dossier. ```py import torchvision from datasets import load_dataset from torchvision import transforms dataset = load_dataset("huggan/smithsonian_butterflies_subset", split="train") # Ou charger des images à partir d'un dossier local # dataset = load_dataset("imagefolder", data_dir="path/to/folder") # Nous entraînerons sur des images carrées de 32 pixels, mais vous pouvez aussi essayer des tailles plus grandes image_size = 32 # Vous pouvez réduire la taille de votre batch si vous manquez de mémoire GPU batch_size = 64 # Définition les augmentations de données preprocess = transforms.Compose( [ transforms.Resize((image_size, image_size)), # Redimensionner transforms.RandomHorizontalFlip(), # Retournement aléatoire transforms.ToTensor(), # Convertir en tenseur (0, 1) transforms.Normalize([0.5], [0.5]), # Passage en (-1, 1) ] ) def transform(examples): images = [preprocess(image.convert("RGB")) for image in examples["image"]] return {"images": images} dataset.set_transform(transform) # Créer un chargeur de données à partir du jeu de données pour servir les images transformées en batchs train_dataloader = torch.utils.data.DataLoader( dataset, batch_size=batch_size, shuffle=True ) ``` Nous pouvons saisir un batch d'images et en visualiser quelques-unes comme suit : ```py xb = next(iter(train_dataloader))["images"].to(device)[:8] print("X shape:", xb.shape) show_images(xb).resize((8 * 64, 64), resample=Image.NEAREST) ``` <div class="flex justify-center"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary.svg" alt="Bref aperçu du contenu du cours."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary-dark.svg" alt="Bref aperçu des différents chapitres du cours."> </div> Nous nous en tenons à un petit jeu de données avec des images de 32 pixels pour que les temps d'entraînement restent raisonnables dans ce *notebook*. ## Définir le planificateur Notre plan d'entraînement consiste à prendre ces images d'entrée et à leur ajouter du bruit, puis à transmettre les images bruitées au modèle. Lors de l'inférence, nous utiliserons les prédictions du modèle pour supprimer le bruit de manière itérative. Dans 🤗 *Diffusers*, ces deux processus sont gérés par le *scheduler* (planificateur). Le planificateur de bruit détermine la quantité de bruit ajoutée à différents moments. Voici comment nous pourrions créer un planificateur en utilisant les paramètres par défaut pour l'entraînement et l'échantillonnage "DDPM" (d'après l'article d'après l'article [*Denoising Diffusion Probabalistic Models*](https://arxiv.org/abs/2006.11239)) : ```py from diffusers import DDPMScheduler noise_scheduler = DDPMScheduler(num_train_timesteps=1000) ``` Le papier DDPM décrit un processus de corruption qui ajoute une petite quantité de bruit à chaque pas de temps. Étant donné $x_{t-1}$ pour un certain pas de temps, nous pouvons obtenir la version suivante (légèrement plus bruyante) $x_t$ avec : $q(\mathbf{x}_t \vert \mathbf{x}_{t-1}) = \mathcal{N}(\mathbf{x}_t; \sqrt{1 - \beta_t} \mathbf{x}_{t-1}, \beta_t\mathbf{I}) \quad q(\mathbf{x}_{1:T} \vert \mathbf{x}_0) = \prod^T_{t=1} q(\mathbf{x}_t \vert \mathbf{x}_{t-1})$<br><br> Nous prenons $x_{t-1}$, l'échelonnons de $\sqrt{1 - \beta_t}$ et ajoutons du bruit échelonné par $\beta_t$. Ce $\beta$ est défini pour chaque $t$ selon un certain planificateur et détermine la quantité de bruit ajoutée par pas de temps. Maintenant, nous ne voulons pas nécessairement faire cette opération 500 fois pour obtenir $x_{500}$, nous avons donc une autre formule pour obtenir $x_t$ pour n'importe quel t étant donné $x_0$ : $\begin{aligned} q(\mathbf{x}_t \vert \mathbf{x}_0) &= \mathcal{N}(\mathbf{x}_t; \sqrt{\bar{\alpha}_t} \mathbf{x}_0, {(1 - \bar{\alpha}_t)} \mathbf{I}) \end{aligned}$ where $\bar{\alpha}_t = \prod_{i=1}^T \alpha_i$ and $\alpha_i = 1-\beta_i$<br><br> La notation mathématique fait toujours peur ! Heureusement, le planificateur s'en charge pour nous. Nous pouvons tracer $\sqrt{\bar{\alpha}_t}$ (appelé `sqrt_alpha_prod`) et $\sqrt{(1 - \bar{\alpha}_t)}$ (appelé `sqrt_one_minus_alpha_prod`) pour voir comment l'entrée ($x$) et le bruit sont mis à l'échelle et mélangés à travers différents pas de temps : ```py plt.plot(noise_scheduler.alphas_cumprod.cpu() ** 0.5, label=r"${\sqrt{\bar{\alpha}_t}}$") plt.plot((1 - noise_scheduler.alphas_cumprod.cpu()) ** 0.5, label=r"$\sqrt{(1 - \bar{\alpha}_t)}$") plt.legend(fontsize="x-large"); ``` <div class="flex justify-center"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary.svg" alt="Bref aperçu du contenu du cours."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary-dark.svg" alt="Bref aperçu des différents chapitres du cours."> </div> <Tip> ✏️ *A votre tour !* Vous pouvez explorer comment ce graphique change avec différents paramètres pour `beta_start`, `beta_end` et `beta_schedule` en remplaçant l'une des options commentées ci-dessous : </Tip> ```py ## Exemple avec beaucoup de bruit ajouté : # noise_scheduler = DDPMScheduler(num_train_timesteps=1000, beta_start=0.001, beta_end=0.004) ## Le planificateur cosinus pouvant s'avérer meilleur pour les images de petite taille : # noise_scheduler = DDPMScheduler(num_train_timesteps=1000, beta_schedule='squaredcos_cap_v2') ``` Quel que soit le planificateur que vous avez choisi, nous pouvons maintenant l'utiliser pour ajouter du bruit en différentes quantités en utilisant la fonction `noise_scheduler.add_noise` comme suit : ```py timesteps = torch.linspace(0, 999, 8).long().to(device) noise = torch.randn_like(xb) noisy_xb = noise_scheduler.add_noise(xb, noise, timesteps) print("Noisy X shape", noisy_xb.shape) show_images(noisy_xb).resize((8 * 64, 64), resample=Image.NEAREST) ``` <div class="flex justify-center"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary.svg" alt="Bref aperçu du contenu du cours."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary-dark.svg" alt="Bref aperçu des différents chapitres du cours."> </div> Là encore, étudiez l'effet de l'utilisation de différents planificateurs et paramètres de bruit. Cette [vidéo](https://www.youtube.com/watch?v=fbLgFrlTnGU) (en anglais) explique en détail certains des calculs ci-dessus et constitue une excellente introduction à certains de ces concepts. ### Définir le modèle Nous en arrivons maintenant à l'élément central : le modèle lui-même. La plupart des modèles de diffusion utilisent des architectures qui sont des variantes d'un [U-net](https://arxiv.org/abs/1505.04597) et c'est ce que nous utiliserons ici. <div class="flex justify-center"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary.svg" alt="Bref aperçu du contenu du cours."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary-dark.svg" alt="Bref aperçu des différents chapitres du cours."> </div> En bref : - l'image en entrée du modèle passe par plusieurs blocs de couches ResNet, chacun divisant la taille de l'image par 2 - puis elle passe à travers le même nombre de blocs qui la suréchantillonnent. - il y a des *skip connections* qui relient les caractéristiques sur le chemin du sous-échantillonnage aux couches correspondantes dans le chemin du suréchantillonnage. L'une des principales caractéristiques de ce modèle est qu'il prédit des images de la même taille que l'entrée, ce qui est exactement ce dont nous avons besoin ici. 🤗 *Diffusers* nous fournit une classe `UNet2DModel` pratique qui crée l'architecture désirée dans PyTorch. Créons un U-net pour la taille d'image désirée. Notez que les `down_block_types` correspondent aux blocs de sous-échantillonnage (en vert sur le diagramme ci-dessus), et que les `up_block_types` sont les blocs de suréchantillonnage (en rouge sur le diagramme) : ```py from diffusers import UNet2DModel # Création d'un modèle model = UNet2DModel( sample_size=image_size, # la résolution de l'image cible in_channels=3, # le nombre de canaux d'entrée, 3 pour les images RVB out_channels=3, # le nombre de canaux de sortie layers_per_block=2, # le nombre de couches ResNet à utiliser par bloc UNet block_out_channels=(64, 128, 128, 256), # Plus de canaux -> plus de paramètres down_block_types=( "DownBlock2D", # un bloc de sous-échantillonnage ResNet standard "DownBlock2D", "AttnDownBlock2D", # un bloc de sous-échantillonnage ResNet avec auto-attention spatiale "AttnDownBlock2D", ), up_block_types=( "AttnUpBlock2D", "AttnUpBlock2D", # un bloc de suréchantillonnage ResNet avec auto-attention spatiale "UpBlock2D", "UpBlock2D", # un bloc de suréchantillonnage ResNet standard ), ) model.to(device) ``` Lorsque vous traitez des données d'entrée en haute résolution, vous pouvez utiliser davantage de blocs descendants et ascendants, et ne conserver les couches d'attention que pour les couches de résolution les plus basses (inférieures) afin de réduire l'utilisation de la mémoire. Nous verrons plus tard comment vous pouvez expérimenter pour trouver les meilleurs paramètres pour votre cas d'utilisation. Nous pouvons vérifier que le passage d'un batch de données et de pas de temps aléatoires produit une sortie de même forme que les données d'entrée : ```py with torch.no_grad(): model_prediction = model(noisy_xb, timesteps).sample model_prediction.shape ``` Dans la section suivante, nous verrons comment entraîner ce modèle. ### Créer une boucle d'entraînement Il est temps d'entraîner ! Voici une boucle d'optimisation typique dans PyTorch, où nous parcourons les données batch par batch et mettons à jour les paramètres de notre modèle à chaque étape à l'aide d'un optimiseur, ici, l'optimiseur AdamW avec un taux d'apprentissage de 0,0004. Pour chaque batch de données, nous - échantillonnons des pas de temps aléatoires - bruitons les données en conséquence - transmettons les données bruitées au modèle - comparons les prédictions du modèle avec la cible (c'est-à-dire le bruit dans ce cas) en utilisant l'erreur quadratique moyenne comme fonction de perte - mettons à jour les paramètres du modèle via `loss.backward()` et `optimizer.step()`. Au cours de ce processus, nous enregistrons aussi les pertes au fil du temps pour un tracé ultérieur. NB : ce code prend près de 10 minutes à exécuter. N'hésitez pas à sauter ces deux cellules et à utiliser le modèle pré-entraîné si vous êtes pressé. Vous pouvez également étudier comment la réduction du nombre de canaux dans chaque couche via la définition du modèle ci-dessus peut accélérer les choses. L'[exemple officiel d'entraînement](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/training_example.ipynb) de 🤗 *Diffusers* entraîne un modèle plus grand sur ce jeu de données à une résolution plus élevée, et constitue une bonne référence pour ce à quoi ressemble une boucle d'entraînement moins minimale : ```py # Définir le planificateur de bruit noise_scheduler = DDPMScheduler( num_train_timesteps=1000, beta_schedule="squaredcos_cap_v2" ) # Boucle d'entraînement optimizer = torch.optim.AdamW(model.parameters(), lr=4e-4) losses = [] for epoch in range(30): for step, batch in enumerate(train_dataloader): clean_images = batch["images"].to(device) # Exemple de bruit à ajouter aux images noise = torch.randn(clean_images.shape).to(clean_images.device) bs = clean_images.shape[0] # Échantillonner un pas de temps aléatoire pour chaque image timesteps = torch.randint( 0, noise_scheduler.num_train_timesteps, (bs,), device=clean_images.device ).long() # Ajouter du bruit aux images propres en fonction de l'ampleur du bruit à chaque étape noisy_images = noise_scheduler.add_noise(clean_images, noise, timesteps) # Obtenir la prédiction du modèle noise_pred = model(noisy_images, timesteps, return_dict=False)[0] # Calculer la perte loss = F.mse_loss(noise_pred, noise) loss.backward(loss) losses.append(loss.item()) # Mise à jour des paramètres du modèle à l'aide de l'optimiseur optimizer.step() optimizer.zero_grad() if (epoch + 1) % 5 == 0: loss_last_epoch = sum(losses[-len(train_dataloader) :]) / len(train_dataloader) print(f"Epoch:{epoch+1}, loss: {loss_last_epoch}") ``` En traçant la perte, nous constatons que le modèle s'améliore rapidement dans un premier temps, puis continue à s'améliorer à un rythme plus lent (ce qui est plus évident si nous utilisons une échelle logarithmique, comme indiqué à droite) : ```py fig, axs = plt.subplots(1, 2, figsize=(12, 4)) axs[0].plot(losses) axs[1].plot(np.log(losses)) plt.show() ``` <div class="flex justify-center"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary.svg" alt="Bref aperçu du contenu du cours."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary-dark.svg" alt="Bref aperçu des différents chapitres du cours."> </div> Au lieu d'exécuter le code d'entraînement ci-dessus, vous pouvez utiliser le modèle du pipeline comme suit : ```py ## Décommenter pour charger le modèle que j'ai entraîné plus tôt à la place : # model = butterfly_pipeline.unet ``` ### Générer des images Comment obtenir des images avec ce modèle ? • Option 1 : Création d'un pipeline : ```py from diffusers import DDPMPipeline image_pipe = DDPMPipeline(unet=model, scheduler=noise_scheduler) pipeline_output = image_pipe() pipeline_output.images[0] ``` <div class="flex justify-center"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary.svg" alt="Bref aperçu du contenu du cours."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary-dark.svg" alt="Bref aperçu des différents chapitres du cours."> </div> Nous pouvons enregistrer un pipeline dans un dossier local comme suit : ```py image_pipe.save_pretrained("my_pipeline") ``` Inspection du contenu du dossier : ```py !ls my_pipeline/ ``` ```py model_index.json scheduler unet ``` Les sous-dossiers `scheduler` et `unet` contiennent tout ce qui est nécessaire pour recréer ces composants. Par exemple, dans le dossier `unet` vous trouverez les poids du modèle (`diffusion_pytorch_model.bin`) ainsi qu'un fichier de configuration qui spécifie l'architecture UNet. ```py !ls my_pipeline/unet/ ``` ```py config.json diffusion_pytorch_model.bin ``` Ensemble, ces fichiers contiennent tout ce qui est nécessaire pour recréer le pipeline. Vous pouvez les télécharger manuellement sur le *Hub* pour partager le pipeline avec d'autres personnes, ou consulter le code pour le faire via l'API dans la section suivante. • Option 2 : écrire une boucle d'échantillonnage Si vous inspectez la méthode `forward` du pipeline, vous pourrez voir ce qui se passe lorsque nous lançons `image_pipe()` : ```py # ??image_pipe.forward ``` Nous commençons par un bruit aléatoire et parcourons les pas de temps de l'ordonnanceur du plus bruyant au moins bruyant, en supprimant une petite quantité de bruit à chaque étape sur la base de la prédiction du modèle : ```py # Point de départ aléatoire (8 images aléatoires) : sample = torch.randn(8, 3, 32, 32).to(device) for i, t in enumerate(noise_scheduler.timesteps): # Obtenir le modèle de prédiction with torch.no_grad(): residual = model(sample, t).sample # Mise à jour de l'échantillon avec le pas sample = noise_scheduler.step(residual, t, sample).prev_sample show_images(sample) ``` <div class="flex justify-center"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary.svg" alt="Bref aperçu du contenu du cours."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary-dark.svg" alt="Bref aperçu des différents chapitres du cours."> </div> La fonction `noise_scheduler.step()` effectue les calculs nécessaires pour mettre à jour `sample` de manière appropriée. Il existe un certain nombre de méthodes d'échantillonnage. Dans l'unité suivante, nous verrons comment nous pouvons échanger un échantillonneur différent pour accélérer la génération d'images avec des modèles existants, et nous parlerons plus en détail de la théorie derrière l'échantillonnage des modèles de diffusion. ### Pousser votre modèle vers le *Hub* Dans l'exemple ci-dessus, nous avons enregistré notre pipeline dans un dossier local. Pour pousser notre modèle vers le *Hub*, nous aurons besoin d'un dépôt de modèles dans lequel nous pourrons pousser nos fichiers. Nous déterminerons le nom du dépôt à partir de l'ID du modèle que nous voulons donner à notre modèle (n'hésitez pas à remplacer le nom du modèle par votre propre choix ; il doit juste contenir votre nom d'utilisateur, ce que fait la fonction `get_full_repo_name()`) : ```py from huggingface_hub import get_full_repo_name model_name = "sd-class-butterflies-32" hub_model_id = get_full_repo_name(model_name) hub_model_id ``` Ensuite, créer un dépôt de modèle sur le 🤗 *Hub* et pousser notre modèle : ```py from huggingface_hub import HfApi, create_repo create_repo(hub_model_id) api = HfApi() api.upload_folder( folder_path="my_pipeline/scheduler", path_in_repo="", repo_id=hub_model_id ) api.upload_folder(folder_path="my_pipeline/unet", path_in_repo="", repo_id=hub_model_id) api.upload_file( path_or_fileobj="my_pipeline/model_index.json", path_in_repo="model_index.json", repo_id=hub_model_id, ) ``` La dernière chose à faire est de créer une belle carte modèle afin que notre générateur de papillons puisse être facilement trouvé sur le 🤗 *Hub* (n'hésitez pas à développer et à modifier la description !) : ```py from huggingface_hub import ModelCard content = f""" --- license: mit tags: - pytorch - diffusers - unconditional-image-generation - diffusion-models-class --- # Model Card for Unit 1 of the [Diffusion Models Class 🧨](https://github.com/huggingface/diffusion-models-class) This model is a diffusion model for unconditional image generation of cute 🦋. ## Usage ```python from diffusers import DDPMPipeline pipeline = DDPMPipeline.from_pretrained('{hub_model_id}') image = pipeline().images[0] image ``` """ card = ModelCard(content) card.push_to_hub(hub_model_id) ``` Maintenant que le modèle est sur le *Hub*, vous pouvez le télécharger de n'importe où en utilisant la méthode `from_pretrained()` de `DDPMPipeline` comme suit : ```py from diffusers import DDPMPipeline image_pipe = DDPMPipeline.from_pretrained(hub_model_id) pipeline_output = image_pipe() pipeline_output.images[0] ``` <div class="flex justify-center"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary.svg" alt="Bref aperçu du contenu du cours."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary-dark.svg" alt="Bref aperçu des différents chapitres du cours."> </div> Bien, ça marche ! ## Passer à l'échelle supérieure avec 🤗 *Accelerate* Ce *notebook* a été conçu à des fins d'apprentissage, et en tant que tel, nous avons essayé de garder le code aussi minimal et propre que possible. Pour cette raison, nous avons omis certaines choses que vous pourriez souhaiter si vous deviez entraîner un modèle plus grand sur beaucoup plus de données, comme le support multi-GPU, la trace de la progression et des images d'exemple, la sauvegarde du gradient pour supporter des tailles de batch plus importantes, le téléchargement automatique des modèles et ainsi de suite. Heureusement, la plupart de ces fonctionnalités sont disponibles dans l'exemple de script d'entraînement [ici](https://github.com/huggingface/diffusers/raw/main/examples/unconditional_image_generation/train_unconditional.py).. Vous pouvez télécharger le fichier comme suit : ```py !wget https://github.com/huggingface/diffusers/raw/main/examples/unconditional_image_generation/train_unconditional.py ``` Ouvrez le fichier et vous verrez où le modèle est défini et quels sont les paramètres disponibles. Nous exécutons le script à l'aide de la commande suivante : ```py # Donnons un nom à notre nouveau modèle pour le Hub model_name = "sd-class-butterflies-64" hub_model_id = get_full_repo_name(model_name) ``` ```py !accelerate launch train_unconditional.py \ --dataset_name="huggan/smithsonian_butterflies_subset" \ --resolution=64 \ --output_dir={model_name} \ --train_batch_size=32 \ --num_epochs=50 \ --gradient_accumulation_steps=1 \ --learning_rate=1e-4 \ --lr_warmup_steps=500 \ --mixed_precision="no" ``` Comme précédemment, poussons le modèle vers le *Hub* et créons une belle carte de modèle (et n'hésitez pas à l'éditer comme vous le souhaitez !): ```py create_repo(hub_model_id) api = HfApi() api.upload_folder( folder_path=f"{model_name}/scheduler", path_in_repo="", repo_id=hub_model_id ) api.upload_folder( folder_path=f"{model_name}/unet", path_in_repo="", repo_id=hub_model_id ) api.upload_file( path_or_fileobj=f"{model_name}/model_index.json", path_in_repo="model_index.json", repo_id=hub_model_id, ) content = f""" --- license: mit tags: - pytorch - diffusers - unconditional-image-generation - diffusion-models-class --- # Model Card for Unit 1 of the [Diffusion Models Class 🧨](https://github.com/huggingface/diffusion-models-class) This model is a diffusion model for unconditional image generation of cute 🦋. ## Usage ```python from diffusers import DDPMPipeline pipeline = DDPMPipeline.from_pretrained('{hub_model_id}') image = pipeline().images[0] image ``` """ card = ModelCard(content) card.push_to_hub(hub_model_id) ``` Environ 45 minutes plus tard, voici le résultat : ```py pipeline = DDPMPipeline.from_pretrained(hub_model_id).to(device) images = pipeline(batch_size=8).images make_grid(images) ``` <div class="flex justify-center"> <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary.svg" alt="Bref aperçu du contenu du cours."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter1/summary-dark.svg" alt="Bref aperçu des différents chapitres du cours."> </div> <Tip> ✏️ *A votre tour !* essayez de trouver des paramètres d'entraînement/de modèle qui donnent de bons résultats en un minimum de temps, et partagez vos résultats avec la communauté. Fouillez dans le script pour voir si vous pouvez comprendre le code, et demandez des éclaircissements sur tout ce qui vous semble confus. </Tip> # Pistes d'approfondissement Nous espérons vous avoir donné un avant-goût de ce que vous pouvez faire avec la bibliothèque 🤗 *Diffusers* ! Voici quelques pistes possibles pour la suite : - Essayez d'entraîner un modèle de diffusion inconditionnel sur un nouveau jeu de données. Points bonus si vous en [créez un vous-même](https://huggingface.co/docs/datasets/image_dataset). Vous pouvez trouver d'excellents jeux de données d'images pour cette tâche dans l'[organisation HugGan](https://huggingface.co/huggan) sur le *Hub*. Assurez-vous simplement de les sous-échantillonner si vous ne voulez pas attendre très longtemps pour que le modèle s'entraîne ! - Essayez DreamBooth pour créer votre propre pipeline de Stable Diffusion personnalisé en utilisant ce [Space]((https://huggingface.co/spaces/multimodalart/dreambooth-training) ou ce [*notebook*](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/sd_dreambooth_training.ipynb). - Modifiez le script d'entraînement pour explorer différents hyperparamètres UNet (nombre de couches, canaux, etc.), différents schémas de bruit, etc. - Consultez le *notebook* [*Implémentation à partir de 0*](https://github.com/huggingface/diffusion-models-class/blob/main/fr/unit1/diffusion_models_from_scratch.ipynb) pour une approche différente des idées fondamentales que nous avons abordées dans cette unité.

diffusion-models-class/units/fr/unit1/2.mdx/0

<jupyter_start><jupyter_text>Bias et limitations Installez la bibliothèque 🤗 *Transformers* pour exécuter ce *notebook*.<jupyter_code>!pip install transformers[sentencepiece] from transformers import pipeline unmasker = pipeline("fill-mask", model="camembert-base") result = unmasker("Cet homme travaille comme <mask>.") print([r["token_str"] for r in result]) result = unmasker("Cette femme travaille comme <mask>.") print([r["token_str"] for r in result])<jupyter_output><empty_output>

notebooks/course/fr/chapter1/section8.ipynb/0

<jupyter_start><jupyter_text>Finetuner un modèle avec Keras Installez les bibliothèques 🤗 Transformers et 🤗 Datasets pour exécuter ce notebook.<jupyter_code>!pip install datasets transformers[sentencepiece] from datasets import load_dataset from transformers import AutoTokenizer, DataCollatorWithPadding import numpy as np raw_datasets = load_dataset("paws-x", "fr") checkpoint = "camembert-base" tokenizer = AutoTokenizer.from_pretrained(checkpoint) def tokenize_function(example): return tokenizer(example["sentence1"], example["sentence2"], truncation=True) tokenized_datasets = raw_datasets.map(tokenize_function, batched=True) data_collator = DataCollatorWithPadding(tokenizer=tokenizer, return_tensors="tf") tf_train_dataset = tokenized_datasets["train"].to_tf_dataset( columns=["attention_mask", "input_ids", "token_type_ids"], label_cols=["labels"], shuffle=True, collate_fn=data_collator, batch_size=8, ) tf_validation_dataset = tokenized_datasets["validation"].to_tf_dataset( columns=["attention_mask", "input_ids", "token_type_ids"], label_cols=["labels"], shuffle=False, collate_fn=data_collator, batch_size=8, ) from transformers import TFAutoModelForSequenceClassification model = TFAutoModelForSequenceClassification.from_pretrained(checkpoint, num_labels=2) from tensorflow.keras.losses import SparseCategoricalCrossentropy model.compile( optimizer="adam", loss=SparseCategoricalCrossentropy(from_logits=True), metrics=["accuracy"], ) model.fit( tf_train_dataset, validation_data=tf_validation_dataset, ) from tensorflow.keras.optimizers.schedules import PolynomialDecay batch_size = 8 num_epochs = 3 # Le nombre d'étapes d'entraînement est le nombre d'échantillons dans le jeu de données, divisé par la taille du batch, puis multiplié par le nombre total d'époques. # Notez que le jeu de données tf_train_dataset est ici un batch de données tf.data.Dataset # et non le jeu de données original Hugging Face, donc sa len() est déjà num_samples // batch_size. num_train_steps = len(tf_train_dataset) * num_epochs lr_scheduler = PolynomialDecay( initial_learning_rate=5e-5, end_learning_rate=0.0, decay_steps=num_train_steps ) from tensorflow.keras.optimizers import Adam opt = Adam(learning_rate=lr_scheduler) import tensorflow as tf model = TFAutoModelForSequenceClassification.from_pretrained(checkpoint, num_labels=2) loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True) model.compile(optimizer=opt, loss=loss, metrics=["accuracy"]) model.fit(tf_train_dataset, validation_data=tf_validation_dataset, epochs=3) preds = model.predict(tf_validation_dataset)["logits"] class_preds = np.argmax(preds, axis=1) print(preds.shape, class_preds.shape) from datasets import load_metric metric = load_metric("glue", "mrpc") metric.compute(predictions=class_preds, references=raw_datasets["validation"]["label"])<jupyter_output><empty_output>

notebooks/course/fr/chapter3/section3_tf.ipynb/0

<jupyter_start><jupyter_text>Fast tokenizers in the QA pipeline (PyTorch) Installez les bibliothèques 🤗 *Transformers* et 🤗 *Datasets* pour exécuter ce *notebook*.<jupyter_code>!pip install datasets transformers[sentencepiece] from transformers import pipeline question_answerer = pipeline("question-answering", model="etalab-ia/camembert-base-squadFR-fquad-piaf") context = """ 🤗 Transformers s'appuie sur les trois bibliothèques d'apprentissage profond les plus populaires (Jax, PyTorch et TensorFlow) avec une intégration transparente entre elles. C'est simple d'entraîner vos modèles avec l'une avant de les charger pour l'inférence avec l'autre. """ question = "Quelles bibliothèques d'apprentissage profond derrière 🤗 Transformers ?" question_answerer(question=question, context=context) long_context = """ 🤗 Transformers : l'état de l'art du NLP 🤗 Transformers fournit des milliers de modèles pré-entraînés pour effectuer des tâches sur des textes telles que la classification, l'extraction d'informations, la réponse à des questions, le résumé de textes, la traduction, la génération de texte et plus encore dans plus de 100 langues. Son objectif est de rendre le traitement automatique des langues de pointe plus facile à utiliser pour tout le monde. 🤗 Transformers fournit des API permettant de télécharger et d'utiliser rapidement ces modèles pré-entraînés sur un texte donné, de les affiner sur vos propres ensembles de données et de les partager avec la communauté sur notre site Web. puis de les partager avec la communauté sur notre hub de modèles. En même temps, chaque module python définissant une architecture est entièrement autonome et peut être modifié pour permettre des expériences de recherche rapides. peut être modifié pour permettre des expériences de recherche rapides. Pourquoi devrais-je utiliser des transformateurs ? 1. Des modèles de pointe faciles à utiliser : - Haute performance sur les tâches NLU et NLG. - Faible barrière à l'entrée pour les éducateurs et les praticiens. - Peu d'abstractions pour l'utilisateur avec seulement trois classes à apprendre. - Une API unifiée pour utiliser tous nos modèles pré-entraînés. - Des coûts de calcul plus faibles, une empreinte carbone réduite : 2. Les chercheurs peuvent partager les modèles formés au lieu de toujours les reformer. - Les praticiens peuvent réduire le temps de calcul et les coûts de production. - Des dizaines d'architectures avec plus de 10 000 modèles pré-formés, certains dans plus de 100 langues. 3. Choisissez le cadre approprié pour chaque étape de la vie d'un modèle : - Entraînez des modèles de pointe en 3 lignes de code. - Déplacez un seul modèle entre les frameworks TF2.0/PyTorch à volonté. - Choisissez de manière transparente le bon framework pour l'entraînement, l'évaluation et la production. 4. Adaptez facilement un modèle ou un exemple à vos besoins : - Nous fournissons des exemples pour chaque architecture afin de reproduire les résultats publiés par ses auteurs originaux. - Les éléments internes des modèles sont exposés de manière aussi cohérente que possible. - Les fichiers de modèles peuvent être utilisés indépendamment de la bibliothèque pour des expériences rapides. 🤗 Transformers s'appuie sur les trois bibliothèques d'apprentissage profond les plus populaires (Jax, PyTorch et TensorFlow) avec une intégration parfaite entre elles. Il est simple d'entraîner vos modèles avec l'une avant de les charger pour l'inférence avec l'autre. """ question_answerer(question=question, context=long_context) from transformers import AutoTokenizer, AutoModelForQuestionAnswering model_checkpoint = "etalab-ia/camembert-base-squadFR-fquad-piaf" tokenizer = AutoTokenizer.from_pretrained(model_checkpoint) model = AutoModelForQuestionAnswering.from_pretrained(model_checkpoint) inputs = tokenizer(question, context, return_tensors="pt") outputs = model(**inputs) start_logits = outputs.start_logits end_logits = outputs.end_logits print(start_logits.shape, end_logits.shape) import torch sequence_ids = inputs.sequence_ids() # Masque tout sauf les tokens du contexte mask = [i != 1 for i in sequence_ids] # Démasque le token [CLS] mask[0] = False mask = torch.tensor(mask)[None] start_logits[mask] = -10000 end_logits[mask] = -10000 start_probabilities = torch.nn.functional.softmax(start_logits, dim=-1)[0] end_probabilities = torch.nn.functional.softmax(end_logits, dim=-1)[0] scores = start_probabilities[:, None] * end_probabilities[None, :] scores = torch.triu(scores) max_index = scores.argmax().item() start_index = max_index // scores.shape[1] end_index = max_index % scores.shape[1] print(scores[start_index, end_index]) inputs_with_offsets = tokenizer(question, context, return_offsets_mapping=True) offsets = inputs_with_offsets["offset_mapping"] start_char, _ = offsets[start_index] _, end_char = offsets[end_index] answer = context[start_char:end_char] result = { "answer": answer, "start": start_char, "end": end_char, "score": scores[start_index, end_index], } print(result) inputs = tokenizer(question, long_context) print(len(inputs["input_ids"])) inputs = tokenizer(question, long_context, max_length=384, truncation="only_second") print(tokenizer.decode(inputs["input_ids"])) sentence = "Cette phrase n'est pas trop longue mais nous allons la diviser quand même." inputs = tokenizer( sentence, truncation=True, return_overflowing_tokens=True, max_length=6, stride=2 ) for ids in inputs["input_ids"]: print(tokenizer.decode(ids)) print(inputs.keys()) print(inputs["overflow_to_sample_mapping"]) sentences = [ "Cette phrase n'est pas trop longue mais on va quand même la diviser.", "Cette phrase est plus courte mais sera quand même divisée.", ] inputs = tokenizer( sentences, truncation=True, return_overflowing_tokens=True, max_length=6, stride=2 ) print(inputs["overflow_to_sample_mapping"]) inputs = tokenizer( question, long_context, stride=128, max_length=384, padding="longest", truncation="only_second", return_overflowing_tokens=True, return_offsets_mapping=True, ) _ = inputs.pop("overflow_to_sample_mapping") offsets = inputs.pop("offset_mapping") inputs = inputs.convert_to_tensors("pt") print(inputs["input_ids"].shape) outputs = model(**inputs) start_logits = outputs.start_logits end_logits = outputs.end_logits print(start_logits.shape, end_logits.shape) sequence_ids = inputs.sequence_ids() # Masque tout sauf les tokens du contexte. mask = [i != 1 for i in sequence_ids] # Démasquer le token [CLS] mask[0] = False # Masquer tous les tokens [PAD] mask = torch.logical_or(torch.tensor(mask)[None], (inputs["attention_mask"] == 0)) start_logits[mask] = -10000 end_logits[mask] = -10000 start_probabilities = torch.nn.functional.softmax(start_logits, dim=-1) end_probabilities = torch.nn.functional.softmax(end_logits, dim=-1) candidates = [] for start_probs, end_probs in zip(start_probabilities, end_probabilities): scores = start_probs[:, None] * end_probs[None, :] idx = torch.triu(scores).argmax().item() start_idx = idx // scores.shape[0] end_idx = idx % scores.shape[0] score = scores[start_idx, end_idx].item() candidates.append((start_idx, end_idx, score)) print(candidates) for candidate, offset in zip(candidates, offsets): start_token, end_token, score = candidate start_char, _ = offset[start_token] _, end_char = offset[end_token] answer = long_context[start_char:end_char] result = {"answer": answer, "start": start_char, "end": end_char, "score": score} print(result)<jupyter_output><empty_output>

notebooks/course/fr/chapter6/section3b_pt.ipynb/0

<jupyter_start><jupyter_text>Entraîner un modèle de langage causal de zéro (TensorFlow)Ici nous entraînons un modèle à générer du code Python. Le Python utilisant des fonctions basées sur des mots anglais, nous gardons un gpt-2 anglais dans l'optique d'obtenir de meilleures performances que ce que l'on pourrait s'attendre en utilisant un gpt-2 en français. Installez les bibliothèques 🤗 *Datasets* et 🤗 *Transformers* pour exécuter ce *notebook*.<jupyter_code>!pip install datasets transformers[sentencepiece] !apt install git-lfs<jupyter_output><empty_output><jupyter_text>Vous aurez besoin de configurer git, adaptez votre email et votre nom dans la cellule suivante.<jupyter_code>!git config --global user.email "you@example.com" !git config --global user.name "Your Name"<jupyter_output><empty_output><jupyter_text>Vous devrez également être connecté au Hub d'Hugging Face. Exécutez ce qui suit et entrez vos informations d'identification.<jupyter_code>from huggingface_hub import notebook_login notebook_login() def any_keyword_in_string(string, keywords): for keyword in keywords: if keyword in string: return True return False filters = ["pandas", "sklearn", "matplotlib", "seaborn"] example_1 = "import numpy as np" example_2 = "import pandas as pd" print( any_keyword_in_string(example_1, filters), any_keyword_in_string(example_2, filters) ) from collections import defaultdict from tqdm import tqdm from datasets import Dataset def filter_streaming_dataset(dataset, filters): filtered_dict = defaultdict(list) total = 0 for sample in tqdm(iter(dataset)): total += 1 if any_keyword_in_string(sample["content"], filters): for k, v in sample.items(): filtered_dict[k].append(v) print(f"{len(filtered_dict['content'])/total:.2%} of data after filtering.") return Dataset.from_dict(filtered_dict) # Cette cellule prendra beaucoup de temps à s'exécuter, donc vous devriez la sauter et aller à la suivante ! from datasets import load_dataset split = "train" # "valid" filters = ["pandas", "sklearn", "matplotlib", "seaborn"] data = load_dataset(f"transformersbook/codeparrot-{split}", split=split, streaming=True) filtered_data = filter_streaming_dataset(data, filters) from datasets import load_dataset, DatasetDict ds_train = load_dataset("huggingface-course/codeparrot-ds-train", split="train") ds_valid = load_dataset("huggingface-course/codeparrot-ds-valid", split="validation") raw_datasets = DatasetDict( { "train": ds_train, # .shuffle().select(range(50000)), "valid": ds_valid, # .shuffle().select(range(500)) } ) raw_datasets for key in raw_datasets["train"][0]: print(f"{key.upper()}: {raw_datasets['train'][0][key][:200]}") from transformers import AutoTokenizer context_length = 128 tokenizer = AutoTokenizer.from_pretrained("huggingface-course/code-search-net-tokenizer") outputs = tokenizer( raw_datasets["train"][:2]["content"], truncation=True, max_length=context_length, return_overflowing_tokens=True, return_length=True, ) print(f"Input IDs length: {len(outputs['input_ids'])}") print(f"Input chunk lengths: {(outputs['length'])}") print(f"Chunk mapping: {outputs['overflow_to_sample_mapping']}") def tokenize(element): outputs = tokenizer( element["content"], truncation=True, max_length=context_length, return_overflowing_tokens=True, return_length=True, ) input_batch = [] for length, input_ids in zip(outputs["length"], outputs["input_ids"]): if length == context_length: input_batch.append(input_ids) return {"input_ids": input_batch} tokenized_datasets = raw_datasets.map( tokenize, batched=True, remove_columns=raw_datasets["train"].column_names ) tokenized_datasets from transformers import AutoTokenizer, TFGPT2LMHeadModel, AutoConfig config = AutoConfig.from_pretrained( "gpt2", vocab_size=len(tokenizer), n_ctx=context_length, bos_token_id=tokenizer.bos_token_id, eos_token_id=tokenizer.eos_token_id, ) model = TFGPT2LMHeadModel(config) model(model.dummy_inputs) # Construit le modèle model.summary() from transformers import DataCollatorForLanguageModeling tokenizer.pad_token = tokenizer.eos_token data_collator = DataCollatorForLanguageModeling(tokenizer, mlm=False, return_tensors="tf") out = data_collator([tokenized_datasets["train"][i] for i in range(5)]) for key in out: print(f"{key} shape: {out[key].shape}") tf_train_dataset = model.prepare_tf_dataset( train_dataset, collate_fn=data_collator, shuffle=True, batch_size=16, ) tf_eval_dataset = model.prepare_tf_dataset( tokenized_dataset["valid"], collate_fn=data_collator, shuffle=False, batch_size=32, ) from huggingface_hub import notebook_login notebook_login() from transformers import create_optimizer import tensorflow as tf num_train_steps = len(tf_train_dataset) optimizer, schedule = create_optimizer( init_lr=5e-5, num_warmup_steps=1_000, num_train_steps=num_train_steps, weight_decay_rate=0.01, ) model.compile(optimizer=optimizer) # Entraîner en mixed-precision float16 tf.keras.mixed_precision.set_global_policy("mixed_float16") from transformers.keras_callbacks import PushToHubCallback callback = PushToHubCallback(output_dir="codeparrot-ds", tokenizer=tokenizer) model.fit(tf_train_dataset, validation_data=tf_eval_dataset, callbacks=[callback]) from transformers import pipeline course_model = TFGPT2LMHeadModel.from_pretrained("huggingface-course/codeparrot-ds") course_tokenizer = AutoTokenizer.from_pretrained("huggingface-course/codeparrot-ds") pipe = pipeline( "text-generation", model=course_model, tokenizer=course_tokenizer, device=0 ) txt = """\ # create some data x = np.random.randn(100) y = np.random.randn(100) # create scatter plot with x, y """ print(pipe(txt, num_return_sequences=1)[0]["generated_text"]) txt = """\ # create some data x = np.random.randn(100) y = np.random.randn(100) # create dataframe from x and y """ print(pipe(txt, num_return_sequences=1)[0]["generated_text"]) txt = """\ # dataframe with profession, income and name df = pd.DataFrame({'profession': x, 'income':y, 'name': z}) # calculate the mean income per profession """ print(pipe(txt, num_return_sequences=1)[0]["generated_text"]) txt = """ # import random forest regressor from scikit-learn from sklearn.ensemble import RandomForestRegressor # fit random forest model with 300 estimators on X, y: """ print(pipe(txt, num_return_sequences=1)[0]["generated_text"])<jupyter_output><empty_output>

notebooks/course/fr/chapter7/section6_tf.ipynb/0

<jupyter_start><jupyter_text>Fine-tuning Stable Diffusion XL with DreamBooth and LoRA on a free-tier Colab Notebook 🧨In this notebook, we show how to fine-tune [Stable Diffusion XL (SDXL)](https://huggingface.co/docs/diffusers/main/en/api/pipelines/stable_diffusion/stable_diffusion_xl) with [DreamBooth](https://huggingface.co/docs/diffusers/main/en/training/dreambooth) and [LoRA](https://huggingface.co/docs/diffusers/main/en/training/lora) on a T4 GPU.SDXL consists of a much larger UNet and two text encoders that make the cross-attention context quite larger than the previous variants.So, to pull this off, we will make use of several tricks such as gradient checkpointing, mixed-precision, and 8-bit Adam. So, hang tight and let's get started 🧪 Setup 🪓<jupyter_code># Check the GPU !nvidia-smi # Install dependencies. !pip install bitsandbytes transformers accelerate peft -q<jupyter_output> ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 211.8/211.8 MB 2.9 MB/s eta 0:00:00  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 92.6/92.6 MB 9.0 MB/s eta 0:00:00  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 261.4/261.4 kB 21.5 MB/s eta 0:00:00 [?25h<jupyter_text>Make sure to install `diffusers` from `main`.<jupyter_code>!pip install git+https://github.com/huggingface/diffusers.git -q<jupyter_output>Installing build dependencies ... [?25l[?25hdone Getting requirements to build wheel ... [?25l[?25hdone Preparing metadata (pyproject.toml) ... [?25l[?25hdone Building wheel for diffusers (pyproject.toml) ... [?25l[?25hdone<jupyter_text>Download diffusers SDXL DreamBooth training script.<jupyter_code>!wget https://raw.githubusercontent.com/huggingface/diffusers/main/examples/dreambooth/train_dreambooth_lora_sdxl.py<jupyter_output>--2023-11-23 06:48:12-- https://raw.githubusercontent.com/huggingface/diffusers/main/examples/dreambooth/train_dreambooth_lora_sdxl.py Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.108.133, 185.199.109.133, 185.199.110.133, ... Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.108.133|:443... connected. HTTP request sent, awaiting response... 200 OK Length: 72845 (71K) [text/plain] Saving to: ‘train_dreambooth_lora_sdxl.py’ train_dreambooth_lo 100%[===================>] 71.14K --.-KB/s in 0.001s 2023-11-23 06:48:12 (46.6 MB/s) - ‘train_dreambooth_lora_sdxl.py’ saved [72845/72845]<jupyter_text>Dataset 🐶 **Let's get our training data!**For this example, we'll download some images from the hubIf you already have a dataset on the hub you wish to use, you can skip this part and go straight to: "Prep fortraining 💻" section, where you'll simply specify the dataset name.If your images are saved locally, and/or you want to add BLIP generated captions,pick option 1 or 2 below. **Option 1:** upload example images from your local files:<jupyter_code>import os from google.colab import files # pick a name for the image folder local_dir = "./dog/" #@param os.makedirs(local_dir) os.chdir(local_dir) # choose and upload local images into the newly created directory uploaded_images = files.upload() os.chdir("/content") # back to parent directory<jupyter_output><empty_output><jupyter_text>**Option 2:** download example images from the hub:<jupyter_code>from huggingface_hub import snapshot_download local_dir = "./dog/" snapshot_download( "diffusers/dog-example", local_dir=local_dir, repo_type="dataset", ignore_patterns=".gitattributes", )<jupyter_output><empty_output><jupyter_text>Preview the images:<jupyter_code>from PIL import Image def image_grid(imgs, rows, cols, resize=256): if resize is not None: imgs = [img.resize((resize, resize)) for img in imgs] w, h = imgs[0].size grid = Image.new("RGB", size=(cols * w, rows * h)) grid_w, grid_h = grid.size for i, img in enumerate(imgs): grid.paste(img, box=(i % cols * w, i // cols * h)) return grid import glob # change path to display images from your local dir img_paths = "./dog/*.jpeg" imgs = [Image.open(path) for path in glob.glob(img_paths)] num_imgs_to_preview = 5 image_grid(imgs[:num_imgs_to_preview], 1, num_imgs_to_preview)<jupyter_output><empty_output><jupyter_text>Generate custom captions with BLIPLoad BLIP to auto caption your images:<jupyter_code>import requests from transformers import AutoProcessor, BlipForConditionalGeneration import torch device = "cuda" if torch.cuda.is_available() else "cpu" # load the processor and the captioning model blip_processor = AutoProcessor.from_pretrained("Salesforce/blip-image-captioning-base") blip_model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base",torch_dtype=torch.float16).to(device) # captioning utility def caption_images(input_image): inputs = blip_processor(images=input_image, return_tensors="pt").to(device, torch.float16) pixel_values = inputs.pixel_values generated_ids = blip_model.generate(pixel_values=pixel_values, max_length=50) generated_caption = blip_processor.batch_decode(generated_ids, skip_special_tokens=True)[0] return generated_caption import glob from PIL import Image # create a list of (Pil.Image, path) pairs local_dir = "./dog/" imgs_and_paths = [(path,Image.open(path)) for path in glob.glob(f"{local_dir}*.jpeg")]<jupyter_output><empty_output><jupyter_text>Now let's add the concept token identifier (e.g. TOK) to each caption using a caption prefix.Feel free to change the prefix according to the concept you're training on!- for this example we can use "a photo of TOK," other options include: - For styles - "In the style of TOK" - For faces - "photo of a TOK person"- You can add additional identifiers to the prefix that can help steer the model in the right direction.-- e.g. for this example, instead of "a photo of TOK" we can use "a photo of TOK dog" / "a photo of TOK corgi dog"<jupyter_code>import json caption_prefix = "a photo of TOK dog, " #@param with open(f'{local_dir}metadata.jsonl', 'w') as outfile: for img in imgs_and_paths: caption = caption_prefix + caption_images(img[1]).split("\n")[0] entry = {"file_name":img[0].split("/")[-1], "prompt": caption} json.dump(entry, outfile) outfile.write('\n')<jupyter_output><empty_output><jupyter_text>Free some memory:<jupyter_code>import gc # delete the BLIP pipelines and free up some memory del blip_processor, blip_model gc.collect() torch.cuda.empty_cache()<jupyter_output><empty_output><jupyter_text>Prep for training 💻 Initialize `accelerate`:<jupyter_code>import locale locale.getpreferredencoding = lambda: "UTF-8" !accelerate config default<jupyter_output>accelerate configuration saved at /root/.cache/huggingface/accelerate/default_config.yaml<jupyter_text>Log into your Hugging Face accountPass [your **write** access token](https://huggingface.co/settings/tokens) so that we can push the trained checkpoints to the Hugging Face Hub:<jupyter_code>from huggingface_hub import notebook_login notebook_login()<jupyter_output><empty_output><jupyter_text>Train! 🔬 Set Hyperparameters ⚡To ensure we can DreamBooth with LoRA on a heavy pipeline like Stable Diffusion XL, we're using:* Gradient checkpointing (`--gradient_accumulation_steps`)* 8-bit Adam (`--use_8bit_adam`)* Mixed-precision training (`--mixed-precision="fp16"`) Launch training 🚀🚀🚀 To allow for custom captions we need to install the `datasets` library, you can skip that if you want to train solely with `--instance_prompt`.In that case, specify `--instance_data_dir` instead of `--dataset_name`<jupyter_code>!pip install datasets -q<jupyter_output> ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 485.6/485.6 kB 7.9 MB/s eta 0:00:00  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 110.5/110.5 kB 13.0 MB/s eta 0:00:00  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 134.8/134.8 kB 16.0 MB/s eta 0:00:00  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 134.3/134.3 kB 15.4 MB/s eta 0:00:00 [?25h<jupyter_text>- Use `--output_dir` to specify your LoRA model repository name! - Use `--caption_column` to specify name of the cpation column in your dataset. In this example we used "prompt" to save our captions in the metadata file, change this according to your needs.<jupyter_code>#!/usr/bin/env bash !accelerate launch train_dreambooth_lora_sdxl.py \ --pretrained_model_name_or_path="stabilityai/stable-diffusion-xl-base-1.0" \ --pretrained_vae_model_name_or_path="madebyollin/sdxl-vae-fp16-fix" \ --dataset_name="dog" \ --output_dir="corgy_dog_LoRA" \ --caption_column="prompt"\ --mixed_precision="fp16" \ --instance_prompt="a photo of TOK dog" \ --resolution=1024 \ --train_batch_size=1 \ --gradient_accumulation_steps=3 \ --gradient_checkpointing \ --learning_rate=1e-4 \ --snr_gamma=5.0 \ --lr_scheduler="constant" \ --lr_warmup_steps=0 \ --mixed_precision="fp16" \ --use_8bit_adam \ --max_train_steps=500 \ --checkpointing_steps=717 \ --seed="0"<jupyter_output>WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for: PyTorch 2.1.0+cu121 with CUDA 1201 (you have 2.1.0+cu118) Python 3.10.13 (you have 3.10.12) Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers) Memory-efficient attention, SwiGLU, sparse and more won't be available. Set XFORMERS_MORE_DETAILS=1 for more details 2023-11-23 07:06:49.633870: E tensorflow/compiler/xla/stream_executor/cuda/cuda_dnn.cc:9342] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered 2023-11-23 07:06:49.633948: E tensorflow/compiler/xla/stream_executor/cuda/cuda_fft.cc:609] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered 2023-11-23 07:06:49.638631: E tensorflow/compiler/xla/stream_executor/cuda/cuda_blas.cc:1518] Unable to register cuBLAS factory: Attempting to register factory for plugin c[...]<jupyter_text>Save your model to the hub and check it out 🔥<jupyter_code>from huggingface_hub import whoami from pathlib import Path #@markdown make sure the `output_dir` you specify here is the same as the one used for training output_dir = "corgy_dog_LoRA" #@param username = whoami(token=Path("/root/.cache/huggingface/"))["name"] repo_id = f"{username}/{output_dir}" # @markdown Sometimes training finishes succesfuly (i.e. a **.safetensores** file with the LoRA weights saved properly to your local `output_dir`) but there's not enough RAM in the free tier to push the model to the hub 🙁 # @markdown # @markdown To mitigate this, run this cell with your training arguments to make sure your model is uploaded! 🤗 # push to the hub🔥 from train_dreambooth_lora_sdxl import save_model_card from huggingface_hub import upload_folder, create_repo repo_id = create_repo(repo_id, exist_ok=True).repo_id # change the params below according to your training arguments save_model_card( repo_id = repo_id, images=[], base_model="stabilityai/stable-diffusion-xl-base-1.0", train_text_encoder=False, instance_prompt="a photo of TOK dog", validation_prompt=None, repo_folder=output_dir, vae_path="madebyollin/sdxl-vae-fp16-fix", ) upload_folder( repo_id=repo_id, folder_path=output_dir, commit_message="End of training", ignore_patterns=["step_*", "epoch_*"], ) from IPython.display import display, Markdown link_to_model = f"https://huggingface.co/{repo_id}" display(Markdown("### Your model has finished training.\nAccess it here: {}".format(link_to_model)))<jupyter_output><empty_output><jupyter_text>Let's generate some images with it! Inference 🐕<jupyter_code>import torch from diffusers import DiffusionPipeline, AutoencoderKL vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16) pipe = DiffusionPipeline.from_pretrained( "stabilityai/stable-diffusion-xl-base-1.0", vae=vae, torch_dtype=torch.float16, variant="fp16", use_safetensors=True ) pipe.load_lora_weights(repo_id) _ = pipe.to("cuda") prompt = "a photo of TOK dog in a bucket at the beach" # @param image = pipe(prompt=prompt, num_inference_steps=25).images[0] image<jupyter_output><empty_output>

notebooks/diffusers/SDXL_DreamBooth_LoRA_.ipynb/0

<jupyter_start><jupyter_text>Stable Conceptualizer - Stable Diffusion using learned conceptsThe Stable Conceptualizer enables you to use pre-learned concepts on Stable Diffusion via textual-inversion using 🤗 Hugging Face [🧨 Diffusers library](https://github.com/huggingface/diffusers). Navigate the [library of pre-learned concepts](https://huggingface.co/sd-concepts-library) here. For teaching the model new concepts using Textual Inversion, [use this notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/sd_textual_inversion_training.ipynb). Initial setup<jupyter_code>#@title Install the required libs !pip install -qq diffusers==0.16.1 transformers ftfy accelerate #@title Login to the Hugging Face Hub from huggingface_hub import notebook_login notebook_login() #@title Import required libraries import os import torch import PIL from PIL import Image from diffusers import StableDiffusionPipeline from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer def image_grid(imgs, rows, cols): assert len(imgs) == rows*cols w, h = imgs[0].size grid = Image.new('RGB', size=(cols*w, rows*h)) grid_w, grid_h = grid.size for i, img in enumerate(imgs): grid.paste(img, box=(i%cols*w, i//cols*h)) return grid<jupyter_output><empty_output><jupyter_text>Run Stable Diffusion with pre-trained Learned ConceptsIf you want to teach Stable Diffusion your own concepts, [use this notebook]().<jupyter_code>#@markdown `pretrained_model_name_or_path` which Stable Diffusion checkpoint you want to use. This should match the one used for training the embeddings. pretrained_model_name_or_path = "CompVis/stable-diffusion-v1-4" #@param {type:"string"} #@title Load your concept here #@markdown Enter the `repo_id` for a concept you like (you can find pre-learned concepts in the public [SD Concepts Library](https://huggingface.co/sd-concepts-library)) repo_id_embeds = "sd-concepts-library/cat-toy" #@param {type:"string"} #@title Load the Stable Diffusion pipeline pipe = StableDiffusionPipeline.from_pretrained( pretrained_model_name_or_path, torch_dtype=torch.float16 ).to("cuda") #@title Load the concept into pipeline pipe.load_textual_inversion(repo_id_embeds) #@title Run the Stable Diffusion pipeline #@markdown Don't forget to use the placeholder token in your prompt prompt = "a grafitti in a favela wall with a <cat-toy> on it" #@param {type:"string"} num_samples = 2 #@param {type:"number"} num_rows = 2 #@param {type:"number"} all_images = [] for _ in range(num_rows): images = pipe(prompt, num_images_per_prompt=num_samples, num_inference_steps=50, guidance_scale=7.5).images all_images.extend(images) grid = image_grid(all_images, num_samples, num_rows) grid<jupyter_output><empty_output>

notebooks/diffusers/stable_conceptualizer_inference.ipynb/0

# this is a demo of inference of IDEFICS-9B using 4bit-quantization which needs about 7GB of GPU memory # which makes it possible to run even on Google Colab import torch from transformers import IdeficsForVisionText2Text, AutoProcessor, BitsAndBytesConfig device = "cuda" if torch.cuda.is_available() else "cpu" checkpoint = "HuggingFaceM4/idefics-9b" #checkpoint = "HuggingFaceM4/tiny-random-idefics" quantization_config = BitsAndBytesConfig( load_in_4bit=True, bnb_4bit_compute_dtype="float16", ) model = IdeficsForVisionText2Text.from_pretrained(checkpoint, quantization_config=quantization_config, device_map="auto") processor = AutoProcessor.from_pretrained(checkpoint) prompts = [ "Instruction: provide an answer to the question. Use the image to answer.\n", "https://hips.hearstapps.com/hmg-prod/images/cute-photos-of-cats-in-grass-1593184777.jpg", "Question: What's on the picture? Answer: \n" ] inputs = processor(prompts, return_tensors="pt") generated_ids = model.generate(**inputs, max_length=150) generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True) print(generated_text[0])

notebooks/examples/idefics/inference_4bit.py/0

<jupyter_start><jupyter_text>If you're opening this Notebook on colab, you will probably need to install 🤗 Transformers and 🤗 Datasets. Uncomment the following cell and run it.<jupyter_code>#! pip install transformers #! pip install datasets #! pip install huggingface_hub<jupyter_output><empty_output><jupyter_text>If you're opening this notebook locally, make sure your environment has an install from the latest version of those libraries.To be able to share your model with the community and generate results like the one shown in the picture below via the inference API, there are a few more steps to follow.First you have to store your authentication token from the Hugging Face website (sign up [here](https://huggingface.co/join) if you haven't already!) then run the following cell and input your token:<jupyter_code>from huggingface_hub import notebook_login notebook_login()<jupyter_output><empty_output><jupyter_text>Then you need to install Git-LFS and setup Git if you haven't already. Uncomment the following instructions and adapt with your name and email:<jupyter_code># !apt install git-lfs # !git config --global user.email "you@example.com" # !git config --global user.name "Your Name"<jupyter_output><empty_output><jupyter_text>Make sure your version of Transformers is at least 4.16.0 since some of the functionality we use was only introduced in that version.<jupyter_code>import transformers print(transformers.__version__)<jupyter_output>4.21.0.dev0<jupyter_text>You can find a script version of this notebook to fine-tune your model in a distributed fashion using multiple GPUs or TPUs [here](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/language-modeling). We also quickly upload some telemetry - this tells us which examples and software versions are getting used so we know where to prioritize our maintenance efforts. We don't collect (or care about) any personally identifiable information, but if you'd prefer not to be counted, feel free to skip this step or delete this cell entirely.<jupyter_code>from transformers.utils import send_example_telemetry send_example_telemetry("language_modeling_notebook", framework="tensorflow")<jupyter_output><empty_output><jupyter_text>Fine-tuning a language model In this notebook, we'll see how to fine-tune one of the [🤗 Transformers](https://github.com/huggingface/transformers) model on a language modeling task. We will cover two types of language modeling tasks which are:- Causal language modeling: the model has to predict the next token in the sentence (so the labels are the same as the inputs shifted to the right). To make sure the model does not cheat, its attention computations are masked so that tokens cannot attend to tokens to their right, as this would result in label leakage.- Masked language modeling: the model has to predict some tokens that are masked in the input. It still has access to the whole sentence, so it can use the tokens before and after the masked tokens to predict their value.We will see how to easily load and preprocess the dataset for each one of those tasks, and how to use Keras to fine-tune a model on it.A script version of this notebook you can directly run on a distributed environment or on TPU is available in our [examples folder](https://github.com/huggingface/transformers/tree/master/examples). Preparing the dataset For each of those tasks, we will use the [Wikitext 2]() dataset as an example. You can load it very easily with the 🤗 Datasets library.<jupyter_code>from datasets import load_dataset datasets = load_dataset("wikitext", "wikitext-2-raw-v1")<jupyter_output>Reusing dataset wikitext (/home/matt/.cache/huggingface/datasets/wikitext/wikitext-2-raw-v1/1.0.0/a241db52902eaf2c6aa732210bead40c090019a499ceb13bcbfa3f8ab646a126)<jupyter_text>You can replace the dataset above with any dataset hosted on [the hub](https://huggingface.co/datasets) or use your own files. Just uncomment the following cell and replace the paths with your own input files:<jupyter_code># datasets = load_dataset("text", data_files={"train": path_to_train.txt, "validation": path_to_validation.txt}<jupyter_output><empty_output><jupyter_text>You can also load datasets from a csv or a JSON file, see the [full documentation](https://huggingface.co/docs/datasets/loading_datasets.htmlfrom-local-files) for more information. To access an actual element, you need to select a split first, then give an index:<jupyter_code>datasets["train"][10]<jupyter_output><empty_output><jupyter_text>To get a sense of what the data looks like, the following function will show some examples picked randomly in the dataset.<jupyter_code>from datasets import ClassLabel import random import pandas as pd from IPython.display import display, HTML def show_random_elements(dataset, num_examples=10): assert num_examples <= len( dataset ), "Can't pick more elements than there are in the dataset." picks = [] for _ in range(num_examples): pick = random.randint(0, len(dataset) - 1) while pick in picks: pick = random.randint(0, len(dataset) - 1) picks.append(pick) df = pd.DataFrame(dataset[picks]) for column, typ in dataset.features.items(): if isinstance(typ, ClassLabel): df[column] = df[column].transform(lambda i: typ.names[i]) display(HTML(df.to_html())) show_random_elements(datasets["train"])<jupyter_output><empty_output><jupyter_text>As we can see, some of the texts are a full paragraph of a Wikipedia article while others are just titles or empty lines. Causal Language modeling For causal language modeling (CLM) we are going to take all the texts in our dataset, tokenize them and concatenate them. Then we will split them into examples of a fixed sequence length. This way the model will receive chunks of contiguous text that may look like:```part of text 1```or ```end of text 1 [BOS_TOKEN] beginning of text 2```depending on whether they span multiple original texts or not. The labels will be the same as the inputs, shifted to the right.We will use the [`distilgpt2`](https://huggingface.co/distilgpt2) model for this example. You can pick any of the checkpoints listed [here](https://huggingface.co/models?filter=causal-lm) instead:<jupyter_code>model_checkpoint = "distilgpt2"<jupyter_output><empty_output><jupyter_text>To tokenize all our texts with the same vocabulary that was used when training the model, we have to download a pretrained tokenizer. This is all done by the `AutoTokenizer` class:<jupyter_code>from transformers import AutoTokenizer tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)<jupyter_output><empty_output><jupyter_text>We can now call the tokenizer on all our texts. This is very simple, using the [`map`](https://huggingface.co/docs/datasets/package_reference/main_classes.htmldatasets.Dataset.map) method from the Datasets library. First we define a function that calls the tokenizer on our texts:<jupyter_code>def tokenize_function(examples): return tokenizer(examples["text"])<jupyter_output><empty_output><jupyter_text>Then we apply it to all the splits in our `datasets` object, using `batched=True` and 4 processes to speed up the preprocessing. We won't need the `text` column afterward, so we discard it.<jupyter_code>tokenized_datasets = datasets.map( tokenize_function, batched=True, num_proc=4, remove_columns=["text"] )<jupyter_output><jupyter_text>If we now look at an element of our datasets, we will see the text have been replaced by the `input_ids` the model will need:<jupyter_code>tokenized_datasets["train"][1]<jupyter_output><empty_output><jupyter_text>Now for the harder part: We need to concatenate all our texts together, and then split the result into chunks of a fixed size, which we will call `block_size`. To do this, we will use the `map` method again, with the option `batched=True`. When we use `batched=True`, the function we pass to `map()` will be passed multiple inputs at once, allowing us to group them into more or fewer examples than we had in the input. This allows us to create our new fixed-length samples.We can use any `block_size` up to the the maximum length our model was pretrained with, which for models in the `gpt2` family is usually something in the range 512-1024. This might be a bit too big to fit in your GPU RAM, though, so let's use something a bit smaller: 128.<jupyter_code># block_size = tokenizer.model_max_length block_size = 128<jupyter_output><empty_output><jupyter_text>Then we write the preprocessing function that will group our texts:<jupyter_code>def group_texts(examples): # Concatenate all texts. concatenated_examples = {k: sum(examples[k], []) for k in examples.keys()} total_length = len(concatenated_examples[list(examples.keys())[0]]) # We drop the small remainder, though you could add padding instead if the model supports it # In this, as in all things, we advise you to follow your heart total_length = (total_length // block_size) * block_size # Split by chunks of max_len. result = { k: [t[i : i + block_size] for i in range(0, total_length, block_size)] for k, t in concatenated_examples.items() } result["labels"] = result["input_ids"].copy() return result<jupyter_output><empty_output><jupyter_text>Note that we duplicate the inputs for our labels, without shifting them, even though we told you the labels need to be shifted! This is because CausalLM models in the 🤗 Transformers library automatically apply right-shifting to the inputs, so we don't need to do it manually.Also note that by default, the `map` method will send a batch of 1,000 examples to be treated by the preprocessing function. So here, we will drop the remainder to make the concatenated tokenized texts a multiple of `block_size` every 1,000 examples. You can adjust this behavior by passing a higher batch size (which will also be processed slower). You can also speed-up the preprocessing by using multiprocessing:<jupyter_code>lm_datasets = tokenized_datasets.map( group_texts, batched=True, batch_size=1000, num_proc=4, )<jupyter_output><jupyter_text>And we can check our datasets have changed: now the samples contain chunks of `block_size` contiguous tokens, potentially spanning several of our original texts.<jupyter_code>tokenizer.decode(lm_datasets["train"][1]["input_ids"])<jupyter_output><empty_output><jupyter_text>Now that the data has been cleaned, we're ready to initialize our model:<jupyter_code>from transformers import TFAutoModelForCausalLM model = TFAutoModelForCausalLM.from_pretrained(model_checkpoint)<jupyter_output>2022-07-20 14:42:37.104406: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:975] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero 2022-07-20 14:42:37.141555: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:975] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero 2022-07-20 14:42:37.142578: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:975] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero 2022-07-20 14:42:37.144764: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2 FMA To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags[...]<jupyter_text>Once we've done that, it's time for our optimizer! We can initialize our `AdamWeightDecay` optimizer directly, or we can use the [`create_optimizer`](https://huggingface.co/docs/transformers/main_classes/optimizer_schedulestransformers.create_optimizer) function to generate an `AdamWeightDecay` optimizer with a learning rate schedule. In this case, we'll just stick with a constant learning rate for simplicity, so let's just use `AdamWeightDecay`.<jupyter_code>from transformers import create_optimizer, AdamWeightDecay optimizer = AdamWeightDecay(lr=2e-5, weight_decay_rate=0.01)<jupyter_output>/home/matt/miniconda3/envs/tensorflow28/lib/python3.10/site-packages/keras/optimizers/optimizer_v2/adam.py:110: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead. super(Adam, self).__init__(name, **kwargs)<jupyter_text>Next, we compile our model. Note that most Transformers models compute loss internally, so we actually don't have to specify anything for that argument! You can of course set your own loss function if you want, but by default our models will choose the 'obvious' loss that matches their task, such as cross-entropy in the case of language modelling. The built-in loss will also correctly handle things like masking the loss on padding tokens, or unlabelled tokens in the case of masked language modelling, so we recommend using it unless you're an advanced user!We also use the `jit_compile` argument to compile the model with [XLA](https://www.tensorflow.org/xla). XLA compilation adds a delay at the start of training, but this is quickly repaid by faster training iterations after that. It has one downside, though - if the shape of your input changes at all, then it will need to rerun the compilation again! This isn't a problem for us in this notebook, because all of our examples are exactly the same length. Be careful with it when that isn't true, though - if you have a variable sequence length in your batches, then you might spend more time compiling your model than actually training, especially for small datasets!If you encounter difficulties when training with XLA, it's a good idea to remove the `jit_compile` argument and see if that fixes things. In fact, when debugging, it can be helpful to skip graph compilation entirely with the `run_eagerly=True` argument to [`compile()`](https://www.tensorflow.org/api_docs/python/tf/keras/Modelcompile). This will let you identify the exact line of code where problems arise, but it will significantly reduce your performance, so make sure to remove it again when you've fixed the problem!<jupyter_code>import tensorflow as tf model.compile(optimizer=optimizer, jit_compile=True)<jupyter_output>No loss specified in compile() - the model's internal loss computation will be used as the loss. Don't panic - this is a common way to train TensorFlow models in Transformers! To disable this behaviour please pass a loss argument, or explicitly pass `loss=None` if you do not want your model to compute a loss.<jupyter_text>Next, we convert our datasets to `tf.data.Dataset`, which Keras understands natively. There are two ways to do this - we can use the slightly more low-level [`Dataset.to_tf_dataset()`](https://huggingface.co/docs/datasets/package_reference/main_classesdatasets.Dataset.to_tf_dataset) method, or we can use [`Model.prepare_tf_dataset()`](https://huggingface.co/docs/transformers/main_classes/modeltransformers.TFPreTrainedModel.prepare_tf_dataset). The main difference between these two is that the `Model` method can inspect the model to determine which column names it can use as input, which means you don't need to specify them yourself. It also supplies a data collator by default which is appropriate for most tasks.<jupyter_code>train_set = model.prepare_tf_dataset( lm_datasets["train"], shuffle=True, batch_size=16, ) validation_set = model.prepare_tf_dataset( lm_datasets["validation"], shuffle=False, batch_size=16, )<jupyter_output><empty_output><jupyter_text>Now we can train our model. We can also add a callback to sync up our model with the Hub - this allows us to resume training from other machines and even test the model's inference quality midway through training! If you don't want to do this, simply remove the callbacks argument in the call to `fit()`.<jupyter_code>from transformers.keras_callbacks import PushToHubCallback from tensorflow.keras.callbacks import TensorBoard model_name = model_checkpoint.split("/")[-1] push_to_hub_model_id = f"{model_name}-finetuned-wikitext2" tensorboard_callback = TensorBoard(log_dir="./clm_model_save/logs") push_to_hub_callback = PushToHubCallback( output_dir="./clm_model_save", tokenizer=tokenizer, hub_model_id=push_to_hub_model_id, ) callbacks = [tensorboard_callback, push_to_hub_callback] model.fit(train_set, validation_data=validation_set, epochs=1, callbacks=callbacks)<jupyter_output>/home/matt/PycharmProjects/notebooks/examples/clm_model_save is already a clone of https://huggingface.co/Rocketknight1/distilgpt2-finetuned-wikitext2. Make sure you pull the latest changes with `repo.git_pull()`. 2022-07-20 14:43:40.621522: I tensorflow/compiler/xla/service/service.cc:170] XLA service 0x55b0e1d74cd0 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices: 2022-07-20 14:43:40.621560: I tensorflow/compiler/xla/service/service.cc:178] StreamExecutor device (0): NVIDIA GeForce RTX 3090, Compute Capability 8.6 2022-07-20 14:43:40.793088: I tensorflow/compiler/mlir/tensorflow/utils/dump_mlir_util.cc:263] disabling MLIR crash reproducer, set env var `MLIR_CRASH_REPRODUCER_DIRECTORY` to enable. 2022-07-20 14:43:40.845975: W tensorflow/compiler/tf2xla/kernels/random_ops.cc:57] Warning: Using tf.random.uniform with XLA compilation will ignore seeds; consider using tf.random.stateless_uniform instead if reproducible behavior is desired. tfgpt2lm_[...]<jupyter_text>Once the training is completed, we can evaluate our model and get its cross-entropy loss on the validation set like this:<jupyter_code>eval_loss = model.evaluate(validation_set)<jupyter_output>121/121 [==============================] - 3s 22ms/step - loss: 3.6744<jupyter_text>The quality of language models is often measured in 'perplexity' rather than cross-entropy. To convert to perplexity, we simply raise e to the power of the cross-entropy loss.<jupyter_code>import math print(f"Perplexity: {math.exp(eval_loss):.2f}")<jupyter_output>Perplexity: 39.43<jupyter_text>If you saved the model with the callback, you can now share this model with all your friends, family, favorite pets: they can all load it with the identifier `"your-username/the-name-you-picked"` so for instance:```pythonfrom transformers import AutoModelForCausalLMmodel = AutoModelForCausalLM.from_pretrained("sgugger/my-awesome-model")``` Inference Now we've trained our model, let's see how we could load it and use it to generate text in future! First, let's load it from the hub. This means we can resume the code from here without needing to rerun everything above every time.<jupyter_code>from transformers import AutoTokenizer, TFAutoModelForCausalLM # You can, of course, use your own username and model name here # once you've pushed your model using the code above! checkpoint = "Rocketknight1/distilgpt2-finetuned-wikitext2" model = TFAutoModelForCausalLM.from_pretrained(checkpoint) tokenizer = AutoTokenizer.from_pretrained(checkpoint)<jupyter_output><empty_output><jupyter_text>Now, let's make up a sentence and see if the model can continue it for us!<jupyter_code>test_sentence = "The Gulf War was a conflict between"<jupyter_output><empty_output><jupyter_text>We'll need to tokenize our input(s) and then use the `generate()` method.<jupyter_code>tokenized = tokenizer(test_sentence, return_tensors="np") outputs = model.generate(**tokenized, max_length=24) print(outputs)<jupyter_output>Setting `pad_token_id` to 50256 (first `eos_token_id`) to generate sequence<jupyter_text>Those are definitely tokens! We should probably turn them back into text so that we can read them:<jupyter_code>tokenizer.decode(outputs[0])<jupyter_output><empty_output><jupyter_text>This combination of quick, fluent responses with a complete incomprehension of the actual underlying facts will be familiar to anyone who's ever scrolled through their Twitter timeline. `distilgpt2` is a very small LM compared to some others, and as you scale them up and train them for longer on larger datasets to get lower losses they often stop making errors as obvious as this one, but their tendency to confidently hallucinate their own "alternative facts" rather than admit ignorance never fully goes away, so bear this in mind before you deploy a generative language model in production! Pipeline API An alternative way to quickly perform inference with any model on the hub is to use the [Pipeline API](https://huggingface.co/docs/transformers/main_classes/pipelines), which abstracts away all the steps we did manually above. It will perform the preprocessing, forward pass and postprocessing all in a single object.Let's showcase this for our trained model:<jupyter_code>from transformers import pipeline text_generator = pipeline( "text-generation", "Rocketknight1/distilgpt2-finetuned-wikitext2", framework="tf", ) text_generator(test_sentence)<jupyter_output>Setting `pad_token_id` to 50256 (first `eos_token_id`) to generate sequence<jupyter_text>Note that we got a different (and equally historically, uh, interesting) response this time! Because [`generate()`](https://huggingface.co/docs/transformers/v4.20.1/en/main_classes/text_generationtransformers.generation_tf_utils.TFGenerationMixin.generate) samples from the model's probability outputs each time, it can return multiple different outputs from the same starting text. If consistency is important to you, you can use parameters like `temperature` to control how variable the outputs should be. Masked language modeling For masked language modeling (MLM) we are going to use the same preprocessing as before for our dataset with one additional step: we will randomly mask some tokens (by replacing them by `[MASK]`) and the labels will be adjusted to only include the masked tokens (we don't have to predict the non-masked tokens).We will use the [`distilroberta-base`](https://huggingface.co/distilroberta-base) model for this example. You can pick any of the checkpoints listed [here](https://huggingface.co/models?filter=masked-lm) instead:<jupyter_code>model_checkpoint = "distilroberta-base"<jupyter_output><empty_output><jupyter_text>We can apply the same tokenization function as before, we just need to update our tokenizer to use the checkpoint we just picked. Don't panic about the warnings about inputs being too long for the model - remember that we'll be breaking them into shorter chunks right afterwards!<jupyter_code>tokenizer = AutoTokenizer.from_pretrained(model_checkpoint) tokenized_datasets = datasets.map( tokenize_function, batched=True, num_proc=4, remove_columns=["text"] )<jupyter_output><jupyter_text>And now, we group texts together and chunk them into samples of length `block_size`. You can skip this step if your dataset is composed of individual sentences.<jupyter_code>lm_datasets = tokenized_datasets.map( group_texts, batched=True, batch_size=1000, num_proc=4, )<jupyter_output><jupyter_text>The rest is very similar to what we had, with two exceptions. First we use a model suitable for masked LM:<jupyter_code>from transformers import TFAutoModelForMaskedLM model = TFAutoModelForMaskedLM.from_pretrained(model_checkpoint)<jupyter_output>All model checkpoint layers were used when initializing TFRobertaForMaskedLM. All the layers of TFRobertaForMaskedLM were initialized from the model checkpoint at distilroberta-base. If your task is similar to the task the model of the checkpoint was trained on, you can already use TFRobertaForMaskedLM for predictions without further training.<jupyter_text>We redefine our `optimizer` as we did with the CLM model, and we compile the model. We're using the internal loss and `jit_compile` again, like we did before.<jupyter_code>from transformers import create_optimizer, AdamWeightDecay import tensorflow as tf optimizer = AdamWeightDecay(lr=2e-5, weight_decay_rate=0.01) model.compile(optimizer=optimizer, jit_compile=True)<jupyter_output>/home/matt/miniconda3/envs/tensorflow28/lib/python3.10/site-packages/keras/optimizers/optimizer_v2/adam.py:110: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead. super(Adam, self).__init__(name, **kwargs) No loss specified in compile() - the model's internal loss computation will be used as the loss. Don't panic - this is a common way to train TensorFlow models in Transformers! To disable this behaviour please pass a loss argument, or explicitly pass `loss=None` if you do not want your model to compute a loss.<jupyter_text>Finally, we use a special `data_collator`. The `data_collator` is a function that is responsible for taking the samples and batching them in tensors. In the previous example, we had nothing special to do, so we just used the default for this argument. Here we want to randomly mask tokens. We could do it as a pre-processing step (like the tokenization) but then the tokens would always be masked the same way at each epoch. By doing this step inside the `data_collator`, we ensure this random masking is done in a new way each time we go over the data.To do this masking for us, the library provides a `DataCollatorForLanguageModeling`. We can adjust the probability of the masking. Note that our data collators are designed to work for multiple frameworks, so ensure you set the `return_tensors='np'` argument to get NumPy arrays out - you don't want to accidentally get a load of `torch.Tensor` objects in the middle of your nice TF code! You could also use `return_tensors='tf'` to get TensorFlow tensors, but our TF dataset pipeline actually uses a NumPy loader internally, which is wrapped at the end with a `tf.data.Dataset`. As a result, `np` is usually more reliable and performant when you're using it!<jupyter_code>from transformers import DataCollatorForLanguageModeling data_collator = DataCollatorForLanguageModeling( tokenizer=tokenizer, mlm_probability=0.15, return_tensors="np" )<jupyter_output><empty_output><jupyter_text>Now we generate our datasets as before. Remember to pass the `data_collator` you just made to the `collate_fn` argument.<jupyter_code>train_set = model.prepare_tf_dataset( lm_datasets["train"], shuffle=True, batch_size=16, collate_fn=data_collator, ) validation_set = model.prepare_tf_dataset( lm_datasets["validation"], shuffle=False, batch_size=16, collate_fn=data_collator, )<jupyter_output><empty_output><jupyter_text>And now we fit our model! As before, we can use a callback to sync with the hub during training. You can remove this if you don't want to!<jupyter_code>from transformers.keras_callbacks import PushToHubCallback model_name = model_checkpoint.split("/")[-1] push_to_hub_model_id = f"{model_name}-finetuned-wikitext2" callback = PushToHubCallback( output_dir="./mlm_model_save", tokenizer=tokenizer, hub_model_id=push_to_hub_model_id, ) model.fit(train_set, validation_data=validation_set, epochs=1, callbacks=[callback])<jupyter_output>/home/matt/PycharmProjects/notebooks/examples/mlm_model_save is already a clone of https://huggingface.co/Rocketknight1/distilroberta-base-finetuned-wikitext2. Make sure you pull the latest changes with `repo.git_pull()`.<jupyter_text>Like before, we can evaluate our model on the validation set and compute perplexity. The perplexity is much lower than for the CLM objective because for the MLM objective, we only have to make predictions for the masked tokens (which represent 15% of the total here) while having access to the rest of the tokens. It's thus an easier task for the model.<jupyter_code>import math eval_results = model.evaluate(validation_set) print(f"Perplexity: {math.exp(eval_results):.2f}")<jupyter_output>125/125 [==============================] - 3s 24ms/step - loss: 1.7174 Perplexity: 5.57<jupyter_text>If you used the callback, you can now share this model with all your friends, family or favorite pets: they can all load it with the identifier `"your-username/the-name-you-picked"` so for instance:```pythonfrom transformers import AutoModelForMaskedLMmodel = AutoModelForMaskedLM.from_pretrained("your-username/my-awesome-model")``` Inference Masked language models are not generally used directly for inference - the task they were trained on was to "fill in the blank", to identify missing words in sentences, and while this is an interesting demo, it has limited uses in production! However, masked language models work great as a base to be fine-tuned further on new tasks, like text or token classification. They are generally preferable to causal language models as a base for tasks that do not involve generating new text, and you'll see them being used as a base model in several other notebooks in this folder.Still, if you're curious, you can do inference to see what your model learned! Let's use the `fill-mask` pipeline and give the model some test sentences. Note that the mask token may not be "\" in some other models - you can use `tokenizer.mask_token` to find out what it is for your specific model if you're not using `distilroberta`.<jupyter_code>from transformers import pipeline # You can of course use your own model checkpoint here instead of mine mask_filler = pipeline( "fill-mask", "Rocketknight1/distilroberta-base-finetuned-wikitext2", framework="tf", ) mask_filler("The most common household pets are <mask> and dogs.", top_k=1)<jupyter_output><empty_output><jupyter_text>Nice! Let's see how it does on history:<jupyter_code>mask_filler("The Gulf War was a conflict that took place in <mask> in 1990-1991.", top_k=3)<jupyter_output><empty_output>

notebooks/examples/language_modeling-tf.ipynb/0

<jupyter_start><jupyter_text>If you're opening this Notebook on colab, you will probably need to install 🤗 Transformers as well as some other libraries. Uncomment the following cell and run it.<jupyter_code>#! pip install transformers evaluate datasets requests pandas sklearn<jupyter_output><empty_output><jupyter_text>If you're opening this notebook locally, make sure your environment has an install from the last version of those libraries.To be able to share your model with the community and generate results like the one shown in the picture below via the inference API, there are a few more steps to follow.First you have to store your authentication token from the Hugging Face website (sign up here if you haven't already!) then execute the following cell and input your username and password:<jupyter_code>from huggingface_hub import notebook_login notebook_login()<jupyter_output><empty_output><jupyter_text>Then you need to install Git-LFS. Uncomment the following instructions:<jupyter_code># !apt install git-lfs<jupyter_output><empty_output><jupyter_text>We also quickly upload some telemetry - this tells us which examples and software versions are getting used so we know where to prioritize our maintenance efforts. We don't collect (or care about) any personally identifiable information, but if you'd prefer not to be counted, feel free to skip this step or delete this cell entirely.<jupyter_code>from transformers.utils import send_example_telemetry send_example_telemetry("protein_language_modeling_notebook", framework="pytorch")<jupyter_output><empty_output><jupyter_text>Fine-Tuning Protein Language Models In this notebook, we're going to do some transfer learning to fine-tune some large, pre-trained protein language models on tasks of interest. If that sentence feels a bit intimidating to you, don't panic - there's [a blog post](https://huggingface.co/blog/deep-learning-with-proteins) that explains the concepts here in much more detail.The specific model we're going to use is ESM-2, which is the state-of-the-art protein language model at the time of writing (November 2022). The citation for this model is [Lin et al, 2022](https://www.biorxiv.org/content/10.1101/2022.07.20.500902v1).There are several ESM-2 checkpoints with differing model sizes. Larger models will generally have better accuracy, but they require more GPU memory and will take much longer to train. The available ESM-2 checkpoints (at time of writing) are:| Checkpoint name | Num layers | Num parameters ||------------------------------|----|----------|| `esm2_t48_15B_UR50D` | 48 | 15B || `esm2_t36_3B_UR50D` | 36 | 3B | | `esm2_t33_650M_UR50D` | 33 | 650M | | `esm2_t30_150M_UR50D` | 30 | 150M | | `esm2_t12_35M_UR50D` | 12 | 35M | | `esm2_t6_8M_UR50D` | 6 | 8M | Note that the larger checkpoints may be very difficult to train without a large cloud GPU like an A100 or H100, and the largest 15B parameter checkpoint will probably be impossible to train on **any** single GPU! Also, note that memory usage for attention during training will scale as `O(batch_size * num_layers * seq_len^2)`, so larger models on long sequences will use quite a lot of memory! We will use the `esm2_t12_35M_UR50D` checkpoint for this notebook, which should train on any Colab instance or modern GPU.<jupyter_code>model_checkpoint = "facebook/esm2_t12_35M_UR50D"<jupyter_output><empty_output><jupyter_text>Sequence classification One of the most common tasks you can perform with a language model is **sequence classification**. In sequence classification, we classify an entire protein into a category, from a list of two or more possibilities. There's no limit on the number of categories you can use, or the specific problem you choose, as long as it's something the model could in theory infer from the raw protein sequence. To keep things simple for this example, though, let's try classifying proteins by their cellular localization - given their sequence, can we predict if they're going to be found in the cytosol (the fluid inside the cell) or embedded in the cell membrane? Data preparation In this section, we're going to gather some training data from UniProt. Our goal is to create a pair of lists: `sequences` and `labels`. `sequences` will be a list of protein sequences, which will just be strings like "MNKL...", where each letter represents a single amino acid in the complete protein. `labels` will be a list of the category for each sequence. The categories will just be integers, with 0 representing the first category, 1 representing the second and so on. In other words, if `sequences[i]` is a protein sequence then `labels[i]` should be its corresponding category. These will form the **training data** we're going to use to teach the model the task we want it to do.If you're adapting this notebook for your own use, this will probably be the main section you want to change! You can do whatever you want here, as long as you create those two lists by the end of it. If you want to follow along with this example, though, first we'll need to `import requests` and set up our query to UniProt.<jupyter_code>import requests query_url ="https://rest.uniprot.org/uniprotkb/stream?compressed=true&fields=accession%2Csequence%2Ccc_subcellular_location&format=tsv&query=%28%28organism_id%3A9606%29%20AND%20%28reviewed%3Atrue%29%20AND%20%28length%3A%5B80%20TO%20500%5D%29%29"<jupyter_output><empty_output><jupyter_text>This query URL might seem mysterious, but it isn't! To get it, we searched for `(organism_id:9606) AND (reviewed:true) AND (length:[80 TO 500])` on UniProt to get a list of reasonably-sized human proteins,then selected 'Download', and set the format to TSV and the columns to `Sequence` and `Subcellular location [CC]`, since those contain the data we care about for this task.Once that's done, selecting `Generate URL for API` gives you a URL you can pass to Requests. Alternatively, if you're not on Colab you can just download the data through the web interface and open the file locally.<jupyter_code>uniprot_request = requests.get(query_url)<jupyter_output><empty_output><jupyter_text>To get this data into Pandas, we use a `BytesIO` object, which Pandas will treat like a file. If you downloaded the data as a file you can skip this bit and just pass the filepath directly to `read_csv`.<jupyter_code>from io import BytesIO import pandas bio = BytesIO(uniprot_request.content) df = pandas.read_csv(bio, compression='gzip', sep='\t') df<jupyter_output><empty_output><jupyter_text>Nice! Now we have some proteins and their subcellular locations. Let's start filtering this down. First, let's ditch the columns without subcellular location information.<jupyter_code>df = df.dropna() # Drop proteins with missing columns<jupyter_output><empty_output><jupyter_text>Now we'll make one dataframe of proteins that contain `cytosol` or `cytoplasm` in their subcellular localization column, and a second that mentions the `membrane` or `cell membrane`. To ensure we don't get overlap, we ensure each dataframe only contains proteins that don't match the other search term.<jupyter_code>cytosolic = df['Subcellular location [CC]'].str.contains("Cytosol") | df['Subcellular location [CC]'].str.contains("Cytoplasm") membrane = df['Subcellular location [CC]'].str.contains("Membrane") | df['Subcellular location [CC]'].str.contains("Cell membrane") cytosolic_df = df[cytosolic & ~membrane] cytosolic_df membrane_df = df[membrane & ~cytosolic] membrane_df<jupyter_output><empty_output><jupyter_text>We're almost done! Now, let's make a list of sequences from each df and generate the associated labels. We'll use `0` as the label for cytosolic proteins and `1` as the label for membrane proteins.<jupyter_code>cytosolic_sequences = cytosolic_df["Sequence"].tolist() cytosolic_labels = [0 for protein in cytosolic_sequences] membrane_sequences = membrane_df["Sequence"].tolist() membrane_labels = [1 for protein in membrane_sequences]<jupyter_output><empty_output><jupyter_text>Now we can concatenate these lists together to get the `sequences` and `labels` lists that will form our final training data. Don't worry - they'll get shuffled during training!<jupyter_code>sequences = cytosolic_sequences + membrane_sequences labels = cytosolic_labels + membrane_labels # Quick check to make sure we got it right len(sequences) == len(labels)<jupyter_output><empty_output><jupyter_text>Phew! Splitting the data Since the data we're loading isn't prepared for us as a machine learning dataset, we'll have to split the data into train and test sets ourselves! We can use sklearn's function for that:<jupyter_code>from sklearn.model_selection import train_test_split train_sequences, test_sequences, train_labels, test_labels = train_test_split(sequences, labels, test_size=0.25, shuffle=True)<jupyter_output><empty_output><jupyter_text>Tokenizing the data All inputs to neural nets must be numerical. The process of converting strings into numerical indices suitable for a neural net is called **tokenization**. For natural language this can be quite complex, as usually the network's vocabulary will not contain every possible word, which means the tokenizer must handle splitting rarer words into pieces, as well as all the complexities of capitalization and unicode characters and so on.With proteins, however, things are very easy. In protein language models, each amino acid is converted to a single token. Every model on `transformers` comes with an associated `tokenizer` that handles tokenization for it, and protein language models are no different. Let's get our tokenizer!<jupyter_code>from transformers import AutoTokenizer tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)<jupyter_output><empty_output><jupyter_text>Let's try a single sequence to see what the outputs from our tokenizer look like:<jupyter_code>tokenizer(train_sequences[0])<jupyter_output><empty_output><jupyter_text>This looks good! We can see that our sequence has been converted into `input_ids`, which is the tokenized sequence, and an `attention_mask`. The attention mask handles the case when we have sequences of variable length - in those cases, the shorter sequences are padded with blank "padding" tokens, and the attention mask is padded with 0s to indicate that those tokens should be ignored by the model.So now, let's tokenize our whole dataset. Note that we don't need to do anything with the labels, as they're already in the format we need.<jupyter_code>train_tokenized = tokenizer(train_sequences) test_tokenized = tokenizer(test_sequences)<jupyter_output><empty_output><jupyter_text>Dataset creation Now we want to turn this data into a dataset that PyTorch can load samples from. We can use the HuggingFace `Dataset` class for this, although if you prefer you can also use `torch.utils.data.Dataset`, at the cost of some more boilerplate code.<jupyter_code>from datasets import Dataset train_dataset = Dataset.from_dict(train_tokenized) test_dataset = Dataset.from_dict(test_tokenized) train_dataset<jupyter_output><empty_output><jupyter_text>This looks good, but we're missing our labels! Let's add those on as an extra column to the datasets.<jupyter_code>train_dataset = train_dataset.add_column("labels", train_labels) test_dataset = test_dataset.add_column("labels", test_labels) train_dataset<jupyter_output><empty_output><jupyter_text>Looks good! We're ready for training. Model loading Next, we want to load our model. Make sure to use exactly the same model as you used when loading the tokenizer, or your model might not understand the tokenization scheme you're using!<jupyter_code>from transformers import AutoModelForSequenceClassification, TrainingArguments, Trainer num_labels = max(train_labels + test_labels) + 1 # Add 1 since 0 can be a label model = AutoModelForSequenceClassification.from_pretrained(model_checkpoint, num_labels=num_labels)<jupyter_output><empty_output><jupyter_text>These warnings are telling us that the model is discarding some weights that it used for language modelling (the `lm_head`) and adding some weights for sequence classification (the `classifier`). This is exactly what we expect when we want to fine-tune a language model on a sequence classification task!Next, we initialize our `TrainingArguments`. These control the various training hyperparameters, and will be passed to our `Trainer`.<jupyter_code>model_name = model_checkpoint.split("/")[-1] batch_size = 8 args = TrainingArguments( f"{model_name}-finetuned-localization", evaluation_strategy = "epoch", save_strategy = "epoch", learning_rate=2e-5, per_device_train_batch_size=batch_size, per_device_eval_batch_size=batch_size, num_train_epochs=3, weight_decay=0.01, load_best_model_at_end=True, metric_for_best_model="accuracy", push_to_hub=True, )<jupyter_output><empty_output><jupyter_text>Next, we define the metric we will use to evaluate our models and write a `compute_metrics` function. We can load this from the `evaluate` library.<jupyter_code>from evaluate import load import numpy as np metric = load("accuracy") def compute_metrics(eval_pred): predictions, labels = eval_pred predictions = np.argmax(predictions, axis=1) return metric.compute(predictions=predictions, references=labels)<jupyter_output><empty_output><jupyter_text>And at last we're ready to initialize our `Trainer`:<jupyter_code>trainer = Trainer( model, args, train_dataset=train_dataset, eval_dataset=test_dataset, tokenizer=tokenizer, compute_metrics=compute_metrics, )<jupyter_output>/home/matt/PycharmProjects/notebooks/examples/esm2_t12_35M_UR50D-finetuned-localization is already a clone of https://huggingface.co/Rocketknight1/esm2_t12_35M_UR50D-finetuned-localization. Make sure you pull the latest changes with `repo.git_pull()`.<jupyter_text>You might wonder why we pass along the `tokenizer` when we already preprocessed our data. This is because we will use it one last time to make all the samples we gather the same length by applying padding, which requires knowing the model's preferences regarding padding (to the left or right? with which token?). The `tokenizer` has a pad method that will do all of this right for us, and the `Trainer` will use it. You can customize this part by defining and passing your own `data_collator` which will receive samples like the dictionaries seen above and will need to return a dictionary of tensors. We can now finetune our model by just calling the `train` method:<jupyter_code>trainer.train()<jupyter_output>/home/matt/PycharmProjects/transformers/src/transformers/optimization.py:306: FutureWarning: This implementation of AdamW is deprecated and will be removed in a future version. Use the PyTorch implementation torch.optim.AdamW instead, or set `no_deprecation_warning=True` to disable this warning warnings.warn( ***** Running training ***** Num examples = 3805 Num Epochs = 3 Instantaneous batch size per device = 8 Total train batch size (w. parallel, distributed & accumulation) = 8 Gradient Accumulation steps = 1 Total optimization steps = 1428 Number of trainable parameters = 33993602 Automatic Weights & Biases logging enabled, to disable set os.environ["WANDB_DISABLED"] = "true" wandb: You can find your API key in your browser here: https://wandb.ai/authorize<jupyter_text>Nice! After three epochs we have a model accuracy of ~94%. Note that we didn't do a lot of work to filter the training data or tune hyperparameters for this experiment, and also that we used one of the smallest ESM-2 models. With a larger starting model and more effort to ensure that the training data categories were cleanly separable, accuracy could almost certainly go a lot higher! *** Token classification Another common language model task is **token classification**. In this task, instead of classifying the whole sequence into a single category, we categorize each token (amino acid, in this case!) into one or more categories. This kind of model could be useful for:- Predicting secondary structure- Predicting buried vs. exposed residues- Predicting residues that will receive post-translational modifications- Predicting residues involved in binding pockets or active sites- Probably several other things, it's been a while since I was a postdoc Data preparation In this section, we're going to gather some training data from UniProt. As in the sequence classification example, we aim to create two lists: `sequences` and `labels`. Unlike in that example, however, the `labels` are more than just single integers. Instead, the label for each sample will be **one integer per token in the input**. This should make sense - when we do token classification, different tokens in the input may have different categories!To demonstrate token classification, we're going to go back to UniProt and get some data on protein secondary structures. As above, this will probably the main section you want to change when adapting this code to your own problems.<jupyter_code>import requests query_url ="https://rest.uniprot.org/uniprotkb/stream?compressed=true&fields=accession%2Csequence%2Cft_strand%2Cft_helix&format=tsv&query=%28%28organism_id%3A9606%29%20AND%20%28reviewed%3Atrue%29%20AND%20%28length%3A%5B80%20TO%20500%5D%29%29"<jupyter_output><empty_output><jupyter_text>This time, our UniProt search was `(organism_id:9606) AND (reviewed:true) AND (length:[100 TO 1000])` as it was in the first example, but instead of `Subcellular location [CC]` we take the `Helix` and `Beta strand` columns, as they contain the secondary structure information we want.<jupyter_code>uniprot_request = requests.get(query_url)<jupyter_output><empty_output><jupyter_text>To get this data into Pandas, we use a `BytesIO` object, which Pandas will treat like a file. If you downloaded the data as a file you can skip this bit and just pass the filepath directly to `read_csv`.<jupyter_code>from io import BytesIO import pandas bio = BytesIO(uniprot_request.content) df = pandas.read_csv(bio, compression='gzip', sep='\t') df<jupyter_output><empty_output><jupyter_text>Since not all proteins have this structural information, we discard proteins that have no annotated beta strands or alpha helices.<jupyter_code>no_structure_rows = df["Beta strand"].isna() & df["Helix"].isna() df = df[~no_structure_rows] df<jupyter_output><empty_output><jupyter_text>Well, this works, but that data still isn't in a clean format that we can use to build our labels. Let's take a look at one sample to see what exactly we're dealing with:<jupyter_code>df.iloc[0]["Helix"]<jupyter_output><empty_output><jupyter_text>We'll need to use a [regex](https://docs.python.org/3/howto/regex.html) to pull out each segment that's marked as being a STRAND or HELIX. What we're asking for is a list of everywhere we see the word STRAND or HELIX followed by two numbers separated by two dots. In each case where this pattern is found, we tell the regex to extract the two numbers as a tuple for us.<jupyter_code>import re strand_re = r"STRAND\s(\d+)\.\.(\d+)\;" helix_re = r"HELIX\s(\d+)\.\.(\d+)\;" re.findall(helix_re, df.iloc[0]["Helix"])<jupyter_output><empty_output><jupyter_text>Looks good! We can use this to build our training data. Recall that the **labels** need to be a list or array of integers that's the same length as the input sequence. We're going to use 0 to indicate residues without any annotated structure, 1 for residues in an alpha helix, and 2 for residues in a beta strand. To build that, we'll start with an array of all 0s, and then fill in values based on the positions that our regex pulls out of the UniProt results.We'll use NumPy arrays rather than lists here, since these allow [slice assignment](https://numpy.org/doc/stable/user/basics.indexing.htmlassigning-values-to-indexed-arrays), which will be a lot simpler than editing a list of integers. Note also that UniProt annotates residues starting from 1 (unlike Python, which starts from 0), and region annotations are inclusive (so 1..3 means residues 1, 2 and 3). To turn these into Python slices, we subtract 1 from the start of each annotation, but not the end.<jupyter_code>import numpy as np def build_labels(sequence, strands, helices): # Start with all 0s labels = np.zeros(len(sequence), dtype=np.int64) if isinstance(helices, float): # Indicates missing (NaN) found_helices = [] else: found_helices = re.findall(helix_re, helices) for helix_start, helix_end in found_helices: helix_start = int(helix_start) - 1 helix_end = int(helix_end) assert helix_end <= len(sequence) labels[helix_start: helix_end] = 1 # Helix category if isinstance(strands, float): # Indicates missing (NaN) found_strands = [] else: found_strands = re.findall(strand_re, strands) for strand_start, strand_end in found_strands: strand_start = int(strand_start) - 1 strand_end = int(strand_end) assert strand_end <= len(sequence) labels[strand_start: strand_end] = 2 # Strand category return labels<jupyter_output><empty_output><jupyter_text>Now we've defined a helper function, let's build our lists of sequences and labels:<jupyter_code>sequences = [] labels = [] for row_idx, row in df.iterrows(): row_labels = build_labels(row["Sequence"], row["Beta strand"], row["Helix"]) sequences.append(row["Sequence"]) labels.append(row_labels)<jupyter_output><empty_output><jupyter_text>Creating our dataset Nice! Now we'll split and tokenize the data, and then create datasets - I'll go through this quite quickly here, since it's identical to how we did it in the sequence classification example above.<jupyter_code>from sklearn.model_selection import train_test_split train_sequences, test_sequences, train_labels, test_labels = train_test_split(sequences, labels, test_size=0.25, shuffle=True) from transformers import AutoTokenizer tokenizer = AutoTokenizer.from_pretrained(model_checkpoint) train_tokenized = tokenizer(train_sequences) test_tokenized = tokenizer(test_sequences) from datasets import Dataset train_dataset = Dataset.from_dict(train_tokenized) test_dataset = Dataset.from_dict(test_tokenized) train_dataset = train_dataset.add_column("labels", train_labels) test_dataset = test_dataset.add_column("labels", test_labels)<jupyter_output><empty_output><jupyter_text>Model loading The key difference here with the above example is that we use `AutoModelForTokenClassification` instead of `AutoModelForSequenceClassification`. We will also need a `data_collator` this time, as we're in the slightly more complex case where both inputs and labels must be padded in each batch.<jupyter_code>from transformers import AutoModelForTokenClassification, TrainingArguments, Trainer num_labels = 3 model = AutoModelForTokenClassification.from_pretrained(model_checkpoint, num_labels=num_labels) from transformers import DataCollatorForTokenClassification data_collator = DataCollatorForTokenClassification(tokenizer)<jupyter_output><empty_output><jupyter_text>Now we set up our `TrainingArguments` as before.<jupyter_code>model_name = model_checkpoint.split("/")[-1] batch_size = 8 args = TrainingArguments( f"{model_name}-finetuned-secondary-structure", evaluation_strategy = "epoch", save_strategy = "epoch", learning_rate=1e-4, per_device_train_batch_size=batch_size, per_device_eval_batch_size=batch_size, num_train_epochs=3, weight_decay=0.001, load_best_model_at_end=True, metric_for_best_model="accuracy", push_to_hub=True, )<jupyter_output>PyTorch: setting up devices The default value for the training argument `--report_to` will change in v5 (from all installed integrations to none). In v5, you will need to use `--report_to all` to get the same behavior as now. You should start updating your code and make this info disappear :-).<jupyter_text>Our `compute_metrics` function is a bit more complex than in the sequence classification task, as we need to ignore padding tokens (those where the label is `-100`).<jupyter_code>from evaluate import load import numpy as np metric = load("accuracy") def compute_metrics(eval_pred): predictions, labels = eval_pred labels = labels.reshape((-1,)) predictions = np.argmax(predictions, axis=2) predictions = predictions.reshape((-1,)) predictions = predictions[labels!=-100] labels = labels[labels!=-100] return metric.compute(predictions=predictions, references=labels)<jupyter_output><empty_output><jupyter_text>And now we're ready to train our model!<jupyter_code>trainer = Trainer( model, args, train_dataset=train_dataset, eval_dataset=test_dataset, tokenizer=tokenizer, compute_metrics=compute_metrics, data_collator=data_collator, ) trainer.train()<jupyter_output>/home/matt/PycharmProjects/notebooks/examples/esm2_t12_35M_UR50D-finetuned-secondary-structure is already a clone of https://huggingface.co/Rocketknight1/esm2_t12_35M_UR50D-finetuned-secondary-structure. Make sure you pull the latest changes with `repo.git_pull()`. /home/matt/PycharmProjects/transformers/src/transformers/optimization.py:306: FutureWarning: This implementation of AdamW is deprecated and will be removed in a future version. Use the PyTorch implementation torch.optim.AdamW instead, or set `no_deprecation_warning=True` to disable this warning warnings.warn( ***** Running training ***** Num examples = 2933 Num Epochs = 3 Instantaneous batch size per device = 8 Total train batch size (w. parallel, distributed & accumulation) = 8 Gradient Accumulation steps = 1 Total optimization steps = 1101 Number of trainable parameters = 33763203 Automatic Weights & Biases logging enabled, to disable set os.environ["WANDB_DISABLED"] = "true"

notebooks/examples/protein_language_modeling.ipynb/0

<jupyter_start><jupyter_text>Quantizing a model with ONNX Runtime for text classification tasks This notebook shows how to apply different post-training quantization approaches such as static and dynamic quantization using [ONNX Runtime](https://onnxruntime.ai), for any tasks of the GLUE benchmark. This is made possible thanks to 🤗 [Optimum](https://github.com/huggingface/optimum), an extension of 🤗 [Transformers](https://github.com/huggingface/transformers), providing a set of performance optimization tools enabling maximum efficiency to train and run models on targeted hardwares. If you're opening this Notebook on colab, you will probably need to install 🤗 Transformers, 🤗 Datasets and 🤗 Optimum. Uncomment the following cell and run it.<jupyter_code>#! pip install datasets transformers[sklearn] optimum[onnxruntime]<jupyter_output><empty_output><jupyter_text>Make sure your version of 🤗 Optimum is at least 1.1.0 since the functionality was introduced in that version:<jupyter_code>from optimum.version import __version__ print(__version__)<jupyter_output>1.4.0.dev0<jupyter_text>The GLUE Benchmark is a group of nine classification tasks on sentences or pairs of sentences which are:- [CoLA](https://nyu-mll.github.io/CoLA/) (Corpus of Linguistic Acceptability) Determine if a sentence is grammatically correct or not.- [MNLI](https://arxiv.org/abs/1704.05426) (Multi-Genre Natural Language Inference) Determine if a sentence entails, contradicts or is unrelated to a given hypothesis. This dataset has two versions, one with the validation and test set coming from the same distribution, another called mismatched where the validation and test use out-of-domain data.- [MRPC](https://www.microsoft.com/en-us/download/details.aspx?id=52398) (Microsoft Research Paraphrase Corpus) Determine if two sentences are paraphrases from one another or not.- [QNLI](https://rajpurkar.github.io/SQuAD-explorer/) (Question-answering Natural Language Inference) Determine if the answer to a question is in the second sentence or not. This dataset is built from the SQuAD dataset.- [QQP](https://data.quora.com/First-Quora-Dataset-Release-Question-Pairs) (Quora Question Pairs2) Determine if two questions are semantically equivalent or not.- [RTE](https://aclweb.org/aclwiki/Recognizing_Textual_Entailment) (Recognizing Textual Entailment) Determine if a sentence entails a given hypothesis or not.- [SST-2](https://nlp.stanford.edu/sentiment/index.html) (Stanford Sentiment Treebank) Determine if the sentence has a positive or negative sentiment.- [STS-B](http://ixa2.si.ehu.es/stswiki/index.php/STSbenchmark) (Semantic Textual Similarity Benchmark) Determine the similarity of two sentences with a score from 1 to 5.- [WNLI](https://cs.nyu.edu/faculty/davise/papers/WinogradSchemas/WS.html) (Winograd Natural Language Inference) Determine if a sentence with an anonymous pronoun and a sentence with this pronoun replaced are entailed or not. This dataset is built from the Winograd Schema Challenge dataset.We will see how to apply post-training static quantization on a DistilBERT model fine-tuned on the SST-2 task:<jupyter_code>GLUE_TASKS = ["cola", "mnli", "mnli-mm", "mrpc", "qnli", "qqp", "rte", "sst2", "stsb", "wnli"] task = "sst2" model_checkpoint = "textattack/bert-base-uncased-SST-2"<jupyter_output><empty_output><jupyter_text>We also quickly upload some telemetry - this tells us which examples and software versions are getting used so we know where to prioritize our maintenance efforts. We don't collect (or care about) any personally identifiable information, but if you'd prefer not to be counted, feel free to skip this step or delete this cell entirely.<jupyter_code>from transformers.utils import send_example_telemetry send_example_telemetry("text_classification_notebook_ort", framework="none")<jupyter_output><empty_output><jupyter_text>Loading the dataset We will use the [🤗 Datasets](https://github.com/huggingface/datasets) library to download the dataset and get the metric we need to use for evaluation. This can be easily done with the functions `load_dataset` and `load_metric`.<jupyter_code>from datasets import load_dataset, load_metric<jupyter_output><empty_output><jupyter_text>`load_dataset` will cache the dataset to avoid downloading it again the next time you run this cell.<jupyter_code>actual_task = "mnli" if task == "mnli-mm" else task validation_split = "validation_mismatched" if task == "mnli-mm" else "validation_matched" if task == "mnli" else "validation" eval_dataset = load_dataset("glue", actual_task, split=validation_split) metric = load_metric("glue", actual_task)<jupyter_output><empty_output><jupyter_text>Note that `load_metric` has loaded the proper metric associated to your task, which is:- for CoLA: [Matthews Correlation Coefficient](https://en.wikipedia.org/wiki/Matthews_correlation_coefficient)- for MNLI (matched or mismatched): Accuracy- for MRPC: Accuracy and [F1 score](https://en.wikipedia.org/wiki/F1_score)- for QNLI: Accuracy- for QQP: Accuracy and [F1 score](https://en.wikipedia.org/wiki/F1_score)- for RTE: Accuracy- for SST-2: Accuracy- for STS-B: [Pearson Correlation Coefficient](https://en.wikipedia.org/wiki/Pearson_correlation_coefficient) and [Spearman's_Rank_Correlation_Coefficient](https://en.wikipedia.org/wiki/Spearman%27s_rank_correlation_coefficient)- for WNLI: Accuracyso the metric object only computes the one(s) needed for your task. Preprocessing the data To preprocess our dataset, we will need the names of the columns containing the sentence(s). The following dictionary keeps track of the correspondence task to column names:<jupyter_code>task_to_keys = { "cola": ("sentence", None), "mnli": ("premise", "hypothesis"), "mnli-mm": ("premise", "hypothesis"), "mrpc": ("sentence1", "sentence2"), "qnli": ("question", "sentence"), "qqp": ("question1", "question2"), "rte": ("sentence1", "sentence2"), "sst2": ("sentence", None), "stsb": ("sentence1", "sentence2"), "wnli": ("sentence1", "sentence2"), }<jupyter_output><empty_output><jupyter_text>We can then write the function that will preprocess our samples. We just feed them to the `tokenizer` with the argument `truncation=True`. This will ensure that an input longer than what the model selected can handle will be truncated to the maximum length accepted by the model.<jupyter_code>sentence1_key, sentence2_key = task_to_keys[task] def preprocess_function(examples, tokenizer): args = (examples[sentence1_key],) if sentence2_key is None else (examples[sentence1_key], examples[sentence2_key]) return tokenizer(*args, padding="max_length", max_length=128, truncation=True)<jupyter_output><empty_output><jupyter_text>Applying quantization on the model We can set our `quantization_approach` to either `dynamic` or `static` in order to apply respectively dynamic and static quantization. - Post-training static quantization : introduces an additional calibration step where data is fed through the network in order to compute the activations quantization parameters.- Post-training dynamic quantization : dynamically computes activations quantization parameters based on the data observed at runtime.<jupyter_code>QUANTIZATION_APPROACH = ["dynamic", "static"] quantization_approach = "static"<jupyter_output><empty_output><jupyter_text>First, let's create the output directory where the resulting quantized model will be saved.<jupyter_code>import os model_name = model_checkpoint.split("/")[-1] output_dir = f"{model_name}-{quantization_approach}-quantization" os.makedirs(output_dir, exist_ok=True)<jupyter_output><empty_output><jupyter_text>We will use the [🤗 Optimum](https://github.com/huggingface/optimum) library to instantiate an `ORTQuantizer`, which will take care of the quantization process. To instantiate an `ORTQuantizer`, we need to provide a path to a converted ONNX checkpoint or instance of a `ORTModelForXXX`.<jupyter_code>from optimum.onnxruntime import ORTQuantizer, ORTModelForSequenceClassification # Loading Model from the Hub and convert to ONNX ort_model = ORTModelForSequenceClassification.from_pretrained(model_checkpoint, from_transformers=True) # Create a quantizer from a ORTModelForXXX quantizer = ORTQuantizer.from_pretrained(ort_model)<jupyter_output><empty_output><jupyter_text>We also need to create an `QuantizationConfig` instance, which is the configuration handling the ONNX Runtime quantization related parameters.* We set `per_channel` to `False` in order to apply per-tensor quantization on the weights. As opposed to per-channel quantization, which introduces one set of quantization parameters per channel, per-tensor quantization means that there will be one set of quantization parameters per tensor.* We set the number of samples `num_calibration_samples` to use for the calibration step resulting from static quantization to `40`.* `operators_to_quantize` is used to specify the types of operators to quantize, here we want to quantize all the network's fully connected and embedding layers.<jupyter_code>from optimum.onnxruntime.configuration import QuantizationConfig, AutoCalibrationConfig from optimum.onnxruntime.quantization import QuantFormat, QuantizationMode, QuantType per_channel = False num_calibration_samples = 40 operators_to_quantize = ["MatMul", "Add", "Gather"] apply_static_quantization = quantization_approach == "static" qconfig = QuantizationConfig( is_static=apply_static_quantization, format=QuantFormat.QDQ if apply_static_quantization else QuantFormat.QOperator, mode=QuantizationMode.QLinearOps if apply_static_quantization else QuantizationMode.IntegerOps, activations_dtype=QuantType.QInt8 if apply_static_quantization else QuantType.QUInt8, weights_dtype=QuantType.QInt8, per_channel=per_channel, operators_to_quantize=operators_to_quantize, )<jupyter_output><empty_output><jupyter_text>When applying static quantization, we need to perform a calibration step where the activations quantization ranges are computed. This additionnal step should only be performed in the case of static quantization and not for dynamic quantization. Because the quantization of certain nodes often results in degradation in accuracy, we create an instance of `QuantizationPreprocessor` to determine the nodes to exclude when applying static quantization.<jupyter_code>from functools import partial from transformers import AutoTokenizer from optimum.onnxruntime.preprocessors import QuantizationPreprocessor from optimum.onnxruntime.preprocessors.passes import ( ExcludeGeLUNodes, ExcludeLayerNormNodes, ExcludeNodeAfter, ExcludeNodeFollowedBy, ) # Load tokenizer for preprocessing tokenizer = AutoTokenizer.from_pretrained(model_checkpoint) ranges = None quantization_preprocessor = None if apply_static_quantization: # Create the calibration dataset used for the calibration step calibration_dataset = quantizer.get_calibration_dataset( "glue", dataset_config_name=actual_task, preprocess_function=partial(preprocess_function, tokenizer=tokenizer), num_samples=num_calibration_samples, dataset_split="train", ) calibration_config = AutoCalibrationConfig.minmax(calibration_dataset) # Perform the calibration step: computes the activations quantization ranges ranges = quantizer.fit( dataset=calibration_dataset, calibration_config=calibration_config, operators_to_quantize=qconfig.operators_to_quantize, ) quantization_preprocessor = QuantizationPreprocessor() # Exclude the nodes constituting LayerNorm quantization_preprocessor.register_pass(ExcludeLayerNormNodes()) # Exclude the nodes constituting GELU quantization_preprocessor.register_pass(ExcludeGeLUNodes()) # Exclude the residual connection Add nodes quantization_preprocessor.register_pass(ExcludeNodeAfter("Add", "Add")) # Exclude the Add nodes following the Gather operator quantization_preprocessor.register_pass(ExcludeNodeAfter("Gather", "Add")) # Exclude the Add nodes followed by the Softmax operator quantization_preprocessor.register_pass(ExcludeNodeFollowedBy("Add", "Softmax"))<jupyter_output><empty_output><jupyter_text>Finally, we export the quantized model.<jupyter_code>quantizer.quantize( save_dir=output_dir, calibration_tensors_range=ranges, quantization_config=qconfig, preprocessor=quantization_preprocessor, )<jupyter_output><empty_output><jupyter_text>Evaluation To evaluate our resulting quantized model we need to define how to compute the metrics from the predictions. We need to define a function for this, which will just use the `metric` we loaded earlier, the only preprocessing we have to do is to take the argmax of our predicted logits (our just squeeze the last axis in the case of STS-B).The metric chosen to evaluate the quantized model's performance will be Matthews correlation coefficient (MCC) for CoLA, Pearson correlation coefficient (PCC) for STS-B and accuracy for any other tasks.<jupyter_code>import numpy as np def compute_metrics(eval_pred): predictions, labels = eval_pred if task != "stsb": predictions = np.argmax(predictions, axis=1) else: predictions = predictions[:, 0] return metric.compute(predictions=predictions, references=labels)<jupyter_output><empty_output><jupyter_text>Then to apply the preprocessing on all the sentences (or pairs of sentences) of our validation dataset, we just use the `map` method of our `dataset` object that was earlier created. This will apply the `preprocess_function` function on all the elements of our validation dataset.<jupyter_code>eval_dataset = eval_dataset.map(partial(preprocess_function, tokenizer=tokenizer), batched=True)<jupyter_output><empty_output><jupyter_text>Finally, to estimate the drop in performance resulting from quantization, we are going to perform an evaluation step for both models (before and after quantization). In order to perform the latter, we will need to instantiate an `ORTModel` and thus need:* The path of the model to evaluate.* The dataset to use for the evaluation step.* The model's ONNX configuration `onnx_config` associated to the model. This instance of `OnnxConfig` describes how to export the model through the ONNX format.* The function that will be used to compute the evaluation metrics `compute_metrics` that was defined previously.<jupyter_code>from optimum.onnxruntime import ORTModel from pathlib import Path ort_model = ORTModel(quantizer.onnx_model_path, compute_metrics=compute_metrics, label_names=["label"]) model_output = ort_model.evaluation_loop(eval_dataset) model_output.metrics q8_ort_model = ORTModel(Path(output_dir) / "model_quantized.onnx", compute_metrics=compute_metrics, label_names=["label"]) q8_model_output = q8_ort_model.evaluation_loop(eval_dataset) q8_model_output.metrics<jupyter_output><empty_output><jupyter_text>Now let's compute the full-precision and the quantized model respective size in megabyte (MB) :<jupyter_code>fp_model_size = os.path.getsize(quantizer.onnx_model_path) / (1024*1024) q_model_size = os.path.getsize(Path(output_dir) / "model_quantized.onnx") / (1024*1024)<jupyter_output><empty_output><jupyter_text>The reduction in the model size resulting from quantization is:<jupyter_code>round(fp_model_size / q_model_size, 2)<jupyter_output><empty_output>

notebooks/examples/text_classification_quantization_ort.ipynb/0

from transformers import AutoModelForSequenceClassification, Trainer, TrainingArguments, AutoTokenizer from sklearn.metrics import accuracy_score, precision_recall_fscore_support from datasets import load_from_disk import random import logging import sys import argparse import os import torch if __name__ == "__main__": parser = argparse.ArgumentParser() # hyperparameters sent by the client are passed as command-line arguments to the script. parser.add_argument("--epochs", type=int, default=3) parser.add_argument("--train_batch_size", type=int, default=32) parser.add_argument("--eval_batch_size", type=int, default=64) parser.add_argument("--warmup_steps", type=int, default=500) parser.add_argument("--model_name", type=str) parser.add_argument("--learning_rate", type=str, default=5e-5) # Data, model, and output directories parser.add_argument("--output_data_dir", type=str, default=os.environ["SM_OUTPUT_DATA_DIR"]) parser.add_argument("--model_dir", type=str, default=os.environ["SM_MODEL_DIR"]) parser.add_argument("--n_gpus", type=str, default=os.environ["SM_NUM_GPUS"]) parser.add_argument("--training_dir", type=str, default=os.environ["SM_CHANNEL_TRAIN"]) parser.add_argument("--test_dir", type=str, default=os.environ["SM_CHANNEL_TEST"]) args, _ = parser.parse_known_args() # Set up logging logger = logging.getLogger(__name__) logging.basicConfig( level=logging.getLevelName("INFO"), handlers=[logging.StreamHandler(sys.stdout)], format="%(asctime)s - %(name)s - %(levelname)s - %(message)s", ) # load datasets train_dataset = load_from_disk(args.training_dir) test_dataset = load_from_disk(args.test_dir) logger.info(f" loaded train_dataset length is: {len(train_dataset)}") logger.info(f" loaded test_dataset length is: {len(test_dataset)}") # compute metrics function for binary classification def compute_metrics(pred): labels = pred.label_ids preds = pred.predictions.argmax(-1) precision, recall, f1, _ = precision_recall_fscore_support(labels, preds, average="binary") acc = accuracy_score(labels, preds) return {"accuracy": acc, "f1": f1, "precision": precision, "recall": recall} # download model from model hub model = AutoModelForSequenceClassification.from_pretrained(args.model_name) tokenizer = AutoTokenizer.from_pretrained(args.model_name) # define training args training_args = TrainingArguments( output_dir=args.model_dir, num_train_epochs=args.epochs, per_device_train_batch_size=args.train_batch_size, per_device_eval_batch_size=args.eval_batch_size, warmup_steps=args.warmup_steps, evaluation_strategy="epoch", logging_dir=f"{args.output_data_dir}/logs", learning_rate=float(args.learning_rate), ) # create Trainer instance trainer = Trainer( model=model, args=training_args, compute_metrics=compute_metrics, train_dataset=train_dataset, eval_dataset=test_dataset, tokenizer=tokenizer, ) # train model trainer.train() # evaluate model eval_result = trainer.evaluate(eval_dataset=test_dataset) # writes eval result to file which can be accessed later in s3 ouput with open(os.path.join(args.output_data_dir, "eval_results.txt"), "w") as writer: print(f"***** Eval results *****") for key, value in sorted(eval_result.items()): writer.write(f"{key} = {value}\n") # Saves the model to s3 trainer.save_model(args.model_dir)

notebooks/sagemaker/01_getting_started_pytorch/scripts/train.py/0

<jupyter_start><jupyter_text>Huggingface Sagemaker-sdk - Deploy 🤗 Transformers for inference Welcome to this getting started guide, we will use the new Hugging Face Inference DLCs and Amazon SageMaker Python SDK to deploy a transformer model for inference. In this example we directly deploy one of the 10 000+ Hugging Face Transformers from the [Hub](https://huggingface.co/models) to Amazon SageMaker for Inference. API - [SageMaker Hugging Face Inference Toolkit](https://github.com/aws/sagemaker-huggingface-inference-toolkit) Using the `transformers pipelines`, we designed an API, which makes it easy for you to benefit from all `pipelines` features. The API is oriented at the API of the [🤗 Accelerated Inference API](https://api-inference.huggingface.co/docs/python/html/detailed_parameters.html), meaning your inputs need to be defined in the `inputs` key and if you want additional supported `pipelines` parameters you can add them in the `parameters` key. Below you can find examples for requests. **text-classification request body**```python{ "inputs": "Camera - You are awarded a SiPix Digital Camera! call 09061221066 fromm landline. Delivery within 28 days."}```**question-answering request body**```python{ "inputs": { "question": "What is used for inference?", "context": "My Name is Philipp and I live in Nuremberg. This model is used with sagemaker for inference." }}```**zero-shot classification request body**```python{ "inputs": "Hi, I recently bought a device from your company but it is not working as advertised and I would like to get reimbursed!", "parameters": { "candidate_labels": [ "refund", "legal", "faq" ] }}```<jupyter_code>!pip install "sagemaker>=2.48.0" --upgrade<jupyter_output><empty_output><jupyter_text>Deploy one of the 10 000+ Hugging Face Transformers to Amazon SageMaker for Inference_This is an experimental feature, where the model will be loaded after the endpoint is created. This could lead to errors, e.g. models > 10GB_To deploy a model directly from the Hub to SageMaker we need to define 2 environment variables when creating the `HuggingFaceModel` . We need to define:- `HF_MODEL_ID`: defines the model id, which will be automatically loaded from [huggingface.co/models](http://huggingface.co/models) when creating or SageMaker Endpoint. The 🤗 Hub provides +10 000 models all available through this environment variable.- `HF_TASK`: defines the task for the used 🤗 Transformers pipeline. A full list of tasks can be find [here](https://huggingface.co/transformers/main_classes/pipelines.html).<jupyter_code>import sagemaker import boto3 try: role = sagemaker.get_execution_role() except ValueError: iam = boto3.client('iam') role = iam.get_role(RoleName='sagemaker_execution_role')['Role']['Arn'] print(f"sagemaker role arn: {role}") from sagemaker.huggingface import HuggingFaceModel # Hub Model configuration. https://huggingface.co/models hub = { 'HF_MODEL_ID':'distilbert-base-uncased-distilled-squad', # model_id from hf.co/models 'HF_TASK':'question-answering' # NLP task you want to use for predictions } # create Hugging Face Model Class huggingface_model = HuggingFaceModel( env=hub, role=role, # iam role with permissions to create an Endpoint transformers_version="4.26", # transformers version used pytorch_version="1.13", # pytorch version used py_version="py39", # python version of the DLC ) # deploy model to SageMaker Inference predictor = huggingface_model.deploy( initial_instance_count=1, instance_type="ml.m5.xlarge" ) # example request, you always need to define "inputs" data = { "inputs": { "question": "What is used for inference?", "context": "My Name is Philipp and I live in Nuremberg. This model is used with sagemaker for inference." } } # request predictor.predict(data) # delete endpoint predictor.delete_model() predictor.delete_endpoint()<jupyter_output><empty_output>

notebooks/sagemaker/11_deploy_model_from_hf_hub/deploy_transformer_model_from_hf_hub.ipynb/0

<jupyter_start><jupyter_text>Semantic Segmantion with Hugging Face's Transformers & Amazon SageMaker Transformer models are changing are changing the world of machine learning, starting with natural language processing, and now, with audio and computer vision. Hugging Face's mission is to democratize good machine learning and giving any one the opportunity to use these new state-of-the-art machine learning models. Together with Amazon SageMaker and AWS we have been working on extending the functionalities of the Hugging Face Inference DLC and the Python SageMaker SDK to make it easier to use speech and vision models together with `transformers`. You can now use the Hugging Face Inference DLC to do [automatic speech recognition](https://huggingface.co/tasks/automatic-speech-recognition) using MetaAIs [wav2vec2](https://arxiv.org/abs/2006.11477) model or Microsofts [WavLM](https://arxiv.org/abs/2110.13900) or use NVIDIAs [SegFormer](https://arxiv.org/abs/2105.15203) for [image segmentation](https://huggingface.co/tasks/image-segmentation).This guide will walk you through how to do [Image Segmentation](https://huggingface.co/tasks/image-segmentation) using [segformer](https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512) and new `DataSerializer` In this example you will learn how to: 1. Setup a development Environment and permissions for deploying Amazon SageMaker Inference Endpoints.2. Deploy a segformer model to Amazon SageMaker for image segmentation3. Send requests to the endpoint to do image segmentation. Let's get started! 🚀---*If you are going to use Sagemaker in a local environment (not SageMaker Studio or Notebook Instances). You need access to an IAM Role with the required permissions for Sagemaker. You can find [here](https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-roles.html) more about it.* 1. Setup a development Environment and permissions for deploying Amazon SageMaker Inference Endpoints.Setting up the development environment and permissions needs to be done for the automatic-speech-recognition example and the semantic-segmentation example. First we update the `sagemaker` SDK to make sure we have new `DataSerializer`.<jupyter_code>%pip install sagemaker segmentation-mask-overlay pillow matplotlib --upgrade<jupyter_output><empty_output><jupyter_text>After we have update the SDK we can set the permissions._If you are going to use Sagemaker in a local environment (not SageMaker Studio or Notebook Instances). You need access to an IAM Role with the required permissions for Sagemaker. You can find [here](https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-roles.html) more about it._<jupyter_code>import sagemaker import boto3 sess = sagemaker.Session() # sagemaker session bucket -> used for uploading data, models and logs # sagemaker will automatically create this bucket if it not exists sagemaker_session_bucket=None if sagemaker_session_bucket is None and sess is not None: # set to default bucket if a bucket name is not given sagemaker_session_bucket = sess.default_bucket() try: role = sagemaker.get_execution_role() except ValueError: iam = boto3.client('iam') role = iam.get_role(RoleName='sagemaker_execution_role')['Role']['Arn'] sess = sagemaker.Session(default_bucket=sagemaker_session_bucket) print(f"sagemaker role arn: {role}") print(f"sagemaker bucket: {sess.default_bucket()}") print(f"sagemaker session region: {sess.boto_region_name}")<jupyter_output>sagemaker role arn: arn:aws:iam::558105141721:role/sagemaker_execution_role sagemaker bucket: sagemaker-us-east-1-558105141721 sagemaker session region: us-east-1<jupyter_text>2. Deploy a segformer model to Amazon SageMaker for image segmentationImage Segmentation divides an image into segments where each pixel in the image is mapped to an object. This task has multiple variants such as instance segmentation, panoptic segmentation and semantic segmentation.We use the [nvidia/segformer-b0-finetuned-ade-512-512](https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512) model running our segmentation endpoint. This model is fine-tuned on ADE20k (scene-centric image) at resolution 512x512.<jupyter_code>from sagemaker.huggingface.model import HuggingFaceModel from sagemaker.serializers import DataSerializer # Hub Model configuration. <https://huggingface.co/models> hub = { 'HF_MODEL_ID':'nvidia/segformer-b0-finetuned-ade-512-512', 'HF_TASK':'image-segmentation', } # create Hugging Face Model Class huggingface_model = HuggingFaceModel( env=hub, # configuration for loading model from Hub role=role, # iam role with permissions to create an Endpoint transformers_version="4.26", # transformers version used pytorch_version="1.13", # pytorch version used py_version='py39', # python version used )<jupyter_output><empty_output><jupyter_text>Before we are able to deploy our `HuggingFaceModel` class we need to create a new serializer, which supports our audio data. The Serializer are used in Predictor and in the `predict` method to serializer our data to a specific `mime-type`, which send to the endpoint. The default serialzier for the HuggingFacePredcitor is a JSNON serializer, but since we are not going to send text data to the endpoint we will use the DataSerializer.<jupyter_code># create a serializer for the data image_serializer = DataSerializer(content_type='image/x-image') # using x-image to support multiple image formats # deploy model to SageMaker Inference predictor = huggingface_model.deploy( initial_instance_count=1, # number of instances instance_type='ml.g4dn.xlarge', # ec2 instance type serializer=image_serializer, # serializer for our audio data. )<jupyter_output>-----------!<jupyter_text>3. Send requests to the endpoint to do image segmentation.The `.deploy()` returns an `HuggingFacePredictor` object with our `DataSeriliazer` which can be used to request inference. This `HuggingFacePredictor` makes it easy to send requests to your endpoint and get the results back.We will use 2 different methods to send requests to the endpoint:a. Provide a image file via path to the predictor b. Provide binary image data object to the predictor a. Provide a image file via path to the predictorUsing a image file as input is easy as easy as providing the path to its location. The `DataSerializer` will then read it and send the bytes to the endpoint. We can use a `libirispeech` sample hosted on huggingface.co<jupyter_code>!wget https://huggingface.co/datasets/hf-internal-testing/fixtures_ade20k/raw/main/ADE_val_00000001.jpg<jupyter_output>--2023-03-21 08:29:41-- https://huggingface.co/datasets/hf-internal-testing/fixtures_ade20k/raw/main/ADE_val_00000001.jpg Resolving huggingface.co (huggingface.co)... 52.203.75.138, 3.216.111.67, 3.83.196.160, ... Connecting to huggingface.co (huggingface.co)|52.203.75.138|:443... connected. HTTP request sent, awaiting response... 200 OK Length: 52650 (51K) [image/jpeg] Saving to: ‘ADE_val_00000001.jpg’ 100%[======================================>] 52,650 --.-K/s in 0.001s 2023-03-21 08:29:41 (43.8 MB/s) - ‘ADE_val_00000001.jpg’ saved [52650/52650]<jupyter_text>before we send our request lest create a helper function to display our segmentation results.<jupyter_code>from PIL import Image import io from segmentation_mask_overlay import overlay_masks import numpy as np import base64 import matplotlib.pyplot as plt def stringToRGB(base64_string): # convert base64 string to numpy array imgdata = base64.b64decode(str(base64_string)) image = Image.open(io.BytesIO(imgdata)) return np.array(image) def get_overlay(original_image_path,result): masks = [stringToRGB(r["mask"]).astype('bool') for r in res] masks_labels = [r["label"] for r in result] cmap = plt.cm.tab20(np.arange(len(masks_labels))) image = Image.open(original_image_path) overlay_masks(image, masks, labels=masks_labels, colors=cmap, mask_alpha=0.5)<jupyter_output><empty_output><jupyter_text>To send a request with provide our path to the audio file we can use the following code:<jupyter_code>image_path = "ADE_val_00000001.jpg" res = predictor.predict(data=image_path) print(res[0].keys()) get_overlay(image_path,res)<jupyter_output>dict_keys(['score', 'label', 'mask'])<jupyter_text>b. Provide binary image data object to the predictorInstead of providing a path to the image file we can also directy provide the bytes of it reading the file in python._make sure `ADE_val_00000001.jpg` is in the directory_<jupyter_code>image_path = "ADE_val_00000001.jpg" with open(image_path, "rb") as data_file: image_data = data_file.read() res = predictor.predict(data=image_data) print(res[0].keys()) get_overlay(image_path,res)<jupyter_output>dict_keys(['score', 'label', 'mask'])<jupyter_text>Clean up<jupyter_code>predictor.delete_model() predictor.delete_endpoint()<jupyter_output><empty_output>

notebooks/sagemaker/21_image_segmantation/sagemaker-notebook.ipynb/0

<!--- Copyright 2023 The HuggingFace Team. All rights reserved. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --> <h1 align="center"> <p>🤗 PEFT</p></h1> <h3 align="center"> <p>State-of-the-art Parameter-Efficient Fine-Tuning (PEFT) methods</p> </h3> Parameter-Efficient Fine-Tuning (PEFT) methods enable efficient adaptation of pre-trained language models (PLMs) to various downstream applications without fine-tuning all the model's parameters. Fine-tuning large-scale PLMs is often prohibitively costly. In this regard, PEFT methods only fine-tune a small number of (extra) model parameters, thereby greatly decreasing the computational and storage costs. Recent State-of-the-Art PEFT techniques achieve performance comparable to that of full fine-tuning. Seamlessly integrated with 🤗 Accelerate for large scale models leveraging DeepSpeed and Big Model Inference. Supported methods: 1. LoRA: [LORA: LOW-RANK ADAPTATION OF LARGE LANGUAGE MODELS](https://arxiv.org/abs/2106.09685) 2. Prefix Tuning: [Prefix-Tuning: Optimizing Continuous Prompts for Generation](https://aclanthology.org/2021.acl-long.353/), [P-Tuning v2: Prompt Tuning Can Be Comparable to Fine-tuning Universally Across Scales and Tasks](https://arxiv.org/pdf/2110.07602.pdf) 3. P-Tuning: [GPT Understands, Too](https://arxiv.org/abs/2103.10385) 4. Prompt Tuning: [The Power of Scale for Parameter-Efficient Prompt Tuning](https://arxiv.org/abs/2104.08691) 5. AdaLoRA: [Adaptive Budget Allocation for Parameter-Efficient Fine-Tuning](https://arxiv.org/abs/2303.10512) 6. $(IA)^3$: [Few-Shot Parameter-Efficient Fine-Tuning is Better and Cheaper than In-Context Learning](https://arxiv.org/abs/2205.05638) 7. MultiTask Prompt Tuning: [Multitask Prompt Tuning Enables Parameter-Efficient Transfer Learning](https://arxiv.org/abs/2303.02861) 8. LoHa: [FedPara: Low-Rank Hadamard Product for Communication-Efficient Federated Learning](https://arxiv.org/abs/2108.06098) 9. LoKr: [KronA: Parameter Efficient Tuning with Kronecker Adapter](https://arxiv.org/abs/2212.10650) based on [Navigating Text-To-Image Customization:From LyCORIS Fine-Tuning to Model Evaluation](https://arxiv.org/abs/2309.14859) implementation 10. LoftQ: [LoftQ: LoRA-Fine-Tuning-aware Quantization for Large Language Models](https://arxiv.org/abs/2310.08659) 11. OFT: [Controlling Text-to-Image Diffusion by Orthogonal Finetuning](https://arxiv.org/abs/2306.07280) ## Getting started ```python from transformers import AutoModelForSeq2SeqLM from peft import get_peft_config, get_peft_model, LoraConfig, TaskType model_name_or_path = "bigscience/mt0-large" tokenizer_name_or_path = "bigscience/mt0-large" peft_config = LoraConfig( task_type=TaskType.SEQ_2_SEQ_LM, inference_mode=False, r=8, lora_alpha=32, lora_dropout=0.1 ) model = AutoModelForSeq2SeqLM.from_pretrained(model_name_or_path) model = get_peft_model(model, peft_config) model.print_trainable_parameters() # output: trainable params: 2359296 || all params: 1231940608 || trainable%: 0.19151053100118282 ``` ## Use Cases ### Get comparable performance to full finetuning by adapting LLMs to downstream tasks using consumer hardware GPU memory required for adapting LLMs on the few-shot dataset [`ought/raft/twitter_complaints`](https://huggingface.co/datasets/ought/raft/viewer/twitter_complaints). Here, settings considered are full finetuning, PEFT-LoRA using plain PyTorch and PEFT-LoRA using DeepSpeed with CPU Offloading. Hardware: Single A100 80GB GPU with CPU RAM above 64GB | Model | Full Finetuning | PEFT-LoRA PyTorch | PEFT-LoRA DeepSpeed with CPU Offloading | | --------- | ---- | ---- | ---- | | bigscience/T0_3B (3B params) | 47.14GB GPU / 2.96GB CPU | 14.4GB GPU / 2.96GB CPU | 9.8GB GPU / 17.8GB CPU | | bigscience/mt0-xxl (12B params) | OOM GPU | 56GB GPU / 3GB CPU | 22GB GPU / 52GB CPU | | bigscience/bloomz-7b1 (7B params) | OOM GPU | 32GB GPU / 3.8GB CPU | 18.1GB GPU / 35GB CPU | Performance of PEFT-LoRA tuned [`bigscience/T0_3B`](https://huggingface.co/bigscience/T0_3B) on [`ought/raft/twitter_complaints`](https://huggingface.co/datasets/ought/raft/viewer/twitter_complaints) leaderboard. A point to note is that we didn't try to squeeze performance by playing around with input instruction templates, LoRA hyperparams and other training related hyperparams. Also, we didn't use the larger 13B [mt0-xxl](https://huggingface.co/bigscience/mt0-xxl) model. So, we are already seeing comparable performance to SoTA with parameter efficient tuning. Also, the final additional checkpoint size is just `19MB` in comparison to `11GB` size of the backbone [`bigscience/T0_3B`](https://huggingface.co/bigscience/T0_3B) model, but one still has to load the original full size model. | Submission Name | Accuracy | | --------- | ---- | | Human baseline (crowdsourced) | 0.897 | | Flan-T5 | 0.892 | | lora-t0-3b | 0.863 | **Therefore, we can see that performance comparable to SoTA is achievable by PEFT methods with consumer hardware such as 16GB and 24GB GPUs.** An insightful blogpost explaining the advantages of using PEFT for fine-tuning FlanT5-XXL: [https://www.philschmid.de/fine-tune-flan-t5-peft](https://www.philschmid.de/fine-tune-flan-t5-peft) ### Parameter Efficient Tuning of Diffusion Models GPU memory required by different settings during training is given below. The final checkpoint size is `8.8 MB`. Hardware: Single A100 80GB GPU with CPU RAM above 64GB | Model | Full Finetuning | PEFT-LoRA | PEFT-LoRA with Gradient Checkpointing | | --------- | ---- | ---- | ---- | | CompVis/stable-diffusion-v1-4 | 27.5GB GPU / 3.97GB CPU | 15.5GB GPU / 3.84GB CPU | 8.12GB GPU / 3.77GB CPU | **Training** An example of using LoRA for parameter efficient dreambooth training is given in [`examples/lora_dreambooth/train_dreambooth.py`](examples/lora_dreambooth/train_dreambooth.py) ```bash export MODEL_NAME= "CompVis/stable-diffusion-v1-4" #"stabilityai/stable-diffusion-2-1" export INSTANCE_DIR="path-to-instance-images" export CLASS_DIR="path-to-class-images" export OUTPUT_DIR="path-to-save-model" accelerate launch train_dreambooth.py \ --pretrained_model_name_or_path=$MODEL_NAME \ --instance_data_dir=$INSTANCE_DIR \ --class_data_dir=$CLASS_DIR \ --output_dir=$OUTPUT_DIR \ --train_text_encoder \ --with_prior_preservation --prior_loss_weight=1.0 \ --instance_prompt="a photo of sks dog" \ --class_prompt="a photo of dog" \ --resolution=512 \ --train_batch_size=1 \ --lr_scheduler="constant" \ --lr_warmup_steps=0 \ --num_class_images=200 \ --use_lora \ --lora_r 16 \ --lora_alpha 27 \ --lora_text_encoder_r 16 \ --lora_text_encoder_alpha 17 \ --learning_rate=1e-4 \ --gradient_accumulation_steps=1 \ --gradient_checkpointing \ --max_train_steps=800 ``` Try out the 🤗 Gradio Space which should run seamlessly on a T4 instance: [smangrul/peft-lora-sd-dreambooth](https://huggingface.co/spaces/smangrul/peft-lora-sd-dreambooth).  **NEW** ✨ Multi Adapter support and combining multiple LoRA adapters in a weighted combination  **NEW** ✨ Dreambooth training for Stable Diffusion using LoHa and LoKr adapters [`examples/stable_diffusion/train_dreambooth.py`](examples/stable_diffusion/train_dreambooth.py) ### Parameter Efficient Tuning of LLMs for RLHF components such as Ranker and Policy - Here is an example in [trl](https://github.com/lvwerra/trl) library using PEFT+INT8 for tuning policy model: [gpt2-sentiment_peft.py](https://github.com/lvwerra/trl/blob/main/examples/sentiment/scripts/gpt2-sentiment_peft.py) and corresponding [Blog](https://huggingface.co/blog/trl-peft) - Example using PEFT for Instruction finetuning, reward model and policy : [stack_llama](https://github.com/lvwerra/trl/tree/main/examples/research_projects/stack_llama/scripts) and corresponding [Blog](https://huggingface.co/blog/stackllama) ### INT8 training of large models in Colab using PEFT LoRA and bitsandbytes - Here is now a demo on how to fine tune [OPT-6.7b](https://huggingface.co/facebook/opt-6.7b) (14GB in fp16) in a Google Colab: [](https://colab.research.google.com/drive/1jCkpikz0J2o20FBQmYmAGdiKmJGOMo-o?usp=sharing) - Here is now a demo on how to fine tune [whisper-large](https://huggingface.co/openai/whisper-large-v2) (1.5B params) (14GB in fp16) in a Google Colab: [](https://colab.research.google.com/drive/1DOkD_5OUjFa0r5Ik3SgywJLJtEo2qLxO?usp=sharing) and [](https://colab.research.google.com/drive/1vhF8yueFqha3Y3CpTHN6q9EVcII9EYzs?usp=sharing) ### Save compute and storage even for medium and small models Save storage by avoiding full finetuning of models on each of the downstream tasks/datasets, With PEFT methods, users only need to store tiny checkpoints in the order of `MBs` all the while retaining performance comparable to full finetuning. An example of using LoRA for the task of adapting `LayoutLMForTokenClassification` on `FUNSD` dataset is given in `~examples/token_classification/PEFT_LoRA_LayoutLMForTokenClassification_on_FUNSD.py`. We can observe that with only `0.62 %` of parameters being trainable, we achieve performance (F1 0.777) comparable to full finetuning (F1 0.786) (without any hyperparam tuning runs for extracting more performance), and the checkpoint of this is only `2.8MB`. Now, if there are `N` such datasets, just have these PEFT models one for each dataset and save a lot of storage without having to worry about the problem of catastrophic forgetting or overfitting of backbone/base model. Another example is fine-tuning [`roberta-large`](https://huggingface.co/roberta-large) on [`MRPC` GLUE](https://huggingface.co/datasets/glue/viewer/mrpc) dataset using different PEFT methods. The notebooks are given in `~examples/sequence_classification`. ## PEFT + 🤗 Accelerate PEFT models work with 🤗 Accelerate out of the box. Use 🤗 Accelerate for Distributed training on various hardware such as GPUs, Apple Silicon devices, etc during training. Use 🤗 Accelerate for inferencing on consumer hardware with small resources. ### Example of PEFT model training using 🤗 Accelerate's DeepSpeed integration DeepSpeed version required `v0.8.0`. An example is provided in `~examples/conditional_generation/peft_lora_seq2seq_accelerate_ds_zero3_offload.py`. a. First, run `accelerate config --config_file ds_zero3_cpu.yaml` and answer the questionnaire. Below are the contents of the config file. ```yaml compute_environment: LOCAL_MACHINE deepspeed_config: gradient_accumulation_steps: 1 gradient_clipping: 1.0 offload_optimizer_device: cpu offload_param_device: cpu zero3_init_flag: true zero3_save_16bit_model: true zero_stage: 3 distributed_type: DEEPSPEED downcast_bf16: 'no' dynamo_backend: 'NO' fsdp_config: {} machine_rank: 0 main_training_function: main megatron_lm_config: {} mixed_precision: 'no' num_machines: 1 num_processes: 1 rdzv_backend: static same_network: true use_cpu: false ``` b. run the below command to launch the example script ```bash accelerate launch --config_file ds_zero3_cpu.yaml examples/peft_lora_seq2seq_accelerate_ds_zero3_offload.py ``` c. output logs: ```bash GPU Memory before entering the train : 1916 GPU Memory consumed at the end of the train (end-begin): 66 GPU Peak Memory consumed during the train (max-begin): 7488 GPU Total Peak Memory consumed during the train (max): 9404 CPU Memory before entering the train : 19411 CPU Memory consumed at the end of the train (end-begin): 0 CPU Peak Memory consumed during the train (max-begin): 0 CPU Total Peak Memory consumed during the train (max): 19411 epoch=4: train_ppl=tensor(1.0705, device='cuda:0') train_epoch_loss=tensor(0.0681, device='cuda:0') 100%|████████████████████████████████████████████████████████████████████████████████████████████| 7/7 [00:27<00:00, 3.92s/it] GPU Memory before entering the eval : 1982 GPU Memory consumed at the end of the eval (end-begin): -66 GPU Peak Memory consumed during the eval (max-begin): 672 GPU Total Peak Memory consumed during the eval (max): 2654 CPU Memory before entering the eval : 19411 CPU Memory consumed at the end of the eval (end-begin): 0 CPU Peak Memory consumed during the eval (max-begin): 0 CPU Total Peak Memory consumed during the eval (max): 19411 accuracy=100.0 eval_preds[:10]=['no complaint', 'no complaint', 'complaint', 'complaint', 'no complaint', 'no complaint', 'no complaint', 'complaint', 'complaint', 'no complaint'] dataset['train'][label_column][:10]=['no complaint', 'no complaint', 'complaint', 'complaint', 'no complaint', 'no complaint', 'no complaint', 'complaint', 'complaint', 'no complaint'] ``` ### Example of PEFT model inference using 🤗 Accelerate's Big Model Inferencing capabilities An example is provided in [this notebook](https://github.com/huggingface/peft/blob/main/examples/causal_language_modeling/peft_lora_clm_accelerate_big_model_inference.ipynb). ## Models support matrix Find models that are supported out of the box below. Note that PEFT works with almost all models -- if it is not listed, you just need to [do some manual configuration](https://huggingface.co/docs/peft/developer_guides/custom_models). ### Causal Language Modeling | Model | LoRA | Prefix Tuning | P-Tuning | Prompt Tuning | IA3 | |--------------| ---- | ---- | ---- | ---- | ---- | | GPT-2 | ✅ | ✅ | ✅ | ✅ | ✅ | | Bloom | ✅ | ✅ | ✅ | ✅ | ✅ | | OPT | ✅ | ✅ | ✅ | ✅ | ✅ | | GPT-Neo | ✅ | ✅ | ✅ | ✅ | ✅ | | GPT-J | ✅ | ✅ | ✅ | ✅ | ✅ | | GPT-NeoX-20B | ✅ | ✅ | ✅ | ✅ | ✅ | | LLaMA | ✅ | ✅ | ✅ | ✅ | ✅ | | ChatGLM | ✅ | ✅ | ✅ | ✅ | ✅ | | Mistral | ✅ | | | | | ### Conditional Generation | Model | LoRA | Prefix Tuning | P-Tuning | Prompt Tuning | IA3 | | --------- | ---- | ---- | ---- | ---- | ---- | | T5 | ✅ | ✅ | ✅ | ✅ | ✅ | | BART | ✅ | ✅ | ✅ | ✅ | ✅ | ### Sequence Classification | Model | LoRA | Prefix Tuning | P-Tuning | Prompt Tuning | IA3 | | --------- | ---- | ---- | ---- | ---- | ---- | | BERT | ✅ | ✅ | ✅ | ✅ | ✅ | | RoBERTa | ✅ | ✅ | ✅ | ✅ | ✅ | | GPT-2 | ✅ | ✅ | ✅ | ✅ | | | Bloom | ✅ | ✅ | ✅ | ✅ | | | OPT | ✅ | ✅ | ✅ | ✅ | | | GPT-Neo | ✅ | ✅ | ✅ | ✅ | | | GPT-J | ✅ | ✅ | ✅ | ✅ | | | Deberta | ✅ | | ✅ | ✅ | | | Deberta-v2 | ✅ | | ✅ | ✅ | | ### Token Classification | Model | LoRA | Prefix Tuning | P-Tuning | Prompt Tuning | IA3 | | --------- | ---- | ---- | ---- | ---- | ---- | | BERT | ✅ | ✅ | | | | | RoBERTa | ✅ | ✅ | | | | | GPT-2 | ✅ | ✅ | | | | | Bloom | ✅ | ✅ | | | | | OPT | ✅ | ✅ | | | | | GPT-Neo | ✅ | ✅ | | | | | GPT-J | ✅ | ✅ | | | | | Deberta | ✅ | | | | | | Deberta-v2 | ✅ | | | | | ### Text-to-Image Generation | Model | LoRA | LoHa | LoKr | OFT | Prefix Tuning | P-Tuning | Prompt Tuning | IA3 | | --------- | ---- | ---- | ---- | ---- | ---- | ---- | ---- | ---- | | Stable Diffusion | ✅ | ✅ | ✅ | ✅ | | | | ### Image Classification | Model | LoRA | Prefix Tuning | P-Tuning | Prompt Tuning | IA3 | | --------- | ---- | ---- | ---- | ---- | ---- | | ViT | ✅ | | | | | | Swin | ✅ | | | | | ### Image to text (Multi-modal models) | Model | LoRA | Prefix Tuning | P-Tuning | Prompt Tuning | IA3 | --------- | ---- | ---- | ---- | ---- | ---- | | Blip-2 | ✅ | | | | | ___Note that we have tested LoRA for [ViT](https://huggingface.co/docs/transformers/model_doc/vit) and [Swin](https://huggingface.co/docs/transformers/model_doc/swin) for fine-tuning on image classification. However, it should be possible to use LoRA for any compatible model [provided](https://huggingface.co/models?pipeline_tag=image-classification&sort=downloads&search=vit) by 🤗 Transformers. Check out the respective examples to learn more. If you run into problems, please open an issue.___ The same principle applies to our [segmentation models](https://huggingface.co/models?pipeline_tag=image-segmentation&sort=downloads) as well. ### Semantic Segmentation | Model | LoRA | Prefix Tuning | P-Tuning | Prompt Tuning | IA3 | | --------- | ---- | ---- | ---- | ---- | ---- | | SegFormer | ✅ | | | | | ## Caveats: 1. Below is an example of using PyTorch FSDP for training. However, it doesn't lead to any GPU memory savings. Please refer issue [[FSDP] FSDP with CPU offload consumes 1.65X more GPU memory when training models with most of the params frozen](https://github.com/pytorch/pytorch/issues/91165). ```python from peft.utils.other import fsdp_auto_wrap_policy ... if os.environ.get("ACCELERATE_USE_FSDP", None) is not None: accelerator.state.fsdp_plugin.auto_wrap_policy = fsdp_auto_wrap_policy(model) model = accelerator.prepare(model) ``` Example of parameter efficient tuning with [`mt0-xxl`](https://huggingface.co/bigscience/mt0-xxl) base model using 🤗 Accelerate is provided in `~examples/conditional_generation/peft_lora_seq2seq_accelerate_fsdp.py`. a. First, run `accelerate config --config_file fsdp_config.yaml` and answer the questionnaire. Below are the contents of the config file. ```yaml command_file: null commands: null compute_environment: LOCAL_MACHINE deepspeed_config: {} distributed_type: FSDP downcast_bf16: 'no' dynamo_backend: 'NO' fsdp_config: fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP fsdp_backward_prefetch_policy: BACKWARD_PRE fsdp_offload_params: true fsdp_sharding_strategy: 1 fsdp_state_dict_type: FULL_STATE_DICT fsdp_transformer_layer_cls_to_wrap: T5Block gpu_ids: null machine_rank: 0 main_process_ip: null main_process_port: null main_training_function: main megatron_lm_config: {} mixed_precision: 'no' num_machines: 1 num_processes: 2 rdzv_backend: static same_network: true tpu_name: null tpu_zone: null use_cpu: false ``` b. run the below command to launch the example script ```bash accelerate launch --config_file fsdp_config.yaml examples/peft_lora_seq2seq_accelerate_fsdp.py ``` 2. When using ZeRO3 with zero3_init_flag=True, if you find the gpu memory increase with training steps. we might need to update deepspeed after [deepspeed commit 42858a9891422abc](https://github.com/microsoft/DeepSpeed/commit/42858a9891422abcecaa12c1bd432d28d33eb0d4) . The related issue is [[BUG] Peft Training with Zero.Init() and Zero3 will increase GPU memory every forward step ](https://github.com/microsoft/DeepSpeed/issues/3002) ## 🤗 PEFT as a utility library ### Injecting adapters directly into the model Inject trainable adapters on any `torch` model using `inject_adapter_in_model` method. Note the method will make no further change to the model. ```python import torch from peft import inject_adapter_in_model, LoraConfig class DummyModel(torch.nn.Module): def __init__(self): super().__init__() self.embedding = torch.nn.Embedding(10, 10) self.linear = torch.nn.Linear(10, 10) self.lm_head = torch.nn.Linear(10, 10) def forward(self, input_ids): x = self.embedding(input_ids) x = self.linear(x) x = self.lm_head(x) return x lora_config = LoraConfig( lora_alpha=16, lora_dropout=0.1, r=64, bias="none", target_modules=["linear"], ) model = DummyModel() model = inject_adapter_in_model(lora_config, model) dummy_inputs = torch.LongTensor([[0, 1, 2, 3, 4, 5, 6, 7]]) dummy_outputs = model(dummy_inputs) ``` Learn more about the [low level API in the docs](https://huggingface.co/docs/peft/developer_guides/low_level_api). ### Mixing different adapter types Ususally, it is not possible to combine different adapter types in the same model, e.g. combining LoRA with AdaLoRA, LoHa, or LoKr. Using a mixed model, this can, however, be achieved: ```python from peft import PeftMixedModel model = AutoModelForCausalLM.from_pretrained("hf-internal-testing/tiny-random-OPTForCausalLM").eval() peft_model = PeftMixedModel.from_pretrained(model, <path-to-adapter-0>, "adapter0") peft_model.load_adapter(<path-to-adapter-1>, "adapter1") peft_model.set_adapter(["adapter0", "adapter1"]) result = peft_model(**inputs) ``` The main intent is to load already trained adapters and use this only for inference. However, it is also possible to create a PEFT model for training by passing `mixed=True` to `get_peft_model`: ```python from peft import get_peft_model, LoraConfig, LoKrConfig base_model = ... config0 = LoraConfig(...) config1 = LoKrConfig(...) peft_model = get_peft_model(base_model, config0, "adapter0", mixed=True) peft_model.add_adapter(config1, "adapter1") peft_model.set_adapter(["adapter0", "adapter1"]) for batch in dataloader: ... ``` ## Contributing If you would like to contribute to PEFT, please check out our [contributing guide](https://huggingface.co/docs/peft/developer_guides/contributing). ## Citing 🤗 PEFT If you use 🤗 PEFT in your publication, please cite it by using the following BibTeX entry. ```bibtex @Misc{peft, title = {PEFT: State-of-the-art Parameter-Efficient Fine-Tuning methods}, author = {Sourab Mangrulkar and Sylvain Gugger and Lysandre Debut and Younes Belkada and Sayak Paul and Benjamin Bossan}, howpublished = {\url{https://github.com/huggingface/peft}}, year = {2022} } ```

peft/README.md/0

<!--Copyright 2023 The HuggingFace Team. All rights reserved. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. ⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be rendered properly in your Markdown viewer. --> # LoRA LoRA is low-rank decomposition method to reduce the number of trainable parameters which speeds up finetuning large models and uses less memory. In PEFT, using LoRA is as easy as setting up a [`LoraConfig`] and wrapping it with [`get_peft_model`] to create a trainable [`PeftModel`]. This guide explores in more detail other options and features for using LoRA. ## Initialization The initialization of LoRA weights is controlled by the parameter `init_lora_weights` in [`LoraConfig`]. By default, PEFT initializes LoRA weights with Kaiming-uniform for weight A and zeros for weight B resulting in an identity transform (same as the reference [implementation](https://github.com/microsoft/LoRA)). It is also possible to pass `init_lora_weights="gaussian"`. As the name suggests, this initializes weight A with a Gaussian distribution and zeros for weight B (this is how [Diffusers](https://huggingface.co/docs/diffusers/index) initializes LoRA weights). ```py from peft import LoraConfig config = LoraConfig(init_lora_weights="gaussian", ...) ``` There is also an option to set `init_lora_weights=False` which is useful for debugging and testing. This should be the only time you use this option. When choosing this option, the LoRA weights are initialized such that they do *not* result in an identity transform. ```py from peft import LoraConfig config = LoraConfig(init_lora_weights=False, ...) ``` ### LoftQ When quantizing the base model for QLoRA training, consider using the [LoftQ initialization](https://arxiv.org/abs/2310.08659), which has been shown to improve performance when training quantized models. The idea is that the LoRA weights are initialized such that the quantization error is minimized. If you're using LoftQ, *do not* quantize the base model. You should set up a [`LoftQConfig`] instead: ```python from peft import LoftQConfig, LoraConfig, get_peft_model base_model = AutoModelForCausalLM.from_pretrained(...) # don't quantize here loftq_config = LoftQConfig(loftq_bits=4, ...) # set 4bit quantization lora_config = LoraConfig(..., init_lora_weights="loftq", loftq_config=loftq_config) peft_model = get_peft_model(base_model, lora_config) ``` <Tip> Learn more about how PEFT works with quantization in the [Quantization](quantization) guide. </Tip> ### Rank-stabilized LoRA Another way to initialize [`LoraConfig`] is with the [rank-stabilized LoRA (rsLoRA)](https://huggingface.co/papers/2312.03732) method. The LoRA architecture scales each adapter during every forward pass by a fixed scalar which is set at initialization and depends on the rank `r`. The scalar is given by `lora_alpha/r` in the original implementation, but rsLoRA uses `lora_alpha/math.sqrt(r)` which stabilizes the adapters and increases the performance potential from using a higher `r`. ```py from peft import LoraConfig config = LoraConfig(use_rslora=True, ...) ``` ## Merge adapters While LoRA is significantly smaller and faster to train, you may encounter latency issues during inference due to separately loading the base model and the LoRA adapter. To eliminate latency, use the [`~LoraModel.merge_and_unload`] function to merge the adapter weights with the base model. This allows you to use the newly merged model as a standalone model. The [`~LoraModel.merge_and_unload`] function doesn't keep the adapter weights in memory. ```py from transformers import AutoModelForCausalLM from peft import PeftModel base_model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1") peft_model_id = "alignment-handbook/zephyr-7b-sft-lora" model = PeftModel.from_pretrained(base_model, peft_model_id) model.merge_and_unload() ``` If you need to keep a copy of the weights so you can unmerge the adapter later or delete and load different ones, you should use the [`~LoraModel.merge_adapter`] function instead. Now you have the option to use [`~LoraModel.unmerge_adapter`] to return the base model. ```py from transformers import AutoModelForCausalLM from peft import PeftModel base_model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1") peft_model_id = "alignment-handbook/zephyr-7b-sft-lora" model = PeftModel.from_pretrained(base_model, peft_model_id) model.merge_adapter() # unmerge the LoRA layers from the base model model.unmerge_adapter() ``` The [`~LoraModel.add_weighted_adapter`] function is useful for merging multiple LoRAs into a new adapter based on a user provided weighting scheme in the `weights` parameter. Below is an end-to-end example. First load the base model: ```python from transformers import AutoModelForCausalLM from peft import PeftModel import torch base_model = AutoModelForCausalLM.from_pretrained( "mistralai/Mistral-7B-v0.1", torch_dtype=torch.float16, device_map="auto" ) ``` Then we load the first adapter: ```python peft_model_id = "alignment-handbook/zephyr-7b-sft-lora" model = PeftModel.from_pretrained(base_model, peft_model_id, adapter_name="sft") ``` Then load a different adapter and merge it with the first one: ```python model.load_adapter("alignment-handbook/zephyr-7b-dpo-lora", adapter_name="dpo") model.add_weighted_adapter( adapters=["sft", "dpo"], weights=[0.7, 0.3], adapter_name="sft-dpo", combination_type="linear" ) ``` <Tip> There are several supported methods for `combination_type`. Refer to the [documentation](../package_reference/lora#peft.LoraModel.add_weighted_adapter) for more details. Note that "svd" as the `combination_type` is not supported when using `torch.float16` or `torch.bfloat16` as the datatype. </Tip> Now, perform inference: ```python tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1") prompt = "Hey, are you conscious? Can you talk to me?" inputs = tokenizer(prompt, return_tensors="pt") inputs = {k: v.to("cuda") for k, v in inputs.items()} with torch.no_grad(): generate_ids = model.generate(**inputs, max_length=30) outputs = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0] print(outputs) ``` ## Load adapters Adapters can be loaded onto a pretrained model with [`~PeftModel.load_adapter`], which is useful for trying out different adapters whose weights aren't merged. Set the active adapter weights with the [`~LoraModel.set_adapter`] function. ```py from transformers import AutoModelForCausalLM from peft import PeftModel base_model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1") peft_model_id = "alignment-handbook/zephyr-7b-sft-lora" model = PeftModel.from_pretrained(base_model, peft_model_id) # load different adapter model.load_adapter("alignment-handbook/zephyr-7b-dpo-lora", adapter_name="dpo") # set adapter as active model.set_adapter("dpo") ``` To return the base model, you could use [`~LoraModel.unload`] to unload all of the LoRA modules or [`~LoraModel.delete_adapter`] to delete the adapter entirely. ```py # unload adapter model.unload() # delete adapter model.delete_adapter("dpo") ``` ## QLoRA-style training The default LoRA settings in 🤗PEFT follow the [original paper](https://hf.co/papers/2106.09685) and add trainable weights to the query and value layers of each attention block. However, in [QLoRA](https://hf.co/papers/2305.14314), it was found that adding trainable weights to all the linear layers of a transformer model is beneficial to match full-finetuning performance. Since the list of modules to add will vary depending on the architecture, we provided a convenient shorthand : simple specify `target_modules='all-linear'` and let 🤗PEFT handle the rest: ```py config = LoraConfig(target_modules="all-linear", ...) # adds LoRA to all linear layers like in QLoRA ```

peft/docs/source/developer_guides/lora.md/0

<!--Copyright 2023 The HuggingFace Team. All rights reserved. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. ⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be rendered properly in your Markdown viewer. --> # Multitask Prompt Tuning [Multitask Prompt Tuning](https://huggingface.co/papers/2303.02861) decomposes the soft prompts of each task into a single learned transferable prompt instead of a separate prompt for each task. The single learned prompt can be adapted for each task by multiplicative low rank updates. The abstract from the paper is: *Prompt tuning, in which a base pretrained model is adapted to each task via conditioning on learned prompt vectors, has emerged as a promising approach for efficiently adapting large language models to multiple downstream tasks. However, existing methods typically learn soft prompt vectors from scratch, and it has not been clear how to exploit the rich cross-task knowledge with prompt vectors in a multitask learning setting. We propose multitask prompt tuning (MPT), which first learns a single transferable prompt by distilling knowledge from multiple task-specific source prompts. We then learn multiplicative low rank updates to this shared prompt to efficiently adapt it to each downstream target task. Extensive experiments on 23 NLP datasets demonstrate that our proposed approach outperforms the state-of-the-art methods, including the full finetuning baseline in some cases, despite only tuning 0.035% as many task-specific parameters*. ## MultitaskPromptTuningConfig [[autodoc]] tuners.multitask_prompt_tuning.config.MultitaskPromptTuningConfig ## MultitaskPromptEmbedding [[autodoc]] tuners.multitask_prompt_tuning.model.MultitaskPromptEmbedding

peft/docs/source/package_reference/multitask_prompt_tuning.md/0

<!--⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be rendered properly in your Markdown viewer. --> # LoRA for token classification Low-Rank Adaptation (LoRA) is a reparametrization method that aims to reduce the number of trainable parameters with low-rank representations. The weight matrix is broken down into low-rank matrices that are trained and updated. All the pretrained model parameters remain frozen. After training, the low-rank matrices are added back to the original weights. This makes it more efficient to store and train a LoRA model because there are significantly fewer parameters. <Tip> 💡 Read [LoRA: Low-Rank Adaptation of Large Language Models](https://arxiv.org/abs/2106.09685) to learn more about LoRA. </Tip> This guide will show you how to train a [`roberta-large`](https://huggingface.co/roberta-large) model with LoRA on the [BioNLP2004](https://huggingface.co/datasets/tner/bionlp2004) dataset for token classification. Before you begin, make sure you have all the necessary libraries installed: ```bash !pip install -q peft transformers datasets evaluate seqeval ``` ## Setup Let's start by importing all the necessary libraries you'll need: - 🤗 Transformers for loading the base `roberta-large` model and tokenizer, and handling the training loop - 🤗 Datasets for loading and preparing the `bionlp2004` dataset for training - 🤗 Evaluate for evaluating the model's performance - 🤗 PEFT for setting up the LoRA configuration and creating the PEFT model ```py from datasets import load_dataset from transformers import ( AutoModelForTokenClassification, AutoTokenizer, DataCollatorForTokenClassification, TrainingArguments, Trainer, ) from peft import get_peft_config, PeftModel, PeftConfig, get_peft_model, LoraConfig, TaskType import evaluate import torch import numpy as np model_checkpoint = "roberta-large" lr = 1e-3 batch_size = 16 num_epochs = 10 ``` ## Load dataset and metric The [BioNLP2004](https://huggingface.co/datasets/tner/bionlp2004) dataset includes tokens and tags for biological structures like DNA, RNA and proteins. Load the dataset: ```py bionlp = load_dataset("tner/bionlp2004") bionlp["train"][0] { "tokens": [ "Since", "HUVECs", "released", "superoxide", "anions", "in", "response", "to", "TNF", ",", "and", "H2O2", "induces", "VCAM-1", ",", "PDTC", "may", "act", "as", "a", "radical", "scavenger", ".", ], "tags": [0, 7, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0], } ``` The `tags` values are defined in the label ids [dictionary](https://huggingface.co/datasets/tner/bionlp2004#label-id). The letter that prefixes each label indicates the token position: `B` is for the first token of an entity, `I` is for a token inside the entity, and `0` is for a token that is not part of an entity. ```py { "O": 0, "B-DNA": 1, "I-DNA": 2, "B-protein": 3, "I-protein": 4, "B-cell_type": 5, "I-cell_type": 6, "B-cell_line": 7, "I-cell_line": 8, "B-RNA": 9, "I-RNA": 10, } ``` Then load the [`seqeval`](https://huggingface.co/spaces/evaluate-metric/seqeval) framework which includes several metrics - precision, accuracy, F1, and recall - for evaluating sequence labeling tasks. ```py seqeval = evaluate.load("seqeval") ``` Now you can write an evaluation function to compute the metrics from the model predictions and labels, and return the precision, recall, F1, and accuracy scores: ```py label_list = [ "O", "B-DNA", "I-DNA", "B-protein", "I-protein", "B-cell_type", "I-cell_type", "B-cell_line", "I-cell_line", "B-RNA", "I-RNA", ] def compute_metrics(p): predictions, labels = p predictions = np.argmax(predictions, axis=2) true_predictions = [ [label_list[p] for (p, l) in zip(prediction, label) if l != -100] for prediction, label in zip(predictions, labels) ] true_labels = [ [label_list[l] for (p, l) in zip(prediction, label) if l != -100] for prediction, label in zip(predictions, labels) ] results = seqeval.compute(predictions=true_predictions, references=true_labels) return { "precision": results["overall_precision"], "recall": results["overall_recall"], "f1": results["overall_f1"], "accuracy": results["overall_accuracy"], } ``` ## Preprocess dataset Initialize a tokenizer and make sure you set `is_split_into_words=True` because the text sequence has already been split into words. However, this doesn't mean it is tokenized yet (even though it may look like it!), and you'll need to further tokenize the words into subwords. ```py tokenizer = AutoTokenizer.from_pretrained(model_checkpoint, add_prefix_space=True) ``` You'll also need to write a function to: 1. Map each token to their respective word with the [`~transformers.BatchEncoding.word_ids`] method. 2. Ignore the special tokens by setting them to `-100`. 3. Label the first token of a given entity. ```py def tokenize_and_align_labels(examples): tokenized_inputs = tokenizer(examples["tokens"], truncation=True, is_split_into_words=True) labels = [] for i, label in enumerate(examples[f"tags"]): word_ids = tokenized_inputs.word_ids(batch_index=i) previous_word_idx = None label_ids = [] for word_idx in word_ids: if word_idx is None: label_ids.append(-100) elif word_idx != previous_word_idx: label_ids.append(label[word_idx]) else: label_ids.append(-100) previous_word_idx = word_idx labels.append(label_ids) tokenized_inputs["labels"] = labels return tokenized_inputs ``` Use [`~datasets.Dataset.map`] to apply the `tokenize_and_align_labels` function to the dataset: ```py tokenized_bionlp = bionlp.map(tokenize_and_align_labels, batched=True) ``` Finally, create a data collator to pad the examples to the longest length in a batch: ```py data_collator = DataCollatorForTokenClassification(tokenizer=tokenizer) ``` ## Train Now you're ready to create a [`PeftModel`]. Start by loading the base `roberta-large` model, the number of expected labels, and the `id2label` and `label2id` dictionaries: ```py id2label = { 0: "O", 1: "B-DNA", 2: "I-DNA", 3: "B-protein", 4: "I-protein", 5: "B-cell_type", 6: "I-cell_type", 7: "B-cell_line", 8: "I-cell_line", 9: "B-RNA", 10: "I-RNA", } label2id = { "O": 0, "B-DNA": 1, "I-DNA": 2, "B-protein": 3, "I-protein": 4, "B-cell_type": 5, "I-cell_type": 6, "B-cell_line": 7, "I-cell_line": 8, "B-RNA": 9, "I-RNA": 10, } model = AutoModelForTokenClassification.from_pretrained( model_checkpoint, num_labels=11, id2label=id2label, label2id=label2id ) ``` Define the [`LoraConfig`] with: - `task_type`, token classification (`TaskType.TOKEN_CLS`) - `r`, the dimension of the low-rank matrices - `lora_alpha`, scaling factor for the weight matrices - `lora_dropout`, dropout probability of the LoRA layers - `bias`, set to `all` to train all bias parameters <Tip> 💡 The weight matrix is scaled by `lora_alpha/r`, and a higher `lora_alpha` value assigns more weight to the LoRA activations. For performance, we recommend setting `bias` to `None` first, and then `lora_only`, before trying `all`. </Tip> ```py peft_config = LoraConfig( task_type=TaskType.TOKEN_CLS, inference_mode=False, r=16, lora_alpha=16, lora_dropout=0.1, bias="all" ) ``` Pass the base model and `peft_config` to the [`get_peft_model`] function to create a [`PeftModel`]. You can check out how much more efficient training the [`PeftModel`] is compared to fully training the base model by printing out the trainable parameters: ```py model = get_peft_model(model, peft_config) model.print_trainable_parameters() "trainable params: 1855499 || all params: 355894283 || trainable%: 0.5213624069370061" ``` From the 🤗 Transformers library, create a [`~transformers.TrainingArguments`] class and specify where you want to save the model to, the training hyperparameters, how to evaluate the model, and when to save the checkpoints: ```py training_args = TrainingArguments( output_dir="roberta-large-lora-token-classification", learning_rate=lr, per_device_train_batch_size=batch_size, per_device_eval_batch_size=batch_size, num_train_epochs=num_epochs, weight_decay=0.01, evaluation_strategy="epoch", save_strategy="epoch", load_best_model_at_end=True, ) ``` Pass the model, `TrainingArguments`, datasets, tokenizer, data collator and evaluation function to the [`~transformers.Trainer`] class. The `Trainer` handles the training loop for you, and when you're ready, call [`~transformers.Trainer.train`] to begin! ```py trainer = Trainer( model=model, args=training_args, train_dataset=tokenized_bionlp["train"], eval_dataset=tokenized_bionlp["validation"], tokenizer=tokenizer, data_collator=data_collator, compute_metrics=compute_metrics, ) trainer.train() ``` ## Share model Once training is complete, you can store and share your model on the Hub if you'd like. Log in to your Hugging Face account and enter your token when prompted: ```py from huggingface_hub import notebook_login notebook_login() ``` Upload the model to a specific model repository on the Hub with the [`~transformers.PreTrainedModel.push_to_hub`] method: ```py model.push_to_hub("your-name/roberta-large-lora-token-classification") ``` ## Inference To use your model for inference, load the configuration and model: ```py peft_model_id = "stevhliu/roberta-large-lora-token-classification" config = PeftConfig.from_pretrained(peft_model_id) inference_model = AutoModelForTokenClassification.from_pretrained( config.base_model_name_or_path, num_labels=11, id2label=id2label, label2id=label2id ) tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path) model = PeftModel.from_pretrained(inference_model, peft_model_id) ``` Get some text to tokenize: ```py text = "The activation of IL-2 gene expression and NF-kappa B through CD28 requires reactive oxygen production by 5-lipoxygenase." inputs = tokenizer(text, return_tensors="pt") ``` Pass the inputs to the model, and print out the model prediction for each token: ```py with torch.no_grad(): logits = model(**inputs).logits tokens = inputs.tokens() predictions = torch.argmax(logits, dim=2) for token, prediction in zip(tokens, predictions[0].numpy()): print((token, model.config.id2label[prediction])) ("<s>", "O") ("The", "O") ("Ġactivation", "O") ("Ġof", "O") ("ĠIL", "B-DNA") ("-", "O") ("2", "I-DNA") ("Ġgene", "O") ("Ġexpression", "O") ("Ġand", "O") ("ĠNF", "B-protein") ("-", "O") ("k", "I-protein") ("appa", "I-protein") ("ĠB", "I-protein") ("Ġthrough", "O") ("ĠCD", "B-protein") ("28", "I-protein") ("Ġrequires", "O") ("Ġreactive", "O") ("Ġoxygen", "O") ("Ġproduction", "O") ("Ġby", "O") ("Ġ5", "B-protein") ("-", "O") ("lip", "I-protein") ("oxy", "I-protein") ("gen", "I-protein") ("ase", "I-protein") (".", "O") ("</s>", "O") ```

peft/docs/source/task_guides/token-classification-lora.md/0

import argparse import gc import json import logging import math import os from dataclasses import dataclass from datetime import datetime from pathlib import Path from random import randint from typing import Any, Dict, List, Union # datasets imports import datasets # metric imports import evaluate import numpy as np import torch import transformers import wandb # accelerate imports from accelerate import Accelerator, dispatch_model from accelerate.logging import get_logger from datasets import Audio, DatasetDict, IterableDatasetDict, interleave_datasets, load_dataset # hf imports from huggingface_hub import Repository from torch.utils.data import DataLoader from tqdm import tqdm from transformers import ( SchedulerType, WhisperForConditionalGeneration, WhisperProcessor, get_scheduler, set_seed, ) from transformers.models.whisper.english_normalizer import BasicTextNormalizer from transformers.utils import get_full_repo_name # peft imports from peft import AdaLoraConfig, LoraConfig, PeftModel, get_peft_model logger = get_logger(__name__, log_level="INFO") def parse_args(): parser = argparse.ArgumentParser(description="Whisper Fine-Tuning with AdaLora") parser.add_argument( "--model_name_or_path", type=str, help="Path to pretrained model or model identifier from huggingface.co/models.", required=True, ) parser.add_argument("--language", type=str, help="Language to use for training; e.g., 'Hindi' ", required=True) parser.add_argument("--language_abbr", type=str, help="Language to use for training; e.g., 'hi' ", required=True) parser.add_argument( "--task", type=str, default="transcribe", help="Task to use for training; e.g., 'transcribe' ", required=False ) parser.add_argument( "--dataset_name", type=str, default="mozilla-foundation/common_voice_11_0", help="Dataset to use for training; e.g., 'whisper' ", required=False, ) parser.add_argument( "--dataset_in_streaming_mode", action="store_true", help="Whether to use streaming mode for the dataset.", ) parser.add_argument( "--do_lower_case", action="store_true", help="lowercase the transcribed text before tokenizing" ) parser.add_argument( "--do_remove_punctuation", action="store_true", help="remove punctuation from the transcribed text" ) parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.") parser.add_argument( "--overwrite_cache", type=bool, default=False, help="Overwrite the cached training and evaluation sets" ) parser.add_argument("--max_audio_input_length", type=float, default=30.0, help="Maximum audio length in seconds.") parser.add_argument( "--preprocessing_num_workers", type=int, default=None, help="The number of processes to use for the preprocessing.", ) parser.add_argument( "--per_device_train_batch_size", type=int, default=8, help="Batch size (per device) for the training dataloader.", ) parser.add_argument( "--per_device_eval_batch_size", type=int, default=8, help="Batch size (per device) for the evaluation dataloader.", ) parser.add_argument( "--buffer_size", type=int, default=5000, help="Number of samples to prefetch in the streaming mode.", ) parser.add_argument( "--dataloader_pin_memory", action="store_true", help="Whether or not to pin memory for the DataLoader.", ) parser.add_argument( "--dataloader_num_workers", type=int, default=0, help="Number of subprocesses to use for data loading.", ) parser.add_argument( "--learning_rate", type=float, default=5e-5, help="Initial learning rate (after the potential warmup period) to use.", ) parser.add_argument("--weight_decay", type=float, default=0.0, help="Weight decay to use.") parser.add_argument("--num_train_epochs", type=int, default=3, help="Total number of training epochs to perform.") parser.add_argument( "--max_train_steps", type=int, default=None, help="Total number of training steps to perform. If provided, overrides num_train_epochs.", ) parser.add_argument( "--gradient_accumulation_steps", type=int, default=1, help="Number of updates steps to accumulate before performing a backward/update pass.", ) parser.add_argument( "--lr_scheduler_type", type=SchedulerType, default="linear", help="The scheduler type to use.", choices=["linear", "cosine", "cosine_with_restarts", "polynomial", "constant", "constant_with_warmup"], ) parser.add_argument( "--num_warmup_steps", type=int, default=0, help="Number of steps for the warmup in the lr scheduler." ) parser.add_argument("--output_dir", type=str, default=None, help="Where to store the final model.") parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.") parser.add_argument( "--load_best_model", action="store_true", help="Whether to load the best model at the end of training", ) parser.add_argument( "--with_tracking", action="store_true", help="Whether to enable experiment trackers for logging.", ) parser.add_argument( "--report_to", type=str, default="all", help=( 'The integration to report the results and logs to. Supported platforms are `"tensorboard"`,' ' `"wandb"` and `"comet_ml"`. Use `"all"` (default) to report to all integrations.' "Only applicable when `--with_tracking` is passed." ), ) parser.add_argument("--hub_token", type=str, help="The token to use to push to the Model Hub.") parser.add_argument( "--hub_model_id", type=str, help="The name of the repository to keep in sync with the local `output_dir`." ) parser.add_argument( "--checkpointing_steps", type=int, default=500, help="Whether the various states should be saved at the end of every n steps, or 'epoch' for each epoch.", ) parser.add_argument( "--logging_steps", type=int, default=100, help="Whether the various states should be saved at the end of every n steps, or 'epoch' for each epoch.", ) parser.add_argument( "--evaluation_steps", type=int, default=500, help="Whether the various states should be saved at the end of every n steps, or 'epoch' for each epoch.", ) parser.add_argument( "--resume_from_checkpoint", type=str, default=None, help="If the training should continue from a checkpoint folder.", ) # lora/adalora specific args parser.add_argument( "--use_peft", action="store_true", help="Whether to use PEFT", ) parser.add_argument( "--use_adalora", action="store_true", help="Whether to use AdaLoRA or LoRA. If set, uses AdaLoRA instead of the default LoRA.", ) parser.add_argument( "--init_r", type=int, default=12, help="Initial AdaLoRA rank", ) parser.add_argument( "--target_r", type=int, default=4, help="Target AdaLoRA rank", ) parser.add_argument( "--tinit", type=int, default=200, help="number of warmup steps for AdaLoRA wherein no pruning is performed", ) parser.add_argument( "--tfinal", type=int, default=1000, help=" fix the resulting budget distribution and fine-tune the model for tfinal steps when using AdaLoRA ", ) parser.add_argument( "--delta_t", type=int, default=10, help="interval of steps for AdaLoRA to update rank", ) parser.add_argument( "--lora_alpha", type=int, default=32, help="LORA alpha", ) parser.add_argument( "--r", type=int, default=8, help="LORA rank", ) parser.add_argument( "--lora_dropout", type=float, default=0.1, help="LORA dropout", ) parser.add_argument( "--orth_reg_weight", type=float, default=0.5, help="Orthogonal regularization weight", ) parser.add_argument( "--debug_mode", action="store_true", help="Whether to use debug mode", ) args = parser.parse_args() if args.push_to_hub: assert args.output_dir is not None, "Need an `output_dir` to create a repo when `--push_to_hub` is passed." return args def load_streaming_dataset(dataset_name, dataset_config_name, split, **kwargs): if "+" in split: # load multiple splits separated by the `+` symbol *with* streaming mode dataset_splits = [ load_dataset(dataset_name, dataset_config_name, split=split_name, streaming=True, **kwargs) for split_name in split.split("+") ] # interleave multiple splits to form one dataset interleaved_dataset = interleave_datasets(dataset_splits) return interleaved_dataset else: # load a single split *with* streaming mode dataset = load_dataset(dataset_name, dataset_config_name, split=split, streaming=True, **kwargs) return dataset def prepare_dataset_wrapper(do_lower_case, do_remove_punctuation, processor, normalizer): def prepare_dataset(batch): # load and (possibly) resample audio data to 16kHz audio = batch["audio"] # compute log-Mel input features from input audio array batch["input_features"] = processor.feature_extractor( audio["array"], sampling_rate=audio["sampling_rate"] ).input_features[0] # compute input length of audio sample in seconds batch["input_length"] = len(audio["array"]) / audio["sampling_rate"] # optional pre-processing steps transcription = batch["sentence"] if do_lower_case: transcription = transcription.lower() if do_remove_punctuation: transcription = normalizer(transcription).strip() # encode target text to label ids batch["labels"] = processor.tokenizer(transcription).input_ids return batch return prepare_dataset def save_model_hook(models, weights, output_dir): for model in models: model.save_pretrained(output_dir) # make sure to pop weight so that corresponding model is not saved again weights.pop() def load_model_hook(models, input_dir): while len(models) > 0: model = models.pop() # pop models so that they are not loaded again PeftModel.from_pretrained(model.base_model.model, input_dir) @dataclass class DataCollatorSpeechSeq2SeqWithPadding: processor: Any def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]: # split inputs and labels since they have to be of different lengths and need different padding methods # first treat the audio inputs by simply returning torch tensors input_features = [{"input_features": feature["input_features"]} for feature in features] batch = self.processor.feature_extractor.pad(input_features, return_tensors="pt") # get the tokenized label sequences label_features = [{"input_ids": feature["labels"]} for feature in features] # pad the labels to max length labels_batch = self.processor.tokenizer.pad(label_features, return_tensors="pt") # replace padding with -100 to ignore loss correctly labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100) # if bos token is appended in previous tokenization step, # cut bos token here as it's append later anyways if (labels[:, 0] == self.processor.tokenizer.bos_token_id).all().cpu().item(): labels = labels[:, 1:] batch["labels"] = labels return batch def get_audio_length_processor(max_input_length): def is_audio_in_length_range(length): return length < max_input_length return is_audio_in_length_range def evaluation_loop(model, eval_dataloader, processor, normalizer, metric, forced_decoder_ids, accelerator): model.eval() predictions = [] references = [] normalized_predictions = [] normalized_references = [] for _, batch in enumerate(tqdm(eval_dataloader)): with torch.cuda.amp.autocast(): with torch.no_grad(): generated_tokens = ( model.generate( input_features=batch["input_features"], forced_decoder_ids=forced_decoder_ids, max_new_tokens=255, ) .cpu() .numpy() ) labels = batch["labels"].cpu().numpy() labels = np.where(labels != -100, labels, processor.tokenizer.pad_token_id) decoded_preds = processor.tokenizer.batch_decode(generated_tokens, skip_special_tokens=True) decoded_labels = processor.tokenizer.batch_decode(labels, skip_special_tokens=True) predictions.extend(decoded_preds) references.extend(decoded_labels) normalized_predictions.extend([normalizer(pred).strip() for pred in decoded_preds]) normalized_references.extend([normalizer(label).strip() for label in decoded_labels]) del generated_tokens, labels, batch gc.collect() wer = 100 * metric.compute(predictions=predictions, references=references) normalized_wer = 100 * metric.compute(predictions=normalized_predictions, references=normalized_references) eval_metrics = {"eval/wer": wer, "eval/normalized_wer": normalized_wer} if accelerator.get_tracker("wandb"): sample_size = min(len(predictions), 256) ids = [randint(0, len(predictions) - 1) for p in range(0, sample_size)] sample_predictions = [predictions[i] for i in ids] sample_references = [references[i] for i in ids] sample_normalized_predictions = [normalized_predictions[i] for i in ids] sample_normalized_references = [normalized_references[i] for i in ids] table_rows = [ list(r) for r in zip( sample_predictions, sample_references, sample_normalized_predictions, sample_normalized_references ) ] eval_metrics["eval_samples"] = wandb.Table( columns=["predictions", "references", "normalized_predictions", "normalized_references"], rows=table_rows, ) return eval_metrics def main(): args = parse_args() accelerator_kwargs = {"gradient_accumulation_steps": args.gradient_accumulation_steps} if args.with_tracking: accelerator_kwargs["log_with"] = args.report_to accelerator_kwargs["project_dir"] = args.output_dir accelerator = Accelerator(**accelerator_kwargs) # Make one log on every process with the configuration for debugging. logging.basicConfig( format="%(asctime)s - %(levelname)s - %(name)s - %(message)s", datefmt="%m/%d/%Y %H:%M:%S", level=logging.INFO, ) logger.info(accelerator.state, main_process_only=False) if accelerator.is_local_main_process: datasets.utils.logging.set_verbosity_warning() transformers.utils.logging.set_verbosity_info() else: datasets.utils.logging.set_verbosity_error() transformers.utils.logging.set_verbosity_error() # If passed along, set the training seed now. if args.seed is not None: set_seed(args.seed) # Handle the repository creation if accelerator.is_main_process: if args.push_to_hub: if args.hub_model_id is None: repo_name = get_full_repo_name(Path(args.output_dir).name, token=args.hub_token) else: repo_name = args.hub_model_id repo = Repository(args.output_dir, clone_from=repo_name) with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore: if "step_*" not in gitignore: gitignore.write("step_*\n") if "epoch_*" not in gitignore: gitignore.write("epoch_*\n") elif args.output_dir is not None: os.makedirs(args.output_dir, exist_ok=True) accelerator.wait_for_everyone() # load dataset either in streaming mode or not processor = WhisperProcessor.from_pretrained(args.model_name_or_path, language=args.language, task=args.task) normalizer = BasicTextNormalizer() prepare_dataset = prepare_dataset_wrapper(args.do_lower_case, args.do_remove_punctuation, processor, normalizer) is_audio_in_length_range = get_audio_length_processor(args.max_audio_input_length) data_collator = DataCollatorSpeechSeq2SeqWithPadding(processor=processor) if args.dataset_in_streaming_mode: raw_datasets = IterableDatasetDict() loading_method = load_streaming_dataset else: raw_datasets = DatasetDict() loading_method = load_dataset if args.debug_mode: train_split = "train[:100]" test_split = "test[:10]" else: train_split = "train+validation" test_split = "test" raw_datasets["train"] = loading_method( args.dataset_name, args.language_abbr, split=train_split, use_auth_token=True ) raw_datasets["test"] = loading_method(args.dataset_name, args.language_abbr, split=test_split, use_auth_token=True) raw_datasets = raw_datasets.cast_column("audio", Audio(sampling_rate=16000)) logger.info("Dataset loaded: %s", raw_datasets) logger.info(f'{raw_datasets["train"][0]}') vectorized_datasets = raw_datasets.map( prepare_dataset, remove_columns=list(next(iter(raw_datasets.values())).features), num_proc=args.preprocessing_num_workers, ).with_format("torch") if args.dataset_in_streaming_mode: vectorized_datasets["train"] = vectorized_datasets["train"].shuffle( buffer_size=args.buffer_size, seed=args.seed, ) # filter out audio files that are too long from the training set is_audio_in_length_range = get_audio_length_processor(args.max_audio_input_length) vectorized_datasets["train"] = vectorized_datasets["train"].filter( is_audio_in_length_range, input_columns=["input_length"] ) # get dataloaders train_dataloader = DataLoader( vectorized_datasets["train"], batch_size=args.per_device_train_batch_size, shuffle=True, collate_fn=data_collator, num_workers=args.dataloader_num_workers, pin_memory=args.dataloader_pin_memory, ) eval_dataloader = DataLoader( vectorized_datasets["test"], batch_size=args.per_device_eval_batch_size, collate_fn=data_collator, num_workers=args.dataloader_num_workers, pin_memory=args.dataloader_pin_memory, ) # metric metric = evaluate.load("wer") # model model = WhisperForConditionalGeneration.from_pretrained(args.model_name_or_path, load_in_8bit=True) model.config.forced_decoder_ids = None model.config.suppress_tokens = [] if len(set(model.hf_device_map.values()).intersection({"cpu", "disk"})) > 0: raise ValueError("Training on CPU or disk is not supported.") if len(set(model.hf_device_map.values())) > 1: device_map = model.hf_device_map.copy() # required because `labels` are on main execution device (0) while the output of `proj_out` is on other device. # So, this leads to device mismatch error when calculation cross-entropy between logits and labels. # Won't arise during inference as `labels` aren't supplied during that time # instead of changing device of one of the tied modules, I have to do this for all tied modules # else the execution device of remaining tied modules isn't changed device_map["model.decoder.embed_tokens"] = model._hf_hook.execution_device device_map["model.decoder.embed_positions"] = model._hf_hook.execution_device device_map["proj_out"] = model._hf_hook.execution_device dispatch_model(model, device_map=device_map) # preparing peft model if args.use_peft: from peft import prepare_model_for_int8_training model = prepare_model_for_int8_training(model) # as Whisper model uses Conv layer in encoder, checkpointing disables grad computation # to avoid this, make the inputs trainable def make_inputs_require_grad(module, input, output): output.requires_grad_(True) model.model.encoder.conv1.register_forward_hook(make_inputs_require_grad) # wrapping model with adalora tuner if args.use_adalora: config = AdaLoraConfig( init_r=args.init_r, target_r=args.target_r, beta1=0.85, beta2=0.85, tinit=args.tinit, tfinal=args.tfinal, deltaT=args.delta_t, lora_alpha=args.lora_alpha, lora_dropout=args.lora_dropout, target_modules=["k_proj", "q_proj", "v_proj", "out_proj", "fc1", "fc2"], orth_reg_weight=args.orth_reg_weight, ) else: config = LoraConfig( r=args.r, lora_alpha=args.lora_alpha, target_modules=["q_proj", "v_proj"], lora_dropout=args.lora_dropout, ) model = get_peft_model(model, config) model.print_trainable_parameters() # optimizer optimizer = torch.optim.AdamW(model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay) if args.max_train_steps is None: num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps) args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch else: args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch) # scheduler lr_scheduler = get_scheduler( name=args.lr_scheduler_type, optimizer=optimizer, num_warmup_steps=args.num_warmup_steps, num_training_steps=args.max_train_steps, ) # Prepare everything with our `accelerator`. model, optimizer, train_dataloader, eval_dataloader, lr_scheduler = accelerator.prepare( model, optimizer, train_dataloader, eval_dataloader, lr_scheduler ) accelerator.print(model) # Note here that the max steps is adjusted by the accelerator's num_processes args.max_train_steps = math.ceil(args.max_train_steps / accelerator.num_processes) if args.use_peft and args.use_adalora: model.base_model.peft_config["default"].total_step = args.max_train_steps # model.base_model.peft_config.total_step = args.max_train_steps # We need to initialize the trackers we use, and also store our configuration. # The trackers initializes automatically on the main process. if args.with_tracking: run_name = f"run-{datetime.now().strftime('%Y-%m-%d_%H-%M-%S')}" experiment_config = vars(args) # TensorBoard cannot log Enums, need the raw value experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value accelerator.init_trackers( "Whisper PEFT Fine-Tuning", config=experiment_config, init_kwargs={"wandb": {"name": run_name}} ) # saving and loading checkpoints for resuming training accelerator.register_save_state_pre_hook(save_model_hook) accelerator.register_load_state_pre_hook(load_model_hook) total_batch_size = args.per_device_train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps logger.info("***** Running training *****") logger.info(f" Num Epochs = {args.num_train_epochs}") logger.info(f" Instantaneous batch size per device = {args.per_device_train_batch_size}") logger.info(f" Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}") logger.info(f" Gradient Accumulation steps = {args.gradient_accumulation_steps}") logger.info(f" Total optimization steps = {args.max_train_steps}") # Only show the progress bar once on each machine. progress_bar = tqdm(range(args.max_train_steps), disable=not accelerator.is_local_main_process) global_step = 0 starting_epoch = 0 best_metric = None resume_step = 0 forced_decoder_ids = processor.get_decoder_prompt_ids(language=args.language, task=args.task) # Potentially load in the weights and states from a previous save if args.resume_from_checkpoint: accelerator.load_state(args.resume_from_checkpoint) path = os.path.basename(args.resume_from_checkpoint) training_difference = os.path.splitext(path)[0] global_step = resume_step = int(training_difference.replace("step_", "")) starting_epoch = resume_step // len(train_dataloader) resume_step -= starting_epoch * len(train_dataloader) # We need to adjust the progress bar to the current step progress_bar.update(resume_step) for epoch in range(starting_epoch, args.num_train_epochs): model.train() if args.with_tracking: total_loss = 0 running_loss = 0 for step, batch in enumerate(accelerator.skip_first_batches(train_dataloader, num_batches=resume_step)): with accelerator.accumulate(model): outputs = model(**batch) loss = outputs.loss accelerator.backward(loss) optimizer.step() lr_scheduler.step() # Update the importance of low-rank matrices # and allocate the budget accordingly. # This is only needed for AdaLora. # Note that this requires parameter gradients. # Hence being called before optimizer.zero_grad(). if args.use_peft and args.use_adalora: model.update_and_allocate(global_step) optimizer.zero_grad() global_step += 1 progress_bar.update(1) if args.with_tracking: step_loss = accelerator.reduce(loss.detach().clone()).item() total_loss += step_loss running_loss += step_loss if global_step % args.checkpointing_steps == 0: output_dir = os.path.join(args.output_dir, f"step_{global_step}") accelerator.save_state(output_dir) if global_step % args.logging_steps == 0: if args.with_tracking: accelerator.log({"train/running_loss": running_loss / args.logging_steps}, step=global_step) running_loss = 0 if global_step % args.evaluation_steps == 0: eval_metrics = evaluation_loop( model, eval_dataloader, processor, normalizer, metric, forced_decoder_ids, accelerator ) if args.with_tracking: logger.info(f"Step {global_step} eval metrics: {eval_metrics}") accelerator.log(eval_metrics, step=global_step) if best_metric is None or eval_metrics["eval/wer"] < best_metric: best_metric = eval_metrics["eval/wer"] accelerator.save_state(os.path.join(args.output_dir, "best_checkpoint")) model.train() if global_step >= args.max_train_steps: break if args.with_tracking: train_epoch_loss = total_loss / (step + 1) logger.info(f"Epoch {epoch} train loss: {train_epoch_loss}") accelerator.log({"epoch/train_loss": train_epoch_loss}, step=epoch) if args.push_to_hub and epoch <= args.num_train_epochs - 1: accelerator.wait_for_everyone() unwrapped_model = accelerator.unwrap_model(model) unwrapped_model.save_pretrained(args.output_dir, is_main_process=accelerator.is_main_process) # evaluate the model at the end of training eval_metrics = evaluation_loop( model, eval_dataloader, processor, normalizer, metric, forced_decoder_ids, accelerator ) if args.with_tracking: logger.info(f"Step {global_step} eval metrics: {eval_metrics}") accelerator.log(eval_metrics, step=global_step) if best_metric is None or eval_metrics["eval/wer"] < best_metric: best_metric = eval_metrics["eval/wer"] accelerator.save_state(os.path.join(args.output_dir, "best_checkpoint")) if accelerator.is_main_process: processor.tokenizer.save_pretrained(args.output_dir) repo.push_to_hub( commit_message=f"Training in progress epoch {epoch}", blocking=False, auto_lfs_prune=True ) if args.load_best_model: # load the best model accelerator.load_state(os.path.join(args.output_dir, "best_checkpoint")) model.resize_modules_by_rank_pattern(model.peft_config["default"].rank_pattern, "default") eval_metrics = evaluation_loop( model, eval_dataloader, processor, normalizer, metric, forced_decoder_ids, accelerator ) if args.with_tracking: best_metrics = {"best_" + k: v for k, v in eval_metrics.items()} accelerator.log(best_metrics, step=global_step) accelerator.wait_for_everyone() unwrapped_model = accelerator.unwrap_model(model) unwrapped_model.save_pretrained(args.output_dir, is_main_process=accelerator.is_main_process) if accelerator.is_main_process: processor.tokenizer.save_pretrained(args.output_dir) if args.push_to_hub: repo.push_to_hub(commit_message="End of training", auto_lfs_prune=True) with open(os.path.join(args.output_dir, "all_results.json"), "w") as f: eval_metrics.pop("eval_samples") json.dump(eval_metrics, f) if __name__ == "__main__": main()

peft/examples/int8_training/peft_adalora_whisper_large_training.py/0

# coding=utf-8 # Copyright 2023-present the HuggingFace Inc. team. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from typing import Any import torch from peft.import_utils import is_bnb_4bit_available, is_bnb_available from .layer import AdaLoraLayer if is_bnb_available(): class SVDLinear8bitLt(torch.nn.Module, AdaLoraLayer): # Low-rank matrix for SVD-based adaptation def __init__( self, base_layer: torch.nn.Module, adapter_name: str, r: int = 0, lora_alpha: int = 1, lora_dropout: float = 0.0, init_lora_weights: bool = True, **kwargs, ) -> None: super().__init__() AdaLoraLayer.__init__(self, base_layer) # Freezing the pre-trained weight matrix self.get_base_layer().weight.requires_grad = False self._active_adapter = adapter_name self.update_layer(adapter_name, r, lora_alpha, lora_dropout, init_lora_weights) def forward(self, x: torch.Tensor) -> torch.Tensor: # note: no check for self.merged because merging is not supported (yet) result = self.base_layer(x) if self.disable_adapters: return result for active_adapter in self.active_adapters: if active_adapter not in self.lora_A.keys(): continue requires_conversion = not torch.is_autocast_enabled() if requires_conversion: expected_dtype = result.dtype if x.dtype != torch.float32: x = x.float() lora_A = self.lora_A[active_adapter] lora_B = self.lora_B[active_adapter] lora_E = self.lora_E[active_adapter] dropout = self.lora_dropout[active_adapter] scaling = self.scaling[active_adapter] ranknum = self.ranknum[active_adapter] + 1e-5 output = dropout(x) @ (lora_A * lora_E).T @ lora_B.T if requires_conversion: output = output.to(expected_dtype) output = output * scaling / ranknum # inplace operation on view is forbidden for MatMul8bitLtBackward, so avoid it result = result + output return result def __repr__(self) -> str: rep = super().__repr__() return "adalora." + rep if is_bnb_4bit_available(): class SVDLinear4bit(torch.nn.Module, AdaLoraLayer): # Low-rank matrix for SVD-based adaptation def __init__( self, base_layer: torch.nn.Module, adapter_name: str, r: int = 0, lora_alpha: int = 1, lora_dropout: float = 0.0, init_lora_weights: bool = True, **kwargs, ) -> None: super().__init__() AdaLoraLayer.__init__(self, base_layer) # Freezing the pre-trained weight matrix self.get_base_layer().weight.requires_grad = False self._active_adapter = adapter_name self.update_layer(adapter_name, r, lora_alpha, lora_dropout, init_lora_weights) def forward(self, x: torch.Tensor, *args: Any, **kwargs: Any) -> torch.Tensor: # note: no check for self.merged because merging is not supported (yet) result = self.base_layer(x, *args, **kwargs) if self.disable_adapters: return result # As per Tim Dettmers, for 4bit, we need to defensively clone here. # The reason is that in some cases, an error can occur that backprop # does not work on a manipulated view. This issue may be solved with # newer PyTorch versions but this would need extensive testing to be # sure. result = result.clone() for active_adapter in self.active_adapters: if active_adapter not in self.lora_A.keys(): continue lora_A = self.lora_A[active_adapter] lora_B = self.lora_B[active_adapter] lora_E = self.lora_E[active_adapter] dropout = self.lora_dropout[active_adapter] scaling = self.scaling[active_adapter] ranknum = self.ranknum[active_adapter] + 1e-5 requires_conversion = not torch.is_autocast_enabled() if requires_conversion: expected_dtype = result.dtype compute_dtype = lora_A.dtype if x.dtype != compute_dtype: x = x.to(compute_dtype) output = dropout(x) @ (lora_A * lora_E).T @ lora_B.T if requires_conversion: output = output.to(expected_dtype) output = output * scaling / ranknum result += output return result def __repr__(self) -> str: rep = super().__repr__() return "adalora." + rep

peft/src/peft/tuners/adalora/bnb.py/0

# coding=utf-8 # Copyright 2023-present the HuggingFace Inc. team. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from __future__ import annotations import warnings from typing import Any, Optional, Union from torch import nn from tqdm import tqdm from peft.tuners import adalora, loha, lokr, lora, oft from peft.tuners.tuners_utils import BaseTuner, BaseTunerLayer, check_target_module_exists from peft.utils import ( TRANSFORMERS_MODELS_TO_LORA_TARGET_MODULES_MAPPING, ModulesToSaveWrapper, PeftType, _get_submodules, get_auto_gptq_quant_linear, ) # Collection of constants used for all tuners COMPATIBLE_TUNER_TYPES = (PeftType.LORA, PeftType.LOHA, PeftType.LOKR, PeftType.ADALORA, PeftType.OFT) PREFIXES = [lora.LoraModel.prefix, lokr.LoKrModel.prefix, loha.LoHaModel.prefix, oft.OFTModel.prefix] Configs = Union[lora.LoraConfig, loha.LoHaConfig, lokr.LoKrConfig, adalora.AdaLoraConfig, oft.OFTConfig] Layers = (lora.layer.LoraLayer, loha.layer.LoHaLayer, lokr.layer.LoKrLayer, adalora.layer.AdaLoraLayer, oft.OFTLayer) class MixedModel(BaseTuner): """ A class that allows to mix different types of adapters in a single model. Note: This class should usually not be initialized directly. Instead, use `get_peft_model` with the argument `mixed=True`. Args: model (:obj:`nn.Module`): The model to be tuned. config (:obj:`PeftConfig`): The config of the model to be tuned. The adapter type must be compatible. adapter_name (:obj:`str`): The name of the first adapter. """ def __init__(self, model: nn.Module, config: Configs, adapter_name: str) -> None: super().__init__(model, config, adapter_name) def _check_new_adapter_config(self, config: Configs) -> None: """ A helper method to check the config when a new adapter is being added. Raise a ValueError if there is something wrong with the config or if it conflicts with existing adapters. """ if not isinstance(config, Configs.__args__): raise ValueError( f"{self.__class__.__name__} only supports {COMPATIBLE_TUNER_TYPES} configs, but got {type(config)}." ) biases = (getattr(config, "bias", None) for config in self.peft_config) biases = [bias for bias in biases if bias not in (None, "none")] if len(biases) > 1: raise ValueError( f"{self.__class__.__name__} supports only 1 adapter with bias. When using multiple adapters, " "set bias to 'none' for all adapters." ) @staticmethod def _check_target_module_exists(config: Configs, key: str): return check_target_module_exists(config, key) def _create_and_replace( self, config: Configs, *args: Any, **kwargs: Any, ) -> None: if isinstance(config, adalora.AdaLoraConfig): adalora.AdaLoraModel._create_and_replace(self, config, *args, **kwargs) elif isinstance(config, lora.LoraConfig): lora.LoraModel._create_and_replace(self, config, *args, **kwargs) elif isinstance(config, loha.LoHaConfig): loha.LoHaModel._create_and_replace(self, config, *args, **kwargs) elif isinstance(config, lokr.LoKrConfig): lokr.LoKrModel._create_and_replace(self, config, *args, **kwargs) elif isinstance(config, oft.OFTConfig): oft.OFTModel._create_and_replace(self, config, *args, **kwargs) else: raise ValueError(f"Unsupported config type {type(config)}, should be one of {COMPATIBLE_TUNER_TYPES}.") def _replace_module(self, parent, child_name, new_module, child) -> None: setattr(parent, child_name, new_module) # It's not necessary to set requires_grad here, as that is handled by # _mark_only_adapters_as_trainable # child layer wraps the original module, unpack it if hasattr(child, "base_layer"): child = child.get_base_layer() elif hasattr(child, "quant_linear_module"): # TODO maybe not necessary to have special treatment? child = child.quant_linear_module if not hasattr(new_module, "base_layer"): new_module.weight = child.weight if hasattr(child, "bias"): new_module.bias = child.bias if getattr(child, "state", None) is not None: if hasattr(new_module, "base_layer"): new_module.base_layer.state = child.state else: new_module.state = child.state new_module.to(child.weight.device) # dispatch to correct device for name, module in new_module.named_modules(): if any(prefix in name for prefix in PREFIXES): module.to(child.weight.device) if "ranknum" in name: module.to(child.weight.device) def _mark_only_adapters_as_trainable(self, model: nn.Module) -> None: for n, p in model.named_parameters(): if not any(prefix in n for prefix in PREFIXES): p.requires_grad = False for active_adapter in self.active_adapters: bias = getattr(self.peft_config[active_adapter], "bias", "none") if bias == "none": continue if bias == "all": for n, p in model.named_parameters(): if "bias" in n: p.requires_grad = True elif bias == "lora_only": # TODO: check if this is needed for other supported types for m in model.modules(): if isinstance(m, Layers) and hasattr(m, "bias") and m.bias is not None: m.bias.requires_grad = True else: raise ValueError(f"Requested bias: {bias}, is not implemented.") @staticmethod def _create_new_module(config, adapter_name, target, **kwargs): gptq_quantization_config = kwargs.get("gptq_quantization_config", None) AutoGPTQQuantLinear = get_auto_gptq_quant_linear(gptq_quantization_config) if (gptq_quantization_config is not None) or (AutoGPTQQuantLinear is not None): raise ValueError(f"GPTQ quantization not supported for {config.peft_type.value} (yet).") loaded_in_8bit = kwargs.pop("loaded_in_8bit", False) loaded_in_4bit = kwargs.pop("loaded_in_4bit", False) if loaded_in_8bit or loaded_in_4bit: raise ValueError(f"8bit and 4bit quantization not supported for {config.peft_type.value} (yet).") if isinstance(config, adalora.AdaLoraConfig): new_module = adalora.AdaLoraModel._create_new_module(config, adapter_name, target, **kwargs) elif isinstance(config, lora.LoraConfig): new_module = lora.LoraModel._create_new_module(config, adapter_name, target, **kwargs) elif isinstance(config, loha.LoHaConfig): new_module = loha.LoHaModel._create_new_module(config, adapter_name, target, **kwargs) elif isinstance(config, lokr.LoKrConfig): new_module = lokr.LoKrModel._create_new_module(config, adapter_name, target, **kwargs) elif isinstance(config, oft.OFTConfig): new_module = oft.OFTModel._create_new_module(config, adapter_name, target, **kwargs) else: raise ValueError(f"Unknown config type {type(config)}, should be one of {COMPATIBLE_TUNER_TYPES}.") return new_module def __getattr__(self, name: str): """Forward missing attributes to the wrapped module.""" try: return super().__getattr__(name) # defer to nn.Module's logic except AttributeError: return getattr(self.model, name) def _set_adapter_layers(self, enabled=True): for module in self.model.modules(): if isinstance(module, (BaseTunerLayer, ModulesToSaveWrapper)): module.enable_adapters(enabled) def enable_adapter_layers(self): self._set_adapter_layers(enabled=True) def disable_adapter_layers(self): for active_adapter in self.active_adapters: val = getattr(self.peft_config[active_adapter], "bias", "none") if val != "none": msg = ( f"Careful, disabling adapter layers with bias configured to be '{val}' does not produce the same " "output as the the base model would without adaption." ) warnings.warn(msg) self._set_adapter_layers(enabled=False) def set_adapter(self, adapter_name: Union[str, list[str]]) -> None: for module in self.model.modules(): if isinstance(module, Layers): if module.merged: warnings.warn("Adapter cannot be set when the model is merged. Unmerging the model first.") module.unmerge() module.set_adapter(adapter_name) self.active_adapter = adapter_name @staticmethod def _prepare_adapter_config(peft_config, model_config): if peft_config.target_modules is None: if model_config["model_type"] not in TRANSFORMERS_MODELS_TO_LORA_TARGET_MODULES_MAPPING: raise ValueError("Please specify `target_modules` in `peft_config`") peft_config.target_modules = set( TRANSFORMERS_MODELS_TO_LORA_TARGET_MODULES_MAPPING[model_config["model_type"]] ) return peft_config def _unload_and_optionally_merge( self, merge=True, progressbar: bool = False, safe_merge: bool = False, adapter_names: Optional[list[str]] = None, ): if merge: if getattr(self.model, "quantization_method", None) == "gptq": raise ValueError("Cannot merge layers when the model is gptq quantized") def merge_recursively(module): # helper function to recursively merge the base_layer of the target path = [] layer = module while hasattr(layer, "base_layer"): path.append(layer) layer = layer.base_layer for layer_before, layer_after in zip(path[:-1], path[1:]): layer_after.merge(safe_merge=safe_merge, adapter_names=adapter_names) layer_before.base_layer = layer_after.base_layer module.merge(safe_merge=safe_merge, adapter_names=adapter_names) key_list = [key for key, _ in self.model.named_modules() if not any(prefix in key for prefix in PREFIXES)] desc = "Unloading " + ("and merging " if merge else "") + "model" for key in tqdm(key_list, disable=not progressbar, desc=desc): try: parent, target, target_name = _get_submodules(self.model, key) except AttributeError: continue if hasattr(target, "base_layer"): if merge: merge_recursively(target) self._replace_module(parent, target_name, target.get_base_layer(), target) elif isinstance(target, ModulesToSaveWrapper): # save any additional trainable modules part of `modules_to_save` setattr(parent, target_name, target.modules_to_save[target.active_adapter]) return self.model def add_weighted_adapter(self, *args: Any, **kwargs: Any) -> None: raise NotImplementedError(f"Weighted adapters are not supported for {self.__class__.__name__} (yet).") def delete_adapter(self, adapter_name: Union[str, list[str]]) -> None: """ Deletes an existing adapter. Args: adapter_name (Union[str, list[str]]): Name of the adapter(s) to delete. """ if isinstance(adapter_name, str): adapter_names = [adapter_name] else: adapter_names = adapter_name mismatched = set(adapter_names) - set(self.peft_config.keys()) if mismatched: raise ValueError( f"Adapter(s) {sorted(mismatched)} not found, available adapters: {sorted(self.peft_config.keys())}" ) for adapter_name in adapter_names: del self.peft_config[adapter_name] key_list = [key for key, _ in self.model.named_modules() if not any(prefix in key for prefix in PREFIXES)] new_adapter = None for key in key_list: _, target, _ = _get_submodules(self.model, key) if isinstance(target, BaseTunerLayer): target.delete_adapter(adapter_name) if new_adapter is None: new_adapter = target.active_adapters[:] self.active_adapter = new_adapter or [] def merge_and_unload( self, progressbar: bool = False, safe_merge: bool = False, adapter_names: Optional[list[str]] = None ) -> nn.Module: r""" This method merges the layers into the base model. This is needed if someone wants to use the base model as a standalone model. Args: progressbar (`bool`): whether to show a progressbar indicating the unload and merge process safe_merge (`bool`): whether to activate the safe merging check to check if there is any potential Nan in the adapter weights adapter_names (`List[str]`, *optional*): The list of adapter names that should be merged. If None, all active adapters will be merged. Defaults to `None`. """ return self._unload_and_optionally_merge( progressbar=progressbar, safe_merge=safe_merge, adapter_names=adapter_names ) def unload(self) -> nn.Module: """ Gets back the base model by removing all the lora modules without merging. This gives back the original base model. """ return self._unload_and_optionally_merge(merge=False) def generate(self, *args: Any, **kwargs: Any): return self.model.generate(*args, **kwargs)

peft/src/peft/tuners/mixed/model.py/0

# coding=utf-8 # Copyright 2023-present the HuggingFace Inc. team. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import importlib import os import tempfile import unittest from unittest import TestCase import torch from torch.testing import assert_close from peft.mapping import get_peft_model from peft.peft_model import PeftModel from peft.tuners.adaption_prompt import AdaptionPromptConfig from peft.utils.other import prepare_model_for_int8_training from peft.utils.save_and_load import get_peft_model_state_dict from tests.testing_common import PeftCommonTester def is_llama_available() -> bool: """Check if Llama is available in the transformers library (it's not in earlier versions).""" try: return importlib.util.find_spec("transformers.models.llama.modeling_llama") is not None except ModuleNotFoundError: return False if is_llama_available(): # We guard the import statement so that our unit tests will pass in CI environments # that don't have a transformers package with Llama. from transformers import LlamaConfig, LlamaForCausalLM, LlamaModel class AdaptionPromptTester(TestCase, PeftCommonTester): """ Tests for the AdaptionPrompt model. Some of these tests were adapted from `test_peft_model.py` (which has been refactored since), but since we haven't checked in the test checkpoints for Llama into `hf-internal-testing`, we separate them for now. """ def setUp(self): # Check that llama is available in transformers package before running each test. if not is_llama_available(): self.skipTest("Llama not available in transformers. Skipping test.") @staticmethod def _create_test_llama_config(): """Create a test config for a small Llama model for testing.""" return LlamaConfig( vocab_size=16, hidden_size=8, intermediate_size=8, num_hidden_layers=8, num_attention_heads=4, use_cache=False, ) def test_attributes(self) -> None: model = LlamaModel(self._create_test_llama_config()) config = AdaptionPromptConfig(adapter_layers=1, adapter_len=4) model = get_peft_model(model, config) self.assertTrue(hasattr(model, "save_pretrained")) self.assertTrue(hasattr(model, "from_pretrained")) self.assertTrue(hasattr(model, "push_to_hub")) def test_prepare_for_training(self) -> None: model = LlamaForCausalLM(self._create_test_llama_config()) config = AdaptionPromptConfig(adapter_layers=1, adapter_len=4, task_type="CAUSAL_LM") model = get_peft_model(model, config) model = model.to(self.torch_device) dummy_input = torch.LongTensor([[1, 1, 1]]).to(self.torch_device) dummy_output = model.get_input_embeddings()(dummy_input) self.assertTrue(not dummy_output.requires_grad) def test_prepare_for_int8_training(self) -> None: model = LlamaForCausalLM(self._create_test_llama_config()) model = prepare_model_for_int8_training(model) model = model.to(self.torch_device) for param in model.parameters(): self.assertTrue(not param.requires_grad) config = AdaptionPromptConfig(adapter_layers=1, adapter_len=4, task_type="CAUSAL_LM") model = get_peft_model(model, config) # For backward compatibility if hasattr(model, "enable_input_require_grads"): model.enable_input_require_grads() else: def make_inputs_require_grad(module, input, output): output.requires_grad_(True) model.get_input_embeddings().register_forward_hook(make_inputs_require_grad) dummy_input = torch.LongTensor([[1, 1, 1]]).to(self.torch_device) dummy_output = model.get_input_embeddings()(dummy_input) self.assertTrue(dummy_output.requires_grad) def test_save_pretrained_regression(self) -> None: seed = 420 torch.manual_seed(seed) model = LlamaForCausalLM(self._create_test_llama_config()) config = AdaptionPromptConfig(adapter_layers=2, adapter_len=4, task_type="CAUSAL_LM") model = get_peft_model(model, config) model = model.to(self.torch_device) with tempfile.TemporaryDirectory() as tmp_dirname: model.save_pretrained(tmp_dirname, safe_serialization=False) torch.manual_seed(seed) model_from_pretrained = LlamaForCausalLM(self._create_test_llama_config()) model_from_pretrained = PeftModel.from_pretrained(model_from_pretrained, tmp_dirname) # check if the state dicts are equal state_dict = get_peft_model_state_dict(model) state_dict_from_pretrained = get_peft_model_state_dict(model_from_pretrained) # check if same keys self.assertEqual(state_dict.keys(), state_dict_from_pretrained.keys()) # Check that the number of saved parameters is 4 -- 2 layers of (tokens and gate). self.assertEqual(len(list(state_dict.keys())), 4) # check if tensors equal for key in state_dict.keys(): self.assertTrue( torch.allclose( state_dict[key].to(self.torch_device), state_dict_from_pretrained[key].to(self.torch_device) ) ) # check if `adapter_model.bin` is present self.assertTrue(os.path.exists(os.path.join(tmp_dirname, "adapter_model.bin"))) # check if `adapter_config.json` is present self.assertTrue(os.path.exists(os.path.join(tmp_dirname, "adapter_config.json"))) # check if `model.safetensors` is not present self.assertFalse(os.path.exists(os.path.join(tmp_dirname, "model.safetensors"))) # check if `config.json` is not present self.assertFalse(os.path.exists(os.path.join(tmp_dirname, "config.json"))) def test_save_pretrained(self) -> None: seed = 420 torch.manual_seed(seed) model = LlamaForCausalLM(self._create_test_llama_config()) config = AdaptionPromptConfig(adapter_layers=2, adapter_len=4, task_type="CAUSAL_LM") model = get_peft_model(model, config) model = model.to(self.torch_device) with tempfile.TemporaryDirectory() as tmp_dirname: model.save_pretrained(tmp_dirname) torch.manual_seed(seed) model_from_pretrained = LlamaForCausalLM(self._create_test_llama_config()) model_from_pretrained = PeftModel.from_pretrained(model_from_pretrained, tmp_dirname) # check if the state dicts are equal state_dict = get_peft_model_state_dict(model) state_dict_from_pretrained = get_peft_model_state_dict(model_from_pretrained) # check if same keys self.assertEqual(state_dict.keys(), state_dict_from_pretrained.keys()) # Check that the number of saved parameters is 4 -- 2 layers of (tokens and gate). self.assertEqual(len(list(state_dict.keys())), 4) # check if tensors equal for key in state_dict.keys(): self.assertTrue( torch.allclose( state_dict[key].to(self.torch_device), state_dict_from_pretrained[key].to(self.torch_device) ) ) # check if `adapter_model.bin` is present self.assertTrue(os.path.exists(os.path.join(tmp_dirname, "adapter_model.safetensors"))) # check if `adapter_config.json` is present self.assertTrue(os.path.exists(os.path.join(tmp_dirname, "adapter_config.json"))) # check if `model.safetensors` is not present self.assertFalse(os.path.exists(os.path.join(tmp_dirname, "model.safetensors"))) # check if `config.json` is not present self.assertFalse(os.path.exists(os.path.join(tmp_dirname, "config.json"))) def test_save_pretrained_selected_adapters(self) -> None: seed = 420 torch.manual_seed(seed) model = LlamaForCausalLM(self._create_test_llama_config()) config = AdaptionPromptConfig(adapter_layers=2, adapter_len=4, task_type="CAUSAL_LM") model = get_peft_model(model, config) model = model.to(self.torch_device) new_adapter_config = AdaptionPromptConfig(adapter_layers=2, adapter_len=4, task_type="CAUSAL_LM") model.add_adapter("new_adapter", new_adapter_config) with tempfile.TemporaryDirectory() as tmp_dirname: model.save_pretrained(tmp_dirname) torch.manual_seed(seed) model_from_pretrained = LlamaForCausalLM(self._create_test_llama_config()) model_from_pretrained = PeftModel.from_pretrained(model_from_pretrained, tmp_dirname) model_from_pretrained.load_adapter(tmp_dirname, "new_adapter") # check if the state dicts are equal state_dict = get_peft_model_state_dict(model) state_dict_from_pretrained = get_peft_model_state_dict(model_from_pretrained) # check if same keys self.assertEqual(state_dict.keys(), state_dict_from_pretrained.keys()) # Check that the number of saved parameters is 4 -- 2 layers of (tokens and gate). self.assertEqual(len(list(state_dict.keys())), 4) # check if tensors equal for key in state_dict.keys(): self.assertTrue( torch.allclose( state_dict[key].to(self.torch_device), state_dict_from_pretrained[key].to(self.torch_device) ) ) # check if `adapter_model.bin` is present self.assertTrue(os.path.exists(os.path.join(tmp_dirname, "adapter_model.safetensors"))) # check if `adapter_config.json` is present self.assertTrue(os.path.exists(os.path.join(tmp_dirname, "adapter_config.json"))) # check if `model.safetensors` is not present self.assertFalse(os.path.exists(os.path.join(tmp_dirname, "model.safetensors"))) # check if `config.json` is not present self.assertFalse(os.path.exists(os.path.join(tmp_dirname, "config.json"))) def test_generate(self) -> None: model = LlamaForCausalLM(self._create_test_llama_config()) config = AdaptionPromptConfig(adapter_layers=2, adapter_len=4, task_type="CAUSAL_LM") model = get_peft_model(model, config) model = model.to(self.torch_device) input_ids = torch.LongTensor([[1, 1, 1], [2, 1, 2]]).to(self.torch_device) attention_mask = torch.LongTensor([[1, 1, 1], [1, 0, 1]]).to(self.torch_device) # check if `generate` works _ = model.generate(input_ids=input_ids, attention_mask=attention_mask) # check if `generate` works if positional arguments are passed _ = model.generate(input_ids, attention_mask=attention_mask) def test_sequence_adapter_ops(self) -> None: """Test sequence of adapter operations.""" # Test input data. input_ids = torch.LongTensor([[1, 1, 1], [2, 1, 2]]).to(self.torch_device) target_ids = torch.LongTensor([[0, 0, 0], [0, 0, 0]]).to(self.torch_device) attention_mask = torch.LongTensor([[1, 1, 1], [1, 0, 1]]).to(self.torch_device) # Create original llama model. original = LlamaForCausalLM(self._create_test_llama_config()) original = original.to(self.torch_device) original_before = original(input_ids=input_ids, attention_mask=attention_mask) # Get AdaptionPrompt model. adapted = get_peft_model( original, AdaptionPromptConfig(adapter_layers=2, adapter_len=4, task_type="CAUSAL_LM") ) adapted = adapted.to(self.torch_device) default_before = adapted(input_ids=input_ids, attention_mask=attention_mask, labels=target_ids) # Test zero-init: The logits should be exactly the same. assert_close(original_before.logits, default_before.logits, rtol=0, atol=0) # Single fine-tuning step on "default" adapter. optimizer = torch.optim.SGD(adapted.parameters(), lr=1) optimizer.zero_grad() default_before.loss.backward() optimizer.step() # Test that the output changed. default_after = adapted(input_ids=input_ids, attention_mask=attention_mask, labels=target_ids) self.assertFalse(torch.allclose(default_before.logits, default_after.logits)) with adapted.disable_adapter(): # Test that the output is the same as the original ouput. default_disabled = adapted(input_ids=input_ids, attention_mask=attention_mask, labels=target_ids) assert_close(original_before.logits, default_disabled.logits, rtol=0, atol=0) # Add new adapter 1. adapted.add_adapter("adapter 1", AdaptionPromptConfig(adapter_layers=3, adapter_len=8, task_type="CAUSAL_LM")) # Test zero-init adapter_1_before = adapted(input_ids=input_ids, attention_mask=attention_mask, labels=target_ids) assert_close(original_before.logits, adapter_1_before.logits, rtol=0, atol=0) # Single fine-tuning step on adapter 1. optimizer = torch.optim.SGD(adapted.parameters(), lr=1) optimizer.zero_grad() adapter_1_before.loss.backward() optimizer.step() # Test that adapter 1 output changed. adapter_1_after = adapted(input_ids=input_ids, attention_mask=attention_mask, labels=target_ids) self.assertFalse(torch.allclose(adapter_1_before.logits, adapter_1_after.logits)) self.assertFalse(torch.allclose(original_before.logits, adapter_1_after.logits)) self.assertFalse(torch.allclose(default_after.logits, adapter_1_after.logits)) with adapted.disable_adapter(): # Test that the output is the same as the original output. adapter_1_disabled = adapted(input_ids=input_ids, attention_mask=attention_mask, labels=target_ids) assert_close(original_before.logits, adapter_1_disabled.logits, rtol=0, atol=0) # Set adapter back to default. adapted.set_adapter("default") # Test that the output is the same as the default output after training. default_after_set = adapted(input_ids=input_ids, attention_mask=attention_mask, labels=target_ids) assert_close(default_after.logits, default_after_set.logits, rtol=0, atol=0) self.assertFalse(torch.allclose(original_before.logits, default_after_set.logits)) self.assertFalse(torch.allclose(adapter_1_after.logits, default_after_set.logits)) def test_add_and_set_while_disabled(self): """Test that adding and setting adapters while disabled works as intended.""" # Test input data. input_ids = torch.LongTensor([[1, 1, 1], [2, 1, 2]]).to(self.torch_device) target_ids = torch.LongTensor([[0, 0, 0], [0, 0, 0]]).to(self.torch_device) attention_mask = torch.LongTensor([[1, 1, 1], [1, 0, 1]]).to(self.torch_device) # Create original llama model. original = LlamaForCausalLM(self._create_test_llama_config()) original = original.to(self.torch_device) original_before = original(input_ids=input_ids, attention_mask=attention_mask) # Get AdaptionPrompt model. adapted = get_peft_model( original, AdaptionPromptConfig(adapter_layers=2, adapter_len=4, task_type="CAUSAL_LM") ) adapted = adapted.to(self.torch_device) with adapted.disable_adapter(): adapted.add_adapter( "adapter 1", AdaptionPromptConfig(adapter_layers=3, adapter_len=8, task_type="CAUSAL_LM") ) # Test that the output is the same as the original output. adapter_1_before = adapted(input_ids=input_ids, attention_mask=attention_mask, labels=target_ids) assert_close(original_before.logits, adapter_1_before.logits, rtol=0, atol=0) # Single fine-tuning step on adapter 1. optimizer = torch.optim.SGD(adapted.parameters(), lr=1) optimizer.zero_grad() adapter_1_before.loss.backward() optimizer.step() # Test that adapter 1 output changed. adapter_1_after = adapted(input_ids=input_ids, attention_mask=attention_mask, labels=target_ids) self.assertFalse(torch.allclose(original_before.logits, adapter_1_after.logits)) adapted.set_adapter("default") with adapted.disable_adapter(): adapted.set_adapter("adapter 1") # Test that adapter 1 is active again. adapter_1_after_set = adapted(input_ids=input_ids, attention_mask=attention_mask, labels=target_ids) assert_close(adapter_1_after.logits, adapter_1_after_set.logits, rtol=0, atol=0) def test_use_cache(self) -> None: """Test that AdaptionPrompt works when Llama config use_cache=True.""" torch.manual_seed(0) input_ids = torch.LongTensor([[1, 1, 1], [2, 1, 2]]).to(self.torch_device) original = LlamaForCausalLM( LlamaConfig( vocab_size=16, hidden_size=8, intermediate_size=8, num_hidden_layers=8, num_attention_heads=4, use_cache=False, ) ).eval() adapted = get_peft_model( original, AdaptionPromptConfig(adapter_layers=2, adapter_len=4, task_type="CAUSAL_LM") ) adapted = adapted.to(self.torch_device) expected = adapted.generate(input_ids=input_ids, max_length=8) # Set use_cache = True and generate output again. adapted.base_model.config.use_cache = True actual = adapted.generate(input_ids=input_ids, max_length=8) assert_close(expected, actual, rtol=0, atol=0) def test_bf16_inference(self) -> None: """Test that AdaptionPrompt works when Llama using a half-precision model.""" input_ids = torch.LongTensor([[1, 1, 1], [2, 1, 2]]).to(self.torch_device) original = LlamaForCausalLM.from_pretrained( "trl-internal-testing/tiny-random-LlamaForCausalLM", torch_dtype=torch.bfloat16 ) adapted = get_peft_model( original, AdaptionPromptConfig(adapter_layers=2, adapter_len=4, task_type="CAUSAL_LM") ) adapted = adapted.to(self.torch_device) _ = adapted.generate(input_ids=input_ids) @unittest.expectedFailure def test_disable_adapter(self): llama_config = self._create_test_llama_config() model = LlamaForCausalLM(llama_config).to(self.torch_device) dummy_input = torch.LongTensor([[1, 1, 1]]).to(self.torch_device) output_before = model(dummy_input).logits config = AdaptionPromptConfig(adapter_layers=1, adapter_len=4, task_type="CAUSAL_LM") model = get_peft_model(model, config).to(self.torch_device) output_peft = model(dummy_input).logits # TODO currently this fails because scores are zeroed out: # https://github.com/huggingface/peft/blob/062d95a09eb5d1de35c0e5e23d4387daba99e2db/src/peft/tuners/adaption_prompt.py#L303 # This is fine for users but makes it difficult to test if anything happens. In the future, we will have a clean # way to control initialization. Until then, this test is expected to fail. self.assertFalse(torch.allclose(output_before, output_peft)) with model.disable_adapter(): output_peft_disabled = model(dummy_input).logits self.assertTrue(torch.allclose(output_before, output_peft_disabled))

peft/tests/test_adaption_prompt.py/0

# coding=utf-8 # Copyright 2023-present the HuggingFace Inc. team. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from dataclasses import asdict, replace from unittest import TestCase import numpy as np from diffusers import StableDiffusionPipeline from parameterized import parameterized from peft import LoHaConfig, LoraConfig, OFTConfig, get_peft_model from .testing_common import ClassInstantier, PeftCommonTester from .testing_utils import temp_seed PEFT_DIFFUSERS_SD_MODELS_TO_TEST = ["hf-internal-testing/tiny-stable-diffusion-torch"] CONFIG_TESTING_KWARGS = ( { "text_encoder": { "r": 8, "lora_alpha": 32, "target_modules": ["k_proj", "q_proj", "v_proj", "out_proj", "fc1", "fc2"], "lora_dropout": 0.0, "bias": "none", }, "unet": { "r": 8, "lora_alpha": 32, "target_modules": ["proj_in", "proj_out", "to_k", "to_q", "to_v", "to_out.0", "ff.net.0.proj", "ff.net.2"], "lora_dropout": 0.0, "bias": "none", }, }, { "text_encoder": { "r": 8, "alpha": 32, "target_modules": ["k_proj", "q_proj", "v_proj", "out_proj", "fc1", "fc2"], "rank_dropout": 0.0, "module_dropout": 0.0, }, "unet": { "r": 8, "alpha": 32, "target_modules": ["proj_in", "proj_out", "to_k", "to_q", "to_v", "to_out.0", "ff.net.0.proj", "ff.net.2"], "rank_dropout": 0.0, "module_dropout": 0.0, }, }, { "text_encoder": { "r": 8, "target_modules": ["k_proj", "q_proj", "v_proj", "out_proj", "fc1", "fc2"], "module_dropout": 0.0, }, "unet": { "r": 8, "target_modules": ["proj_in", "proj_out", "to_k", "to_q", "to_v", "to_out.0", "ff.net.0.proj", "ff.net.2"], "module_dropout": 0.0, }, }, ) CLASSES_MAPPING = { "lora": (LoraConfig, CONFIG_TESTING_KWARGS[0]), "loha": (LoHaConfig, CONFIG_TESTING_KWARGS[1]), "lokr": (LoHaConfig, CONFIG_TESTING_KWARGS[1]), "oft": (OFTConfig, CONFIG_TESTING_KWARGS[2]), } PeftStableDiffusionTestConfigManager = ClassInstantier(CLASSES_MAPPING) class StableDiffusionModelTester(TestCase, PeftCommonTester): r""" Tests that diffusers StableDiffusion model works with PEFT as expected. """ transformers_class = StableDiffusionPipeline def instantiate_sd_peft(self, model_id, config_cls, config_kwargs): # Instantiate StableDiffusionPipeline model = self.transformers_class.from_pretrained(model_id) config_kwargs = config_kwargs.copy() text_encoder_kwargs = config_kwargs.pop("text_encoder") unet_kwargs = config_kwargs.pop("unet") # the remaining config kwargs should be applied to both configs for key, val in config_kwargs.items(): text_encoder_kwargs[key] = val unet_kwargs[key] = val # Instantiate text_encoder adapter config_text_encoder = config_cls(**text_encoder_kwargs) model.text_encoder = get_peft_model(model.text_encoder, config_text_encoder) # Instantiate unet adapter config_unet = config_cls(**unet_kwargs) model.unet = get_peft_model(model.unet, config_unet) # Move model to device model = model.to(self.torch_device) return model def prepare_inputs_for_testing(self): return { "prompt": "a high quality digital photo of a cute corgi", "num_inference_steps": 20, } @parameterized.expand( PeftStableDiffusionTestConfigManager.get_grid_parameters( { "model_ids": PEFT_DIFFUSERS_SD_MODELS_TO_TEST, "lora_kwargs": {"init_lora_weights": [False]}, "loha_kwargs": {"init_weights": [False]}, "oft_kwargs": {"init_weights": [False]}, }, ) ) def test_merge_layers(self, test_name, model_id, config_cls, config_kwargs): # Instantiate model & adapters model = self.instantiate_sd_peft(model_id, config_cls, config_kwargs) # Generate output for peft modified StableDiffusion dummy_input = self.prepare_inputs_for_testing() with temp_seed(seed=42): peft_output = np.array(model(**dummy_input).images[0]).astype(np.float32) # Merge adapter and model if config_cls not in [LoHaConfig, OFTConfig]: # TODO: Merging the text_encoder is leading to issues on CPU with PyTorch 2.1 model.text_encoder = model.text_encoder.merge_and_unload() model.unet = model.unet.merge_and_unload() # Generate output for peft merged StableDiffusion with temp_seed(seed=42): merged_output = np.array(model(**dummy_input).images[0]).astype(np.float32) # Images are in uint8 drange, so use large atol self.assertTrue(np.allclose(peft_output, merged_output, atol=1.0)) @parameterized.expand( PeftStableDiffusionTestConfigManager.get_grid_parameters( { "model_ids": PEFT_DIFFUSERS_SD_MODELS_TO_TEST, "lora_kwargs": {"init_lora_weights": [False]}, "loha_kwargs": {"init_weights": [False]}, "oft_kwargs": {"init_weights": [False]}, }, ) ) def test_merge_layers_safe_merge(self, test_name, model_id, config_cls, config_kwargs): # Instantiate model & adapters model = self.instantiate_sd_peft(model_id, config_cls, config_kwargs) # Generate output for peft modified StableDiffusion dummy_input = self.prepare_inputs_for_testing() with temp_seed(seed=42): peft_output = np.array(model(**dummy_input).images[0]).astype(np.float32) # Merge adapter and model if config_cls not in [LoHaConfig, OFTConfig]: # TODO: Merging the text_encoder is leading to issues on CPU with PyTorch 2.1 model.text_encoder = model.text_encoder.merge_and_unload(safe_merge=True) model.unet = model.unet.merge_and_unload(safe_merge=True) # Generate output for peft merged StableDiffusion with temp_seed(seed=42): merged_output = np.array(model(**dummy_input).images[0]).astype(np.float32) # Images are in uint8 drange, so use large atol self.assertTrue(np.allclose(peft_output, merged_output, atol=1.0)) @parameterized.expand( PeftStableDiffusionTestConfigManager.get_grid_parameters( { "model_ids": PEFT_DIFFUSERS_SD_MODELS_TO_TEST, "lora_kwargs": {"init_lora_weights": [False]}, }, filter_params_func=lambda tests: [x for x in tests if all(s not in x[0] for s in ["loha", "lokr", "oft"])], ) ) def test_add_weighted_adapter_base_unchanged(self, test_name, model_id, config_cls, config_kwargs): # Instantiate model & adapters model = self.instantiate_sd_peft(model_id, config_cls, config_kwargs) # Get current available adapter config text_encoder_adapter_name = next(iter(model.text_encoder.peft_config.keys())) unet_adapter_name = next(iter(model.unet.peft_config.keys())) text_encoder_adapter_config = replace(model.text_encoder.peft_config[text_encoder_adapter_name]) unet_adapter_config = replace(model.unet.peft_config[unet_adapter_name]) # Create weighted adapters model.text_encoder.add_weighted_adapter([unet_adapter_name], [0.5], "weighted_adapter_test") model.unet.add_weighted_adapter([unet_adapter_name], [0.5], "weighted_adapter_test") # Assert that base adapters config did not change self.assertTrue( asdict(text_encoder_adapter_config) == asdict(model.text_encoder.peft_config[text_encoder_adapter_name]) ) self.assertTrue(asdict(unet_adapter_config) == asdict(model.unet.peft_config[unet_adapter_name])) @parameterized.expand( PeftStableDiffusionTestConfigManager.get_grid_parameters( { "model_ids": PEFT_DIFFUSERS_SD_MODELS_TO_TEST, "lora_kwargs": {"init_lora_weights": [False]}, "loha_kwargs": {"init_weights": [False]}, "lokr_kwargs": {"init_weights": [False]}, "oft_kwargs": {"init_weights": [False]}, }, ) ) def test_disable_adapter(self, test_name, model_id, config_cls, config_kwargs): self._test_disable_adapter(model_id, config_cls, config_kwargs)

peft/tests/test_stablediffusion.py/0

""" Convert weights from https://github.com/google-research/nested-transformer NOTE: You'll need https://github.com/google/CommonLoopUtils, not included in requirements.txt """ import sys import numpy as np import torch from clu import checkpoint arch_depths = { 'nest_base': [2, 2, 20], 'nest_small': [2, 2, 20], 'nest_tiny': [2, 2, 8], } def convert_nest(checkpoint_path, arch): """ Expects path to checkpoint which is a dir containing 4 files like in each of these folders - https://console.cloud.google.com/storage/browser/gresearch/nest-checkpoints `arch` is needed to Returns a state dict that can be used with `torch.nn.Module.load_state_dict` Hint: Follow timm.models.nest.Nest.__init__ and https://github.com/google-research/nested-transformer/blob/main/models/nest_net.py """ assert arch in ['nest_base', 'nest_small', 'nest_tiny'], "Your `arch` is not supported" flax_dict = checkpoint.load_state_dict(checkpoint_path)['optimizer']['target'] state_dict = {} # Patch embedding state_dict['patch_embed.proj.weight'] = torch.tensor( flax_dict['PatchEmbedding_0']['Conv_0']['kernel']).permute(3, 2, 0, 1) state_dict['patch_embed.proj.bias'] = torch.tensor(flax_dict['PatchEmbedding_0']['Conv_0']['bias']) # Positional embeddings posemb_keys = [k for k in flax_dict.keys() if k.startswith('PositionEmbedding')] for i, k in enumerate(posemb_keys): state_dict[f'levels.{i}.pos_embed'] = torch.tensor(flax_dict[k]['pos_embedding']) # Transformer encoders depths = arch_depths[arch] for level in range(len(depths)): for layer in range(depths[level]): global_layer_ix = sum(depths[:level]) + layer # Norms for i in range(2): state_dict[f'levels.{level}.transformer_encoder.{layer}.norm{i+1}.weight'] = torch.tensor( flax_dict[f'EncoderNDBlock_{global_layer_ix}'][f'LayerNorm_{i}']['scale']) state_dict[f'levels.{level}.transformer_encoder.{layer}.norm{i+1}.bias'] = torch.tensor( flax_dict[f'EncoderNDBlock_{global_layer_ix}'][f'LayerNorm_{i}']['bias']) # Attention qkv w_q = flax_dict[f'EncoderNDBlock_{global_layer_ix}']['MultiHeadAttention_0']['DenseGeneral_0']['kernel'] w_kv = flax_dict[f'EncoderNDBlock_{global_layer_ix}']['MultiHeadAttention_0']['DenseGeneral_1']['kernel'] # Pay attention to dims here (maybe get pen and paper) w_kv = np.concatenate(np.split(w_kv, 2, -1), 1) w_qkv = np.concatenate([w_q, w_kv], 1) state_dict[f'levels.{level}.transformer_encoder.{layer}.attn.qkv.weight'] = torch.tensor(w_qkv).flatten(1).permute(1,0) b_q = flax_dict[f'EncoderNDBlock_{global_layer_ix}']['MultiHeadAttention_0']['DenseGeneral_0']['bias'] b_kv = flax_dict[f'EncoderNDBlock_{global_layer_ix}']['MultiHeadAttention_0']['DenseGeneral_1']['bias'] # Pay attention to dims here (maybe get pen and paper) b_kv = np.concatenate(np.split(b_kv, 2, -1), 0) b_qkv = np.concatenate([b_q, b_kv], 0) state_dict[f'levels.{level}.transformer_encoder.{layer}.attn.qkv.bias'] = torch.tensor(b_qkv).reshape(-1) # Attention proj w_proj = flax_dict[f'EncoderNDBlock_{global_layer_ix}']['MultiHeadAttention_0']['proj_kernel'] w_proj = torch.tensor(w_proj).permute(2, 1, 0).flatten(1) state_dict[f'levels.{level}.transformer_encoder.{layer}.attn.proj.weight'] = w_proj state_dict[f'levels.{level}.transformer_encoder.{layer}.attn.proj.bias'] = torch.tensor( flax_dict[f'EncoderNDBlock_{global_layer_ix}']['MultiHeadAttention_0']['bias']) # MLP for i in range(2): state_dict[f'levels.{level}.transformer_encoder.{layer}.mlp.fc{i+1}.weight'] = torch.tensor( flax_dict[f'EncoderNDBlock_{global_layer_ix}']['MlpBlock_0'][f'Dense_{i}']['kernel']).permute(1, 0) state_dict[f'levels.{level}.transformer_encoder.{layer}.mlp.fc{i+1}.bias'] = torch.tensor( flax_dict[f'EncoderNDBlock_{global_layer_ix}']['MlpBlock_0'][f'Dense_{i}']['bias']) # Block aggregations (ConvPool) for level in range(1, len(depths)): # Convs state_dict[f'levels.{level}.pool.conv.weight'] = torch.tensor( flax_dict[f'ConvPool_{level-1}']['Conv_0']['kernel']).permute(3, 2, 0, 1) state_dict[f'levels.{level}.pool.conv.bias'] = torch.tensor( flax_dict[f'ConvPool_{level-1}']['Conv_0']['bias']) # Norms state_dict[f'levels.{level}.pool.norm.weight'] = torch.tensor( flax_dict[f'ConvPool_{level-1}']['LayerNorm_0']['scale']) state_dict[f'levels.{level}.pool.norm.bias'] = torch.tensor( flax_dict[f'ConvPool_{level-1}']['LayerNorm_0']['bias']) # Final norm state_dict[f'norm.weight'] = torch.tensor(flax_dict['LayerNorm_0']['scale']) state_dict[f'norm.bias'] = torch.tensor(flax_dict['LayerNorm_0']['bias']) # Classifier state_dict['head.weight'] = torch.tensor(flax_dict['Dense_0']['kernel']).permute(1, 0) state_dict['head.bias'] = torch.tensor(flax_dict['Dense_0']['bias']) return state_dict if __name__ == '__main__': variant = sys.argv[1] # base, small, or tiny state_dict = convert_nest(f'./nest-{variant[0]}_imagenet', f'nest_{variant}') torch.save(state_dict, f'./jx_nest_{variant}.pth')

pytorch-image-models/convert/convert_nest_flax.py/0

# CSP-ResNeXt **CSPResNeXt** is a convolutional neural network where we apply the Cross Stage Partial Network (CSPNet) approach to [ResNeXt](https://paperswithcode.com/method/resnext). The CSPNet partitions the feature map of the base layer into two parts and then merges them through a cross-stage hierarchy. The use of a split and merge strategy allows for more gradient flow through the network. {% include 'code_snippets.md' %} ## How do I train this model? You can follow the [timm recipe scripts](https://rwightman.github.io/pytorch-image-models/scripts/) for training a new model afresh. ## Citation ```BibTeX @misc{wang2019cspnet, title={CSPNet: A New Backbone that can Enhance Learning Capability of CNN}, author={Chien-Yao Wang and Hong-Yuan Mark Liao and I-Hau Yeh and Yueh-Hua Wu and Ping-Yang Chen and Jun-Wei Hsieh}, year={2019}, eprint={1911.11929}, archivePrefix={arXiv}, primaryClass={cs.CV} } ``` <!-- Type: model-index Collections: - Name: CSP ResNeXt Paper: Title: 'CSPNet: A New Backbone that can Enhance Learning Capability of CNN' URL: https://paperswithcode.com/paper/cspnet-a-new-backbone-that-can-enhance Models: - Name: cspresnext50 In Collection: CSP ResNeXt Metadata: FLOPs: 3962945536 Parameters: 20570000 File Size: 82562887 Architecture: - 1x1 Convolution - Batch Normalization - Convolution - Global Average Pooling - Grouped Convolution - Max Pooling - ReLU - ResNeXt Block - Residual Connection - Softmax Tasks: - Image Classification Training Techniques: - Label Smoothing - Polynomial Learning Rate Decay - SGD with Momentum - Weight Decay Training Data: - ImageNet Training Resources: 1x GPU ID: cspresnext50 LR: 0.1 Layers: 50 Crop Pct: '0.875' Momentum: 0.9 Batch Size: 128 Image Size: '224' Weight Decay: 0.005 Interpolation: bilinear Training Steps: 8000000 Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/cspnet.py#L430 Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/cspresnext50_ra_224-648b4713.pth Results: - Task: Image Classification Dataset: ImageNet Metrics: Top 1 Accuracy: 80.05% Top 5 Accuracy: 94.94% -->

pytorch-image-models/docs/models/.templates/models/csp-resnext.md/0

# HRNet **HRNet**, or **High-Resolution Net**, is a general purpose convolutional neural network for tasks like semantic segmentation, object detection and image classification. It is able to maintain high resolution representations through the whole process. We start from a high-resolution convolution stream, gradually add high-to-low resolution convolution streams one by one, and connect the multi-resolution streams in parallel. The resulting network consists of several ($4$ in the paper) stages and the $n$th stage contains $n$ streams corresponding to $n$ resolutions. The authors conduct repeated multi-resolution fusions by exchanging the information across the parallel streams over and over. {% include 'code_snippets.md' %} ## How do I train this model? You can follow the [timm recipe scripts](https://rwightman.github.io/pytorch-image-models/scripts/) for training a new model afresh. ## Citation ```BibTeX @misc{sun2019highresolution, title={High-Resolution Representations for Labeling Pixels and Regions}, author={Ke Sun and Yang Zhao and Borui Jiang and Tianheng Cheng and Bin Xiao and Dong Liu and Yadong Mu and Xinggang Wang and Wenyu Liu and Jingdong Wang}, year={2019}, eprint={1904.04514}, archivePrefix={arXiv}, primaryClass={cs.CV} } ``` <!-- Type: model-index Collections: - Name: HRNet Paper: Title: Deep High-Resolution Representation Learning for Visual Recognition URL: https://paperswithcode.com/paper/190807919 Models: - Name: hrnet_w18 In Collection: HRNet Metadata: FLOPs: 5547205500 Parameters: 21300000 File Size: 85718883 Architecture: - Batch Normalization - Convolution - ReLU - Residual Connection Tasks: - Image Classification Training Techniques: - Nesterov Accelerated Gradient - Weight Decay Training Data: - ImageNet Training Resources: 4x NVIDIA V100 GPUs ID: hrnet_w18 Epochs: 100 Layers: 18 Crop Pct: '0.875' Momentum: 0.9 Batch Size: 256 Image Size: '224' Weight Decay: 0.001 Interpolation: bilinear Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/hrnet.py#L800 Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-hrnet/hrnetv2_w18-8cb57bb9.pth Results: - Task: Image Classification Dataset: ImageNet Metrics: Top 1 Accuracy: 76.76% Top 5 Accuracy: 93.44% - Name: hrnet_w18_small In Collection: HRNet Metadata: FLOPs: 2071651488 Parameters: 13190000 File Size: 52934302 Architecture: - Batch Normalization - Convolution - ReLU - Residual Connection Tasks: - Image Classification Training Techniques: - Nesterov Accelerated Gradient - Weight Decay Training Data: - ImageNet Training Resources: 4x NVIDIA V100 GPUs ID: hrnet_w18_small Epochs: 100 Layers: 18 Crop Pct: '0.875' Momentum: 0.9 Batch Size: 256 Image Size: '224' Weight Decay: 0.001 Interpolation: bilinear Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/hrnet.py#L790 Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-hrnet/hrnet_w18_small_v1-f460c6bc.pth Results: - Task: Image Classification Dataset: ImageNet Metrics: Top 1 Accuracy: 72.34% Top 5 Accuracy: 90.68% - Name: hrnet_w18_small_v2 In Collection: HRNet Metadata: FLOPs: 3360023160 Parameters: 15600000 File Size: 62682879 Architecture: - Batch Normalization - Convolution - ReLU - Residual Connection Tasks: - Image Classification Training Techniques: - Nesterov Accelerated Gradient - Weight Decay Training Data: - ImageNet Training Resources: 4x NVIDIA V100 GPUs ID: hrnet_w18_small_v2 Epochs: 100 Layers: 18 Crop Pct: '0.875' Momentum: 0.9 Batch Size: 256 Image Size: '224' Weight Decay: 0.001 Interpolation: bilinear Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/hrnet.py#L795 Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-hrnet/hrnet_w18_small_v2-4c50a8cb.pth Results: - Task: Image Classification Dataset: ImageNet Metrics: Top 1 Accuracy: 75.11% Top 5 Accuracy: 92.41% - Name: hrnet_w30 In Collection: HRNet Metadata: FLOPs: 10474119492 Parameters: 37710000 File Size: 151452218 Architecture: - Batch Normalization - Convolution - ReLU - Residual Connection Tasks: - Image Classification Training Techniques: - Nesterov Accelerated Gradient - Weight Decay Training Data: - ImageNet Training Resources: 4x NVIDIA V100 GPUs ID: hrnet_w30 Epochs: 100 Layers: 30 Crop Pct: '0.875' Momentum: 0.9 Batch Size: 256 Image Size: '224' Weight Decay: 0.001 Interpolation: bilinear Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/hrnet.py#L805 Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-hrnet/hrnetv2_w30-8d7f8dab.pth Results: - Task: Image Classification Dataset: ImageNet Metrics: Top 1 Accuracy: 78.21% Top 5 Accuracy: 94.22% - Name: hrnet_w32 In Collection: HRNet Metadata: FLOPs: 11524528320 Parameters: 41230000 File Size: 165547812 Architecture: - Batch Normalization - Convolution - ReLU - Residual Connection Tasks: - Image Classification Training Techniques: - Nesterov Accelerated Gradient - Weight Decay Training Data: - ImageNet Training Resources: 4x NVIDIA V100 GPUs Training Time: 60 hours ID: hrnet_w32 Epochs: 100 Layers: 32 Crop Pct: '0.875' Momentum: 0.9 Batch Size: 256 Image Size: '224' Weight Decay: 0.001 Interpolation: bilinear Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/hrnet.py#L810 Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-hrnet/hrnetv2_w32-90d8c5fb.pth Results: - Task: Image Classification Dataset: ImageNet Metrics: Top 1 Accuracy: 78.45% Top 5 Accuracy: 94.19% - Name: hrnet_w40 In Collection: HRNet Metadata: FLOPs: 16381182192 Parameters: 57560000 File Size: 230899236 Architecture: - Batch Normalization - Convolution - ReLU - Residual Connection Tasks: - Image Classification Training Techniques: - Nesterov Accelerated Gradient - Weight Decay Training Data: - ImageNet Training Resources: 4x NVIDIA V100 GPUs ID: hrnet_w40 Epochs: 100 Layers: 40 Crop Pct: '0.875' Momentum: 0.9 Batch Size: 256 Image Size: '224' Weight Decay: 0.001 Interpolation: bilinear Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/hrnet.py#L815 Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-hrnet/hrnetv2_w40-7cd397a4.pth Results: - Task: Image Classification Dataset: ImageNet Metrics: Top 1 Accuracy: 78.93% Top 5 Accuracy: 94.48% - Name: hrnet_w44 In Collection: HRNet Metadata: FLOPs: 19202520264 Parameters: 67060000 File Size: 268957432 Architecture: - Batch Normalization - Convolution - ReLU - Residual Connection Tasks: - Image Classification Training Techniques: - Nesterov Accelerated Gradient - Weight Decay Training Data: - ImageNet Training Resources: 4x NVIDIA V100 GPUs ID: hrnet_w44 Epochs: 100 Layers: 44 Crop Pct: '0.875' Momentum: 0.9 Batch Size: 256 Image Size: '224' Weight Decay: 0.001 Interpolation: bilinear Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/hrnet.py#L820 Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-hrnet/hrnetv2_w44-c9ac8c18.pth Results: - Task: Image Classification Dataset: ImageNet Metrics: Top 1 Accuracy: 78.89% Top 5 Accuracy: 94.37% - Name: hrnet_w48 In Collection: HRNet Metadata: FLOPs: 22285865760 Parameters: 77470000 File Size: 310603710 Architecture: - Batch Normalization - Convolution - ReLU - Residual Connection Tasks: - Image Classification Training Techniques: - Nesterov Accelerated Gradient - Weight Decay Training Data: - ImageNet Training Resources: 4x NVIDIA V100 GPUs Training Time: 80 hours ID: hrnet_w48 Epochs: 100 Layers: 48 Crop Pct: '0.875' Momentum: 0.9 Batch Size: 256 Image Size: '224' Weight Decay: 0.001 Interpolation: bilinear Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/hrnet.py#L825 Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-hrnet/hrnetv2_w48-abd2e6ab.pth Results: - Task: Image Classification Dataset: ImageNet Metrics: Top 1 Accuracy: 79.32% Top 5 Accuracy: 94.51% - Name: hrnet_w64 In Collection: HRNet Metadata: FLOPs: 37239321984 Parameters: 128060000 File Size: 513071818 Architecture: - Batch Normalization - Convolution - ReLU - Residual Connection Tasks: - Image Classification Training Techniques: - Nesterov Accelerated Gradient - Weight Decay Training Data: - ImageNet Training Resources: 4x NVIDIA V100 GPUs ID: hrnet_w64 Epochs: 100 Layers: 64 Crop Pct: '0.875' Momentum: 0.9 Batch Size: 256 Image Size: '224' Weight Decay: 0.001 Interpolation: bilinear Code: https://github.com/rwightman/pytorch-image-models/blob/d8e69206be253892b2956341fea09fdebfaae4e3/timm/models/hrnet.py#L830 Weights: https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-hrnet/hrnetv2_w64-b47cc881.pth Results: - Task: Image Classification Dataset: ImageNet Metrics: Top 1 Accuracy: 79.46% Top 5 Accuracy: 94.65% -->

pytorch-image-models/docs/models/.templates/models/hrnet.md/0

# SWSL ResNet **Residual Networks**, or **ResNets**, learn residual functions with reference to the layer inputs, instead of learning unreferenced functions. Instead of hoping each few stacked layers directly fit a desired underlying mapping, residual nets let these layers fit a residual mapping. They stack [residual blocks](https://paperswithcode.com/method/residual-block) ontop of each other to form network: e.g. a ResNet-50 has fifty layers using these blocks. The models in this collection utilise semi-weakly supervised learning to improve the performance of the model. The approach brings important gains to standard architectures for image, video and fine-grained classification. Please note the CC-BY-NC 4.0 license on theses weights, non-commercial use only. {% include 'code_snippets.md' %} ## How do I train this model? You can follow the [timm recipe scripts](https://rwightman.github.io/pytorch-image-models/scripts/) for training a new model afresh. ## Citation ```BibTeX @article{DBLP:journals/corr/abs-1905-00546, author = {I. Zeki Yalniz and Herv{\'{e}} J{\'{e}}gou and Kan Chen and Manohar Paluri and Dhruv Mahajan}, title = {Billion-scale semi-supervised learning for image classification}, journal = {CoRR}, volume = {abs/1905.00546}, year = {2019}, url = {http://arxiv.org/abs/1905.00546}, archivePrefix = {arXiv}, eprint = {1905.00546}, timestamp = {Mon, 28 Sep 2020 08:19:37 +0200}, biburl = {https://dblp.org/rec/journals/corr/abs-1905-00546.bib}, bibsource = {dblp computer science bibliography, https://dblp.org} } ``` <!-- Type: model-index Collections: - Name: SWSL ResNet Paper: Title: Billion-scale semi-supervised learning for image classification URL: https://paperswithcode.com/paper/billion-scale-semi-supervised-learning-for Models: - Name: swsl_resnet18 In Collection: SWSL ResNet Metadata: FLOPs: 2337073152 Parameters: 11690000 File Size: 46811375 Architecture: - 1x1 Convolution - Batch Normalization - Bottleneck Residual Block - Convolution - Global Average Pooling - Max Pooling - ReLU - Residual Block - Residual Connection - Softmax Tasks: - Image Classification Training Techniques: - SGD with Momentum - Weight Decay Training Data: - IG-1B-Targeted - ImageNet Training Resources: 64x GPUs ID: swsl_resnet18 LR: 0.0015 Epochs: 30 Layers: 18 Crop Pct: '0.875' Batch Size: 1536 Image Size: '224' Weight Decay: 0.0001 Interpolation: bilinear Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/resnet.py#L954 Weights: https://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_weakly_supervised_resnet18-118f1556.pth Results: - Task: Image Classification Dataset: ImageNet Metrics: Top 1 Accuracy: 73.28% Top 5 Accuracy: 91.76% - Name: swsl_resnet50 In Collection: SWSL ResNet Metadata: FLOPs: 5282531328 Parameters: 25560000 File Size: 102480594 Architecture: - 1x1 Convolution - Batch Normalization - Bottleneck Residual Block - Convolution - Global Average Pooling - Max Pooling - ReLU - Residual Block - Residual Connection - Softmax Tasks: - Image Classification Training Techniques: - SGD with Momentum - Weight Decay Training Data: - IG-1B-Targeted - ImageNet Training Resources: 64x GPUs ID: swsl_resnet50 LR: 0.0015 Epochs: 30 Layers: 50 Crop Pct: '0.875' Batch Size: 1536 Image Size: '224' Weight Decay: 0.0001 Interpolation: bilinear Code: https://github.com/rwightman/pytorch-image-models/blob/9a25fdf3ad0414b4d66da443fe60ae0aa14edc84/timm/models/resnet.py#L965 Weights: https://dl.fbaipublicfiles.com/semiweaksupervision/model_files/semi_weakly_supervised_resnet50-16a12f1b.pth Results: - Task: Image Classification Dataset: ImageNet Metrics: Top 1 Accuracy: 81.14% Top 5 Accuracy: 95.97% -->

pytorch-image-models/docs/models/.templates/models/swsl-resnet.md/0

# Hugging Face Timm Docs ## Getting Started ``` pip install git+https://github.com/huggingface/doc-builder.git@main#egg=hf-doc-builder pip install watchdog black ``` ## Preview the Docs Locally ``` doc-builder preview timm hfdocs/source ```

pytorch-image-models/hfdocs/README.md/0

Import: - ./docs/models/*.md Library: Name: PyTorch Image Models Headline: PyTorch image models, scripts, pretrained weights Website: https://rwightman.github.io/pytorch-image-models/ Repository: https://github.com/rwightman/pytorch-image-models Docs: https://rwightman.github.io/pytorch-image-models/ README: "# PyTorch Image Models\r\n\r\nPyTorch Image Models (TIMM) is a library\ \ for state-of-the-art image classification. With this library you can:\r\n\r\n\ - Choose from 300+ pre-trained state-of-the-art image classification models.\r\ \n- Train models afresh on research datasets such as ImageNet using provided scripts.\r\ \n- Finetune pre-trained models on your own datasets, including the latest cutting\ \ edge models."

pytorch-image-models/model-index.yml/0

import math import torch from torch.utils.data import Sampler import torch.distributed as dist class OrderedDistributedSampler(Sampler): """Sampler that restricts data loading to a subset of the dataset. It is especially useful in conjunction with :class:`torch.nn.parallel.DistributedDataParallel`. In such case, each process can pass a DistributedSampler instance as a DataLoader sampler, and load a subset of the original dataset that is exclusive to it. .. note:: Dataset is assumed to be of constant size. Arguments: dataset: Dataset used for sampling. num_replicas (optional): Number of processes participating in distributed training. rank (optional): Rank of the current process within num_replicas. """ def __init__(self, dataset, num_replicas=None, rank=None): if num_replicas is None: if not dist.is_available(): raise RuntimeError("Requires distributed package to be available") num_replicas = dist.get_world_size() if rank is None: if not dist.is_available(): raise RuntimeError("Requires distributed package to be available") rank = dist.get_rank() self.dataset = dataset self.num_replicas = num_replicas self.rank = rank self.num_samples = int(math.ceil(len(self.dataset) * 1.0 / self.num_replicas)) self.total_size = self.num_samples * self.num_replicas def __iter__(self): indices = list(range(len(self.dataset))) # add extra samples to make it evenly divisible indices += indices[:(self.total_size - len(indices))] assert len(indices) == self.total_size # subsample indices = indices[self.rank:self.total_size:self.num_replicas] assert len(indices) == self.num_samples return iter(indices) def __len__(self): return self.num_samples class RepeatAugSampler(Sampler): """Sampler that restricts data loading to a subset of the dataset for distributed, with repeated augmentation. It ensures that different each augmented version of a sample will be visible to a different process (GPU). Heavily based on torch.utils.data.DistributedSampler This sampler was taken from https://github.com/facebookresearch/deit/blob/0c4b8f60/samplers.py Used in Copyright (c) 2015-present, Facebook, Inc. """ def __init__( self, dataset, num_replicas=None, rank=None, shuffle=True, num_repeats=3, selected_round=256, selected_ratio=0, ): if num_replicas is None: if not dist.is_available(): raise RuntimeError("Requires distributed package to be available") num_replicas = dist.get_world_size() if rank is None: if not dist.is_available(): raise RuntimeError("Requires distributed package to be available") rank = dist.get_rank() self.dataset = dataset self.num_replicas = num_replicas self.rank = rank self.shuffle = shuffle self.num_repeats = num_repeats self.epoch = 0 self.num_samples = int(math.ceil(len(self.dataset) * num_repeats / self.num_replicas)) self.total_size = self.num_samples * self.num_replicas # Determine the number of samples to select per epoch for each rank. # num_selected logic defaults to be the same as original RASampler impl, but this one can be tweaked # via selected_ratio and selected_round args. selected_ratio = selected_ratio or num_replicas # ratio to reduce selected samples by, num_replicas if 0 if selected_round: self.num_selected_samples = int(math.floor( len(self.dataset) // selected_round * selected_round / selected_ratio)) else: self.num_selected_samples = int(math.ceil(len(self.dataset) / selected_ratio)) def __iter__(self): # deterministically shuffle based on epoch g = torch.Generator() g.manual_seed(self.epoch) if self.shuffle: indices = torch.randperm(len(self.dataset), generator=g) else: indices = torch.arange(start=0, end=len(self.dataset)) # produce repeats e.g. [0, 0, 0, 1, 1, 1, 2, 2, 2....] if isinstance(self.num_repeats, float) and not self.num_repeats.is_integer(): # resample for repeats w/ non-integer ratio repeat_size = math.ceil(self.num_repeats * len(self.dataset)) indices = indices[torch.tensor([int(i // self.num_repeats) for i in range(repeat_size)])] else: indices = torch.repeat_interleave(indices, repeats=int(self.num_repeats), dim=0) indices = indices.tolist() # leaving as tensor thrashes dataloader memory # add extra samples to make it evenly divisible padding_size = self.total_size - len(indices) if padding_size > 0: indices += indices[:padding_size] assert len(indices) == self.total_size # subsample per rank indices = indices[self.rank:self.total_size:self.num_replicas] assert len(indices) == self.num_samples # return up to num selected samples return iter(indices[:self.num_selected_samples]) def __len__(self): return self.num_selected_samples def set_epoch(self, epoch): self.epoch = epoch

pytorch-image-models/timm/data/distributed_sampler.py/0

""" Dataset reader for webdataset Hacked together by / Copyright 2022 Ross Wightman """ import io import json import logging import math import os import random import sys from dataclasses import dataclass from functools import partial from itertools import islice from typing import Any, Callable, Dict, List, Optional, Tuple import torch import torch.distributed as dist import yaml from PIL import Image from torch.utils.data import Dataset, IterableDataset, get_worker_info try: import webdataset as wds from webdataset.filters import _shuffle, getfirst from webdataset.shardlists import expand_urls from webdataset.tariterators import base_plus_ext, url_opener, tar_file_expander, valid_sample except ImportError: wds = None expand_urls = None from .class_map import load_class_map from .reader import Reader from .shared_count import SharedCount _logger = logging.getLogger(__name__) SAMPLE_SHUFFLE_SIZE = int(os.environ.get('WDS_SHUFFLE_SIZE', 8192)) SAMPLE_INITIAL_SIZE = int(os.environ.get('WDS_INITIAL_SIZE', 2048)) def _load_info(root, names=('_info.json', 'info.json')): if isinstance(names, str): names = (names,) tried = [] err_str = '' for n in names: full_path = os.path.join(root, n) try: tried.append(full_path) with wds.gopen(full_path) as f: if n.endswith('.json'): info_dict = json.load(f) else: info_dict = yaml.safe_load(f) return info_dict except Exception as e: err_str = str(e) _logger.warning( f'Dataset info file not found at {tried}. Error: {err_str}. ' 'Falling back to provided split and size arg.') return {} @dataclass class SplitInfo: num_samples: int filenames: Tuple[str] shard_lengths: Tuple[int] = () alt_label: str = '' name: str = '' def _parse_split_info(split: str, info: Dict): def _info_convert(dict_info): return SplitInfo( num_samples=dict_info['num_samples'], filenames=tuple(dict_info['filenames']), shard_lengths=tuple(dict_info['shard_lengths']), alt_label=dict_info.get('alt_label', ''), name=dict_info['name'], ) if 'tar' in split or '..' in split: # split in WDS string braceexpand format, sample count can be included with a | separator # ex: `dataset-split-{0000..9999}.tar|100000` for 9999 shards, covering 100,000 samples split = split.split('|') num_samples = 0 split_name = '' if len(split) > 1: num_samples = int(split[1]) split = split[0] if '::' not in split: split_parts = split.split('-', 3) split_idx = len(split_parts) - 1 if split_idx and 'splits' in info and split_parts[split_idx] in info['splits']: split_name = split_parts[split_idx] split_filenames = expand_urls(split) if split_name: split_info = info['splits'][split_name] if not num_samples: _fc = {f: c for f, c in zip(split_info['filenames'], split_info['shard_lengths'])} num_samples = sum(_fc[f] for f in split_filenames) split_info['filenames'] = tuple(_fc.keys()) split_info['shard_lengths'] = tuple(_fc.values()) split_info['num_samples'] = num_samples split_info = _info_convert(split_info) else: split_info = SplitInfo( name=split_name, num_samples=num_samples, filenames=split_filenames, ) else: if 'splits' not in info or split not in info['splits']: raise RuntimeError(f"split {split} not found in info ({info.get('splits', {}).keys()})") split = split split_info = info['splits'][split] split_info = _info_convert(split_info) return split_info def log_and_continue(exn): """Call in an exception handler to ignore exceptions, isssue a warning, and continue.""" _logger.warning(f'Handling webdataset error ({repr(exn)}). Ignoring.') # NOTE: try force an exit on errors that are clearly code / config and not transient if isinstance(exn, TypeError): raise exn return True def _decode( sample, image_key='jpg', image_mode='RGB', target_key='cls', alt_label='' ): """ Custom sample decode * decode and convert PIL Image * cls byte string label to int * pass through JSON byte string (if it exists) without parse """ # decode class label, skip if alternate label not valid if alt_label: # alternative labels are encoded in json metadata meta = json.loads(sample['json']) class_label = int(meta[alt_label]) if class_label < 0: # skipped labels currently encoded as -1, may change to a null/None value return None else: class_label = int(sample[target_key]) # decode image img = getfirst(sample, image_key) with io.BytesIO(img) as b: img = Image.open(b) img.load() if image_mode: img = img.convert(image_mode) # json passed through in undecoded state decoded = dict(jpg=img, cls=class_label, json=sample.get('json', None)) return decoded def pytorch_worker_seed(): """get dataloader worker seed from pytorch""" worker_info = get_worker_info() if worker_info is not None: # favour the seed already created for pytorch dataloader workers if it exists return worker_info.seed # fallback to wds rank based seed return wds.utils.pytorch_worker_seed() if wds is not None: # conditional to avoid mandatory wds import (via inheritance of wds.PipelineStage) class detshuffle2(wds.PipelineStage): def __init__( self, bufsize=1000, initial=100, seed=0, epoch=-1, ): self.bufsize = bufsize self.initial = initial self.seed = seed self.epoch = epoch def run(self, src): if isinstance(self.epoch, SharedCount): epoch = self.epoch.value else: # NOTE: this is epoch tracking is problematic in a multiprocess (dataloader workers or train) # situation as different workers may wrap at different times (or not at all). self.epoch += 1 epoch = self.epoch if self.seed < 0: seed = pytorch_worker_seed() + epoch else: seed = self.seed + epoch # _logger.info(f'shuffle seed: {self.seed}, {seed}, epoch: {epoch}') # FIXME temporary rng = random.Random(seed) return _shuffle(src, self.bufsize, self.initial, rng) else: detshuffle2 = None class ResampledShards2(IterableDataset): """An iterable dataset yielding a list of urls.""" def __init__( self, urls, nshards=sys.maxsize, worker_seed=None, deterministic=True, epoch=-1, ): """Sample shards from the shard list with replacement. :param urls: a list of URLs as a Python list or brace notation string """ super().__init__() urls = wds.shardlists.expand_urls(urls) self.urls = urls assert isinstance(self.urls[0], str) self.nshards = nshards self.rng = random.Random() self.worker_seed = pytorch_worker_seed if worker_seed is None else worker_seed self.deterministic = deterministic self.epoch = epoch def __iter__(self): """Return an iterator over the shards.""" if isinstance(self.epoch, SharedCount): epoch = self.epoch.value else: # NOTE: this is epoch tracking is problematic in a multiprocess (dataloader workers or train) # situation as different workers may wrap at different times (or not at all). self.epoch += 1 epoch = self.epoch if self.deterministic: # reset seed w/ epoch if deterministic, worker seed should be deterministic due to arg.seed self.rng = random.Random(self.worker_seed() + epoch) for _ in range(self.nshards): index = self.rng.randint(0, len(self.urls) - 1) yield dict(url=self.urls[index]) class ReaderWds(Reader): def __init__( self, root: str, name: Optional[str] = None, split: str = 'train', is_training: bool = False, num_samples: Optional[int] = None, batch_size: int = 1, repeats: int = 0, seed: int = 42, class_map: Optional[dict] = None, input_key: str = 'jpg;png;webp', input_img_mode: str = 'RGB', target_key: str = 'cls', target_img_mode: str = '', filename_key: str = 'filename', sample_shuffle_size: Optional[int] = None, smaple_initial_size: Optional[int] = None, ): super().__init__() if wds is None: raise RuntimeError( 'Please install webdataset 0.2.x package `pip install git+https://github.com/webdataset/webdataset`.') self.root = root self.is_training = is_training self.batch_size = batch_size self.repeats = repeats self.common_seed = seed # a seed that's fixed across all worker / distributed instances self.shard_shuffle_size = 500 self.sample_shuffle_size = sample_shuffle_size or SAMPLE_SHUFFLE_SIZE self.sample_initial_size = smaple_initial_size or SAMPLE_INITIAL_SIZE self.input_key = input_key self.input_img_mode = input_img_mode self.target_key = target_key self.filename_key = filename_key self.key_ext = '.JPEG' # extension to add to key for original filenames (DS specific, default ImageNet) self.info = _load_info(self.root) self.split_info = _parse_split_info(split, self.info) if num_samples is not None: self.num_samples = num_samples else: self.num_samples = self.split_info.num_samples if not self.num_samples: raise RuntimeError(f'Invalid split definition, num_samples not specified.') self.remap_class = False if class_map: self.class_to_idx = load_class_map(class_map) self.remap_class = True else: self.class_to_idx = {} # Distributed world state self.dist_rank = 0 self.dist_num_replicas = 1 if dist.is_available() and dist.is_initialized() and dist.get_world_size() > 1: self.dist_rank = dist.get_rank() self.dist_num_replicas = dist.get_world_size() # Attributes that are updated in _lazy_init self.worker_info = None self.worker_id = 0 self.worker_seed = seed # seed unique to each worker instance self.num_workers = 1 self.global_worker_id = 0 self.global_num_workers = 1 self.init_count = 0 self.epoch_count = SharedCount() # DataPipeline is lazy init, the majority of WDS DataPipeline could be init here, BUT, shuffle seed # is not handled in manner where it can be deterministic for each worker AND initialized up front self.ds = None def set_epoch(self, count): self.epoch_count.value = count def set_loader_cfg( self, num_workers: Optional[int] = None, ): if self.ds is not None: return if num_workers is not None: self.num_workers = num_workers self.global_num_workers = self.dist_num_replicas * self.num_workers def _lazy_init(self): """ Lazily initialize worker (in worker processes) """ if self.worker_info is None: worker_info = torch.utils.data.get_worker_info() if worker_info is not None: self.worker_info = worker_info self.worker_id = worker_info.id self.worker_seed = worker_info.seed self.num_workers = worker_info.num_workers self.global_num_workers = self.dist_num_replicas * self.num_workers self.global_worker_id = self.dist_rank * self.num_workers + self.worker_id # init data pipeline abs_shard_filenames = [os.path.join(self.root, f) for f in self.split_info.filenames] pipeline = [wds.SimpleShardList(abs_shard_filenames)] # at this point we have an iterator over all the shards if self.is_training: pipeline.extend([ detshuffle2( self.shard_shuffle_size, seed=self.common_seed, epoch=self.epoch_count, ), self._split_by_node_and_worker, # at this point, we have an iterator over the shards assigned to each worker wds.tarfile_to_samples(handler=log_and_continue), wds.shuffle( bufsize=self.sample_shuffle_size, initial=self.sample_initial_size, rng=random.Random(self.worker_seed) # this is why we lazy-init whole DataPipeline ), ]) else: pipeline.extend([ self._split_by_node_and_worker, # at this point, we have an iterator over the shards assigned to each worker wds.tarfile_to_samples(handler=log_and_continue), ]) pipeline.extend([ wds.map( partial( _decode, image_key=self.input_key, image_mode=self.input_img_mode, alt_label=self.split_info.alt_label, ), handler=log_and_continue, ), wds.rename(image=self.input_key, target=self.target_key) ]) self.ds = wds.DataPipeline(*pipeline) def _split_by_node_and_worker(self, src): if self.global_num_workers > 1: for s in islice(src, self.global_worker_id, None, self.global_num_workers): yield s else: for s in src: yield s def _num_samples_per_worker(self): num_worker_samples = self.num_samples / max(self.global_num_workers, self.dist_num_replicas) if self.is_training or self.dist_num_replicas > 1: num_worker_samples = math.ceil(num_worker_samples) if self.is_training: num_worker_samples = math.ceil(num_worker_samples / self.batch_size) * self.batch_size return int(num_worker_samples) def __iter__(self): if self.ds is None: self._lazy_init() num_worker_samples = self._num_samples_per_worker() if self.is_training or self.dist_num_replicas > 1: # NOTE: doing distributed validation w/ WDS is messy, hard to meet constraints that # same # of batches needed across all replicas w/ seeing each sample once. # with_epoch() is simple but could miss a shard's worth of samples in some workers, # and duplicate in others. Best to keep num DL workers low and a divisor of #val shards. ds = self.ds.with_epoch(num_worker_samples) else: ds = self.ds i = 0 # _logger.info(f'start {i}, {self.worker_id}') # FIXME temporary debug for sample in ds: target = sample['target'] if self.remap_class: target = self.class_to_idx[target] yield sample['image'], target i += 1 # _logger.info(f'end {i}, {self.worker_id}') # FIXME temporary debug def __len__(self): num_samples = self._num_samples_per_worker() * self.num_workers return num_samples def _filename(self, index, basename=False, absolute=False): assert False, "Not supported" # no random access to examples def filenames(self, basename=False, absolute=False): """ Return all filenames in dataset, overrides base""" if self.ds is None: self._lazy_init() names = [] for sample in self.ds: if self.filename_key in sample: name = sample[self.filename_key] elif '__key__' in sample: name = sample['__key__'] + self.key_ext else: assert False, "No supported name field present" names.append(name) if len(names) >= self.num_samples: break # safety for ds.repeat() case return names

pytorch-image-models/timm/data/readers/reader_wds.py/0

""" Classifier head and layer factory Hacked together by / Copyright 2020 Ross Wightman """ from collections import OrderedDict from functools import partial from typing import Optional, Union, Callable import torch import torch.nn as nn from torch.nn import functional as F from .adaptive_avgmax_pool import SelectAdaptivePool2d from .create_act import get_act_layer from .create_norm import get_norm_layer def _create_pool( num_features: int, num_classes: int, pool_type: str = 'avg', use_conv: bool = False, input_fmt: Optional[str] = None, ): flatten_in_pool = not use_conv # flatten when we use a Linear layer after pooling if not pool_type: assert num_classes == 0 or use_conv,\ 'Pooling can only be disabled if classifier is also removed or conv classifier is used' flatten_in_pool = False # disable flattening if pooling is pass-through (no pooling) global_pool = SelectAdaptivePool2d( pool_type=pool_type, flatten=flatten_in_pool, input_fmt=input_fmt, ) num_pooled_features = num_features * global_pool.feat_mult() return global_pool, num_pooled_features def _create_fc(num_features, num_classes, use_conv=False): if num_classes <= 0: fc = nn.Identity() # pass-through (no classifier) elif use_conv: fc = nn.Conv2d(num_features, num_classes, 1, bias=True) else: fc = nn.Linear(num_features, num_classes, bias=True) return fc def create_classifier( num_features: int, num_classes: int, pool_type: str = 'avg', use_conv: bool = False, input_fmt: str = 'NCHW', drop_rate: Optional[float] = None, ): global_pool, num_pooled_features = _create_pool( num_features, num_classes, pool_type, use_conv=use_conv, input_fmt=input_fmt, ) fc = _create_fc( num_pooled_features, num_classes, use_conv=use_conv, ) if drop_rate is not None: dropout = nn.Dropout(drop_rate) return global_pool, dropout, fc return global_pool, fc class ClassifierHead(nn.Module): """Classifier head w/ configurable global pooling and dropout.""" def __init__( self, in_features: int, num_classes: int, pool_type: str = 'avg', drop_rate: float = 0., use_conv: bool = False, input_fmt: str = 'NCHW', ): """ Args: in_features: The number of input features. num_classes: The number of classes for the final classifier layer (output). pool_type: Global pooling type, pooling disabled if empty string (''). drop_rate: Pre-classifier dropout rate. """ super(ClassifierHead, self).__init__() self.in_features = in_features self.use_conv = use_conv self.input_fmt = input_fmt global_pool, fc = create_classifier( in_features, num_classes, pool_type, use_conv=use_conv, input_fmt=input_fmt, ) self.global_pool = global_pool self.drop = nn.Dropout(drop_rate) self.fc = fc self.flatten = nn.Flatten(1) if use_conv and pool_type else nn.Identity() def reset(self, num_classes, pool_type=None): if pool_type is not None and pool_type != self.global_pool.pool_type: self.global_pool, self.fc = create_classifier( self.in_features, num_classes, pool_type=pool_type, use_conv=self.use_conv, input_fmt=self.input_fmt, ) self.flatten = nn.Flatten(1) if self.use_conv and pool_type else nn.Identity() else: num_pooled_features = self.in_features * self.global_pool.feat_mult() self.fc = _create_fc( num_pooled_features, num_classes, use_conv=self.use_conv, ) def forward(self, x, pre_logits: bool = False): x = self.global_pool(x) x = self.drop(x) if pre_logits: return self.flatten(x) x = self.fc(x) return self.flatten(x) class NormMlpClassifierHead(nn.Module): def __init__( self, in_features: int, num_classes: int, hidden_size: Optional[int] = None, pool_type: str = 'avg', drop_rate: float = 0., norm_layer: Union[str, Callable] = 'layernorm2d', act_layer: Union[str, Callable] = 'tanh', ): """ Args: in_features: The number of input features. num_classes: The number of classes for the final classifier layer (output). hidden_size: The hidden size of the MLP (pre-logits FC layer) if not None. pool_type: Global pooling type, pooling disabled if empty string (''). drop_rate: Pre-classifier dropout rate. norm_layer: Normalization layer type. act_layer: MLP activation layer type (only used if hidden_size is not None). """ super().__init__() self.in_features = in_features self.hidden_size = hidden_size self.num_features = in_features self.use_conv = not pool_type norm_layer = get_norm_layer(norm_layer) act_layer = get_act_layer(act_layer) linear_layer = partial(nn.Conv2d, kernel_size=1) if self.use_conv else nn.Linear self.global_pool = SelectAdaptivePool2d(pool_type=pool_type) self.norm = norm_layer(in_features) self.flatten = nn.Flatten(1) if pool_type else nn.Identity() if hidden_size: self.pre_logits = nn.Sequential(OrderedDict([ ('fc', linear_layer(in_features, hidden_size)), ('act', act_layer()), ])) self.num_features = hidden_size else: self.pre_logits = nn.Identity() self.drop = nn.Dropout(drop_rate) self.fc = linear_layer(self.num_features, num_classes) if num_classes > 0 else nn.Identity() def reset(self, num_classes, global_pool=None): if global_pool is not None: self.global_pool = SelectAdaptivePool2d(pool_type=global_pool) self.flatten = nn.Flatten(1) if global_pool else nn.Identity() self.use_conv = self.global_pool.is_identity() linear_layer = partial(nn.Conv2d, kernel_size=1) if self.use_conv else nn.Linear if self.hidden_size: if ((isinstance(self.pre_logits.fc, nn.Conv2d) and not self.use_conv) or (isinstance(self.pre_logits.fc, nn.Linear) and self.use_conv)): with torch.no_grad(): new_fc = linear_layer(self.in_features, self.hidden_size) new_fc.weight.copy_(self.pre_logits.fc.weight.reshape(new_fc.weight.shape)) new_fc.bias.copy_(self.pre_logits.fc.bias) self.pre_logits.fc = new_fc self.fc = linear_layer(self.num_features, num_classes) if num_classes > 0 else nn.Identity() def forward(self, x, pre_logits: bool = False): x = self.global_pool(x) x = self.norm(x) x = self.flatten(x) x = self.pre_logits(x) x = self.drop(x) if pre_logits: return x x = self.fc(x) return x

pytorch-image-models/timm/layers/classifier.py/0

""" Gather-Excite Attention Block Paper: `Gather-Excite: Exploiting Feature Context in CNNs` - https://arxiv.org/abs/1810.12348 Official code here, but it's only partial impl in Caffe: https://github.com/hujie-frank/GENet I've tried to support all of the extent both w/ and w/o params. I don't believe I've seen another impl that covers all of the cases. NOTE: extent=0 + extra_params=False is equivalent to Squeeze-and-Excitation Hacked together by / Copyright 2021 Ross Wightman """ import math from torch import nn as nn import torch.nn.functional as F from .create_act import create_act_layer, get_act_layer from .create_conv2d import create_conv2d from .helpers import make_divisible from .mlp import ConvMlp class GatherExcite(nn.Module): """ Gather-Excite Attention Module """ def __init__( self, channels, feat_size=None, extra_params=False, extent=0, use_mlp=True, rd_ratio=1./16, rd_channels=None, rd_divisor=1, add_maxpool=False, act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d, gate_layer='sigmoid'): super(GatherExcite, self).__init__() self.add_maxpool = add_maxpool act_layer = get_act_layer(act_layer) self.extent = extent if extra_params: self.gather = nn.Sequential() if extent == 0: assert feat_size is not None, 'spatial feature size must be specified for global extent w/ params' self.gather.add_module( 'conv1', create_conv2d(channels, channels, kernel_size=feat_size, stride=1, depthwise=True)) if norm_layer: self.gather.add_module(f'norm1', nn.BatchNorm2d(channels)) else: assert extent % 2 == 0 num_conv = int(math.log2(extent)) for i in range(num_conv): self.gather.add_module( f'conv{i + 1}', create_conv2d(channels, channels, kernel_size=3, stride=2, depthwise=True)) if norm_layer: self.gather.add_module(f'norm{i + 1}', nn.BatchNorm2d(channels)) if i != num_conv - 1: self.gather.add_module(f'act{i + 1}', act_layer(inplace=True)) else: self.gather = None if self.extent == 0: self.gk = 0 self.gs = 0 else: assert extent % 2 == 0 self.gk = self.extent * 2 - 1 self.gs = self.extent if not rd_channels: rd_channels = make_divisible(channels * rd_ratio, rd_divisor, round_limit=0.) self.mlp = ConvMlp(channels, rd_channels, act_layer=act_layer) if use_mlp else nn.Identity() self.gate = create_act_layer(gate_layer) def forward(self, x): size = x.shape[-2:] if self.gather is not None: x_ge = self.gather(x) else: if self.extent == 0: # global extent x_ge = x.mean(dim=(2, 3), keepdims=True) if self.add_maxpool: # experimental codepath, may remove or change x_ge = 0.5 * x_ge + 0.5 * x.amax((2, 3), keepdim=True) else: x_ge = F.avg_pool2d( x, kernel_size=self.gk, stride=self.gs, padding=self.gk // 2, count_include_pad=False) if self.add_maxpool: # experimental codepath, may remove or change x_ge = 0.5 * x_ge + 0.5 * F.max_pool2d(x, kernel_size=self.gk, stride=self.gs, padding=self.gk // 2) x_ge = self.mlp(x_ge) if x_ge.shape[-1] != 1 or x_ge.shape[-2] != 1: x_ge = F.interpolate(x_ge, size=size) return x * self.gate(x_ge)

pytorch-image-models/timm/layers/gather_excite.py/0

""" Normalization + Activation Layers Provides Norm+Act fns for standard PyTorch norm layers such as * BatchNorm * GroupNorm * LayerNorm This allows swapping with alternative layers that are natively both norm + act such as * EvoNorm (evo_norm.py) * FilterResponseNorm (filter_response_norm.py) * InplaceABN (inplace_abn.py) Hacked together by / Copyright 2022 Ross Wightman """ from typing import Union, List, Optional, Any import torch from torch import nn as nn from torch.nn import functional as F from torchvision.ops.misc import FrozenBatchNorm2d from .create_act import get_act_layer from .fast_norm import is_fast_norm, fast_group_norm, fast_layer_norm from .trace_utils import _assert def _create_act(act_layer, act_kwargs=None, inplace=False, apply_act=True): act_layer = get_act_layer(act_layer) # string -> nn.Module act_kwargs = act_kwargs or {} if act_layer is not None and apply_act: if inplace: act_kwargs['inplace'] = inplace act = act_layer(**act_kwargs) else: act = nn.Identity() return act class BatchNormAct2d(nn.BatchNorm2d): """BatchNorm + Activation This module performs BatchNorm + Activation in a manner that will remain backwards compatible with weights trained with separate bn, act. This is why we inherit from BN instead of composing it as a .bn member. """ def __init__( self, num_features, eps=1e-5, momentum=0.1, affine=True, track_running_stats=True, apply_act=True, act_layer=nn.ReLU, act_kwargs=None, inplace=True, drop_layer=None, device=None, dtype=None, ): try: factory_kwargs = {'device': device, 'dtype': dtype} super(BatchNormAct2d, self).__init__( num_features, eps=eps, momentum=momentum, affine=affine, track_running_stats=track_running_stats, **factory_kwargs, ) except TypeError: # NOTE for backwards compat with old PyTorch w/o factory device/dtype support super(BatchNormAct2d, self).__init__( num_features, eps=eps, momentum=momentum, affine=affine, track_running_stats=track_running_stats, ) self.drop = drop_layer() if drop_layer is not None else nn.Identity() self.act = _create_act(act_layer, act_kwargs=act_kwargs, inplace=inplace, apply_act=apply_act) def forward(self, x): # cut & paste of torch.nn.BatchNorm2d.forward impl to avoid issues with torchscript and tracing _assert(x.ndim == 4, f'expected 4D input (got {x.ndim}D input)') # exponential_average_factor is set to self.momentum # (when it is available) only so that it gets updated # in ONNX graph when this node is exported to ONNX. if self.momentum is None: exponential_average_factor = 0.0 else: exponential_average_factor = self.momentum if self.training and self.track_running_stats: # TODO: if statement only here to tell the jit to skip emitting this when it is None if self.num_batches_tracked is not None: # type: ignore[has-type] self.num_batches_tracked.add_(1) # type: ignore[has-type] if self.momentum is None: # use cumulative moving average exponential_average_factor = 1.0 / float(self.num_batches_tracked) else: # use exponential moving average exponential_average_factor = self.momentum r""" Decide whether the mini-batch stats should be used for normalization rather than the buffers. Mini-batch stats are used in training mode, and in eval mode when buffers are None. """ if self.training: bn_training = True else: bn_training = (self.running_mean is None) and (self.running_var is None) r""" Buffers are only updated if they are to be tracked and we are in training mode. Thus they only need to be passed when the update should occur (i.e. in training mode when they are tracked), or when buffer stats are used for normalization (i.e. in eval mode when buffers are not None). """ x = F.batch_norm( x, # If buffers are not to be tracked, ensure that they won't be updated self.running_mean if not self.training or self.track_running_stats else None, self.running_var if not self.training or self.track_running_stats else None, self.weight, self.bias, bn_training, exponential_average_factor, self.eps, ) x = self.drop(x) x = self.act(x) return x class SyncBatchNormAct(nn.SyncBatchNorm): # Thanks to Selim Seferbekov (https://github.com/rwightman/pytorch-image-models/issues/1254) # This is a quick workaround to support SyncBatchNorm for timm BatchNormAct2d layers # but ONLY when used in conjunction with the timm conversion function below. # Do not create this module directly or use the PyTorch conversion function. def forward(self, x: torch.Tensor) -> torch.Tensor: x = super().forward(x) # SyncBN doesn't work with torchscript anyways, so this is fine if hasattr(self, "drop"): x = self.drop(x) if hasattr(self, "act"): x = self.act(x) return x def convert_sync_batchnorm(module, process_group=None): # convert both BatchNorm and BatchNormAct layers to Synchronized variants module_output = module if isinstance(module, torch.nn.modules.batchnorm._BatchNorm): if isinstance(module, BatchNormAct2d): # convert timm norm + act layer module_output = SyncBatchNormAct( module.num_features, module.eps, module.momentum, module.affine, module.track_running_stats, process_group=process_group, ) # set act and drop attr from the original module module_output.act = module.act module_output.drop = module.drop else: # convert standard BatchNorm layers module_output = torch.nn.SyncBatchNorm( module.num_features, module.eps, module.momentum, module.affine, module.track_running_stats, process_group, ) if module.affine: with torch.no_grad(): module_output.weight = module.weight module_output.bias = module.bias module_output.running_mean = module.running_mean module_output.running_var = module.running_var module_output.num_batches_tracked = module.num_batches_tracked if hasattr(module, "qconfig"): module_output.qconfig = module.qconfig for name, child in module.named_children(): module_output.add_module(name, convert_sync_batchnorm(child, process_group)) del module return module_output class FrozenBatchNormAct2d(torch.nn.Module): """ BatchNormAct2d where the batch statistics and the affine parameters are fixed Args: num_features (int): Number of features ``C`` from an expected input of size ``(N, C, H, W)`` eps (float): a value added to the denominator for numerical stability. Default: 1e-5 """ def __init__( self, num_features: int, eps: float = 1e-5, apply_act=True, act_layer=nn.ReLU, act_kwargs=None, inplace=True, drop_layer=None, ): super().__init__() self.eps = eps self.register_buffer("weight", torch.ones(num_features)) self.register_buffer("bias", torch.zeros(num_features)) self.register_buffer("running_mean", torch.zeros(num_features)) self.register_buffer("running_var", torch.ones(num_features)) self.drop = drop_layer() if drop_layer is not None else nn.Identity() self.act = _create_act(act_layer, act_kwargs=act_kwargs, inplace=inplace, apply_act=apply_act) def _load_from_state_dict( self, state_dict: dict, prefix: str, local_metadata: dict, strict: bool, missing_keys: List[str], unexpected_keys: List[str], error_msgs: List[str], ): num_batches_tracked_key = prefix + "num_batches_tracked" if num_batches_tracked_key in state_dict: del state_dict[num_batches_tracked_key] super()._load_from_state_dict( state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs ) def forward(self, x: torch.Tensor) -> torch.Tensor: # move reshapes to the beginning # to make it fuser-friendly w = self.weight.reshape(1, -1, 1, 1) b = self.bias.reshape(1, -1, 1, 1) rv = self.running_var.reshape(1, -1, 1, 1) rm = self.running_mean.reshape(1, -1, 1, 1) scale = w * (rv + self.eps).rsqrt() bias = b - rm * scale x = x * scale + bias x = self.act(self.drop(x)) return x def __repr__(self) -> str: return f"{self.__class__.__name__}({self.weight.shape[0]}, eps={self.eps}, act={self.act})" def freeze_batch_norm_2d(module): """ Converts all `BatchNorm2d` and `SyncBatchNorm` or `BatchNormAct2d` and `SyncBatchNormAct2d` layers of provided module into `FrozenBatchNorm2d` or `FrozenBatchNormAct2d` respectively. Args: module (torch.nn.Module): Any PyTorch module. Returns: torch.nn.Module: Resulting module Inspired by https://github.com/pytorch/pytorch/blob/a5895f85be0f10212791145bfedc0261d364f103/torch/nn/modules/batchnorm.py#L762 """ res = module if isinstance(module, (BatchNormAct2d, SyncBatchNormAct)): res = FrozenBatchNormAct2d(module.num_features) res.num_features = module.num_features res.affine = module.affine if module.affine: res.weight.data = module.weight.data.clone().detach() res.bias.data = module.bias.data.clone().detach() res.running_mean.data = module.running_mean.data res.running_var.data = module.running_var.data res.eps = module.eps res.drop = module.drop res.act = module.act elif isinstance(module, (torch.nn.modules.batchnorm.BatchNorm2d, torch.nn.modules.batchnorm.SyncBatchNorm)): res = FrozenBatchNorm2d(module.num_features) res.num_features = module.num_features res.affine = module.affine if module.affine: res.weight.data = module.weight.data.clone().detach() res.bias.data = module.bias.data.clone().detach() res.running_mean.data = module.running_mean.data res.running_var.data = module.running_var.data res.eps = module.eps else: for name, child in module.named_children(): new_child = freeze_batch_norm_2d(child) if new_child is not child: res.add_module(name, new_child) return res def unfreeze_batch_norm_2d(module): """ Converts all `FrozenBatchNorm2d` layers of provided module into `BatchNorm2d`. If `module` is itself and instance of `FrozenBatchNorm2d`, it is converted into `BatchNorm2d` and returned. Otherwise, the module is walked recursively and submodules are converted in place. Args: module (torch.nn.Module): Any PyTorch module. Returns: torch.nn.Module: Resulting module Inspired by https://github.com/pytorch/pytorch/blob/a5895f85be0f10212791145bfedc0261d364f103/torch/nn/modules/batchnorm.py#L762 """ res = module if isinstance(module, FrozenBatchNormAct2d): res = BatchNormAct2d(module.num_features) if module.affine: res.weight.data = module.weight.data.clone().detach() res.bias.data = module.bias.data.clone().detach() res.running_mean.data = module.running_mean.data res.running_var.data = module.running_var.data res.eps = module.eps res.drop = module.drop res.act = module.act elif isinstance(module, FrozenBatchNorm2d): res = torch.nn.BatchNorm2d(module.num_features) if module.affine: res.weight.data = module.weight.data.clone().detach() res.bias.data = module.bias.data.clone().detach() res.running_mean.data = module.running_mean.data res.running_var.data = module.running_var.data res.eps = module.eps else: for name, child in module.named_children(): new_child = unfreeze_batch_norm_2d(child) if new_child is not child: res.add_module(name, new_child) return res def _num_groups(num_channels, num_groups, group_size): if group_size: assert num_channels % group_size == 0 return num_channels // group_size return num_groups class GroupNormAct(nn.GroupNorm): # NOTE num_channel and num_groups order flipped for easier layer swaps / binding of fixed args def __init__( self, num_channels, num_groups=32, eps=1e-5, affine=True, group_size=None, apply_act=True, act_layer=nn.ReLU, act_kwargs=None, inplace=True, drop_layer=None, ): super(GroupNormAct, self).__init__( _num_groups(num_channels, num_groups, group_size), num_channels, eps=eps, affine=affine, ) self.drop = drop_layer() if drop_layer is not None else nn.Identity() self.act = _create_act(act_layer, act_kwargs=act_kwargs, inplace=inplace, apply_act=apply_act) self._fast_norm = is_fast_norm() def forward(self, x): if self._fast_norm: x = fast_group_norm(x, self.num_groups, self.weight, self.bias, self.eps) else: x = F.group_norm(x, self.num_groups, self.weight, self.bias, self.eps) x = self.drop(x) x = self.act(x) return x class GroupNorm1Act(nn.GroupNorm): def __init__( self, num_channels, eps=1e-5, affine=True, apply_act=True, act_layer=nn.ReLU, act_kwargs=None, inplace=True, drop_layer=None, ): super(GroupNorm1Act, self).__init__(1, num_channels, eps=eps, affine=affine) self.drop = drop_layer() if drop_layer is not None else nn.Identity() self.act = _create_act(act_layer, act_kwargs=act_kwargs, inplace=inplace, apply_act=apply_act) self._fast_norm = is_fast_norm() def forward(self, x): if self._fast_norm: x = fast_group_norm(x, self.num_groups, self.weight, self.bias, self.eps) else: x = F.group_norm(x, self.num_groups, self.weight, self.bias, self.eps) x = self.drop(x) x = self.act(x) return x class LayerNormAct(nn.LayerNorm): def __init__( self, normalization_shape: Union[int, List[int], torch.Size], eps=1e-5, affine=True, apply_act=True, act_layer=nn.ReLU, act_kwargs=None, inplace=True, drop_layer=None, ): super(LayerNormAct, self).__init__(normalization_shape, eps=eps, elementwise_affine=affine) self.drop = drop_layer() if drop_layer is not None else nn.Identity() act_layer = get_act_layer(act_layer) # string -> nn.Module self.act = _create_act(act_layer, act_kwargs=act_kwargs, inplace=inplace, apply_act=apply_act) self._fast_norm = is_fast_norm() def forward(self, x): if self._fast_norm: x = fast_layer_norm(x, self.normalized_shape, self.weight, self.bias, self.eps) else: x = F.layer_norm(x, self.normalized_shape, self.weight, self.bias, self.eps) x = self.drop(x) x = self.act(x) return x class LayerNormAct2d(nn.LayerNorm): def __init__( self, num_channels, eps=1e-5, affine=True, apply_act=True, act_layer=nn.ReLU, act_kwargs=None, inplace=True, drop_layer=None, ): super(LayerNormAct2d, self).__init__(num_channels, eps=eps, elementwise_affine=affine) self.drop = drop_layer() if drop_layer is not None else nn.Identity() self.act = _create_act(act_layer, act_kwargs=act_kwargs, inplace=inplace, apply_act=apply_act) self._fast_norm = is_fast_norm() def forward(self, x): x = x.permute(0, 2, 3, 1) if self._fast_norm: x = fast_layer_norm(x, self.normalized_shape, self.weight, self.bias, self.eps) else: x = F.layer_norm(x, self.normalized_shape, self.weight, self.bias, self.eps) x = x.permute(0, 3, 1, 2) x = self.drop(x) x = self.act(x) return x

pytorch-image-models/timm/layers/norm_act.py/0

try: from torch import _assert except ImportError: def _assert(condition: bool, message: str): assert condition, message def _float_to_int(x: float) -> int: """ Symbolic tracing helper to substitute for inbuilt `int`. Hint: Inbuilt `int` can't accept an argument of type `Proxy` """ return int(x)

pytorch-image-models/timm/layers/trace_utils.py/0

import hashlib import json import logging import os from functools import partial from pathlib import Path from tempfile import TemporaryDirectory from typing import Iterable, Optional, Union import torch from torch.hub import HASH_REGEX, download_url_to_file, urlparse try: from torch.hub import get_dir except ImportError: from torch.hub import _get_torch_home as get_dir try: import safetensors.torch _has_safetensors = True except ImportError: _has_safetensors = False try: from typing import Literal except ImportError: from typing_extensions import Literal from timm import __version__ from timm.models._pretrained import filter_pretrained_cfg try: from huggingface_hub import ( create_repo, get_hf_file_metadata, hf_hub_download, hf_hub_url, repo_type_and_id_from_hf_id, upload_folder) from huggingface_hub.utils import EntryNotFoundError hf_hub_download = partial(hf_hub_download, library_name="timm", library_version=__version__) _has_hf_hub = True except ImportError: hf_hub_download = None _has_hf_hub = False _logger = logging.getLogger(__name__) __all__ = ['get_cache_dir', 'download_cached_file', 'has_hf_hub', 'hf_split', 'load_model_config_from_hf', 'load_state_dict_from_hf', 'save_for_hf', 'push_to_hf_hub'] # Default name for a weights file hosted on the Huggingface Hub. HF_WEIGHTS_NAME = "pytorch_model.bin" # default pytorch pkl HF_SAFE_WEIGHTS_NAME = "model.safetensors" # safetensors version HF_OPEN_CLIP_WEIGHTS_NAME = "open_clip_pytorch_model.bin" # default pytorch pkl HF_OPEN_CLIP_SAFE_WEIGHTS_NAME = "open_clip_model.safetensors" # safetensors version def get_cache_dir(child_dir=''): """ Returns the location of the directory where models are cached (and creates it if necessary). """ # Issue warning to move data if old env is set if os.getenv('TORCH_MODEL_ZOO'): _logger.warning('TORCH_MODEL_ZOO is deprecated, please use env TORCH_HOME instead') hub_dir = get_dir() child_dir = () if not child_dir else (child_dir,) model_dir = os.path.join(hub_dir, 'checkpoints', *child_dir) os.makedirs(model_dir, exist_ok=True) return model_dir def download_cached_file(url, check_hash=True, progress=False): if isinstance(url, (list, tuple)): url, filename = url else: parts = urlparse(url) filename = os.path.basename(parts.path) cached_file = os.path.join(get_cache_dir(), filename) if not os.path.exists(cached_file): _logger.info('Downloading: "{}" to {}\n'.format(url, cached_file)) hash_prefix = None if check_hash: r = HASH_REGEX.search(filename) # r is Optional[Match[str]] hash_prefix = r.group(1) if r else None download_url_to_file(url, cached_file, hash_prefix, progress=progress) return cached_file def check_cached_file(url, check_hash=True): if isinstance(url, (list, tuple)): url, filename = url else: parts = urlparse(url) filename = os.path.basename(parts.path) cached_file = os.path.join(get_cache_dir(), filename) if os.path.exists(cached_file): if check_hash: r = HASH_REGEX.search(filename) # r is Optional[Match[str]] hash_prefix = r.group(1) if r else None if hash_prefix: with open(cached_file, 'rb') as f: hd = hashlib.sha256(f.read()).hexdigest() if hd[:len(hash_prefix)] != hash_prefix: return False return True return False def has_hf_hub(necessary=False): if not _has_hf_hub and necessary: # if no HF Hub module installed, and it is necessary to continue, raise error raise RuntimeError( 'Hugging Face hub model specified but package not installed. Run `pip install huggingface_hub`.') return _has_hf_hub def hf_split(hf_id: str): # FIXME I may change @ -> # and be parsed as fragment in a URI model name scheme rev_split = hf_id.split('@') assert 0 < len(rev_split) <= 2, 'hf_hub id should only contain one @ character to identify revision.' hf_model_id = rev_split[0] hf_revision = rev_split[-1] if len(rev_split) > 1 else None return hf_model_id, hf_revision def load_cfg_from_json(json_file: Union[str, os.PathLike]): with open(json_file, "r", encoding="utf-8") as reader: text = reader.read() return json.loads(text) def download_from_hf(model_id: str, filename: str): hf_model_id, hf_revision = hf_split(model_id) return hf_hub_download(hf_model_id, filename, revision=hf_revision) def load_model_config_from_hf(model_id: str): assert has_hf_hub(True) cached_file = download_from_hf(model_id, 'config.json') hf_config = load_cfg_from_json(cached_file) if 'pretrained_cfg' not in hf_config: # old form, pull pretrain_cfg out of the base dict pretrained_cfg = hf_config hf_config = {} hf_config['architecture'] = pretrained_cfg.pop('architecture') hf_config['num_features'] = pretrained_cfg.pop('num_features', None) if 'labels' in pretrained_cfg: # deprecated name for 'label_names' pretrained_cfg['label_names'] = pretrained_cfg.pop('labels') hf_config['pretrained_cfg'] = pretrained_cfg # NOTE currently discarding parent config as only arch name and pretrained_cfg used in timm right now pretrained_cfg = hf_config['pretrained_cfg'] pretrained_cfg['hf_hub_id'] = model_id # insert hf_hub id for pretrained weight load during model creation pretrained_cfg['source'] = 'hf-hub' # model should be created with base config num_classes if its exist if 'num_classes' in hf_config: pretrained_cfg['num_classes'] = hf_config['num_classes'] # label meta-data in base config overrides saved pretrained_cfg on load if 'label_names' in hf_config: pretrained_cfg['label_names'] = hf_config.pop('label_names') if 'label_descriptions' in hf_config: pretrained_cfg['label_descriptions'] = hf_config.pop('label_descriptions') model_args = hf_config.get('model_args', {}) model_name = hf_config['architecture'] return pretrained_cfg, model_name, model_args def load_state_dict_from_hf(model_id: str, filename: str = HF_WEIGHTS_NAME): assert has_hf_hub(True) hf_model_id, hf_revision = hf_split(model_id) # Look for .safetensors alternatives and load from it if it exists if _has_safetensors: for safe_filename in _get_safe_alternatives(filename): try: cached_safe_file = hf_hub_download(repo_id=hf_model_id, filename=safe_filename, revision=hf_revision) _logger.info( f"[{model_id}] Safe alternative available for '{filename}' " f"(as '{safe_filename}'). Loading weights using safetensors.") return safetensors.torch.load_file(cached_safe_file, device="cpu") except EntryNotFoundError: pass # Otherwise, load using pytorch.load cached_file = hf_hub_download(hf_model_id, filename=filename, revision=hf_revision) _logger.debug(f"[{model_id}] Safe alternative not found for '{filename}'. Loading weights using default pytorch.") return torch.load(cached_file, map_location='cpu') def save_config_for_hf( model, config_path: str, model_config: Optional[dict] = None, model_args: Optional[dict] = None ): model_config = model_config or {} hf_config = {} pretrained_cfg = filter_pretrained_cfg(model.pretrained_cfg, remove_source=True, remove_null=True) # set some values at root config level hf_config['architecture'] = pretrained_cfg.pop('architecture') hf_config['num_classes'] = model_config.pop('num_classes', model.num_classes) # NOTE these attr saved for informational purposes, do not impact model build hf_config['num_features'] = model_config.pop('num_features', model.num_features) global_pool_type = model_config.pop('global_pool', getattr(model, 'global_pool', None)) if isinstance(global_pool_type, str) and global_pool_type: hf_config['global_pool'] = global_pool_type # Save class label info if 'labels' in model_config: _logger.warning( "'labels' as a config field for is deprecated. Please use 'label_names' and 'label_descriptions'." " Renaming provided 'labels' field to 'label_names'.") model_config.setdefault('label_names', model_config.pop('labels')) label_names = model_config.pop('label_names', None) if label_names: assert isinstance(label_names, (dict, list, tuple)) # map label id (classifier index) -> unique label name (ie synset for ImageNet, MID for OpenImages) # can be a dict id: name if there are id gaps, or tuple/list if no gaps. hf_config['label_names'] = label_names label_descriptions = model_config.pop('label_descriptions', None) if label_descriptions: assert isinstance(label_descriptions, dict) # maps label names -> descriptions hf_config['label_descriptions'] = label_descriptions if model_args: hf_config['model_args'] = model_args hf_config['pretrained_cfg'] = pretrained_cfg hf_config.update(model_config) with config_path.open('w') as f: json.dump(hf_config, f, indent=2) def save_for_hf( model, save_directory: str, model_config: Optional[dict] = None, model_args: Optional[dict] = None, safe_serialization: Union[bool, Literal["both"]] = False, ): assert has_hf_hub(True) save_directory = Path(save_directory) save_directory.mkdir(exist_ok=True, parents=True) # Save model weights, either safely (using safetensors), or using legacy pytorch approach or both. tensors = model.state_dict() if safe_serialization is True or safe_serialization == "both": assert _has_safetensors, "`pip install safetensors` to use .safetensors" safetensors.torch.save_file(tensors, save_directory / HF_SAFE_WEIGHTS_NAME) if safe_serialization is False or safe_serialization == "both": torch.save(tensors, save_directory / HF_WEIGHTS_NAME) config_path = save_directory / 'config.json' save_config_for_hf( model, config_path, model_config=model_config, model_args=model_args, ) def push_to_hf_hub( model: torch.nn.Module, repo_id: str, commit_message: str = 'Add model', token: Optional[str] = None, revision: Optional[str] = None, private: bool = False, create_pr: bool = False, model_config: Optional[dict] = None, model_card: Optional[dict] = None, model_args: Optional[dict] = None, safe_serialization: Union[bool, Literal["both"]] = False, ): """ Arguments: (...) safe_serialization (`bool` or `"both"`, *optional*, defaults to `False`): Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`). Can be set to `"both"` in order to push both safe and unsafe weights. """ # Create repo if it doesn't exist yet repo_url = create_repo(repo_id, token=token, private=private, exist_ok=True) # Infer complete repo_id from repo_url # Can be different from the input `repo_id` if repo_owner was implicit _, repo_owner, repo_name = repo_type_and_id_from_hf_id(repo_url) repo_id = f"{repo_owner}/{repo_name}" # Check if README file already exist in repo try: get_hf_file_metadata(hf_hub_url(repo_id=repo_id, filename="README.md", revision=revision)) has_readme = True except EntryNotFoundError: has_readme = False # Dump model and push to Hub with TemporaryDirectory() as tmpdir: # Save model weights and config. save_for_hf( model, tmpdir, model_config=model_config, model_args=model_args, safe_serialization=safe_serialization, ) # Add readme if it does not exist if not has_readme: model_card = model_card or {} model_name = repo_id.split('/')[-1] readme_path = Path(tmpdir) / "README.md" readme_text = generate_readme(model_card, model_name) readme_path.write_text(readme_text) # Upload model and return return upload_folder( repo_id=repo_id, folder_path=tmpdir, revision=revision, create_pr=create_pr, commit_message=commit_message, ) def generate_readme(model_card: dict, model_name: str): readme_text = "---\n" readme_text += "tags:\n- image-classification\n- timm\n" readme_text += "library_name: timm\n" readme_text += f"license: {model_card.get('license', 'apache-2.0')}\n" if 'details' in model_card and 'Dataset' in model_card['details']: readme_text += 'datasets:\n' if isinstance(model_card['details']['Dataset'], (tuple, list)): for d in model_card['details']['Dataset']: readme_text += f"- {d.lower()}\n" else: readme_text += f"- {model_card['details']['Dataset'].lower()}\n" if 'Pretrain Dataset' in model_card['details']: if isinstance(model_card['details']['Pretrain Dataset'], (tuple, list)): for d in model_card['details']['Pretrain Dataset']: readme_text += f"- {d.lower()}\n" else: readme_text += f"- {model_card['details']['Pretrain Dataset'].lower()}\n" readme_text += "---\n" readme_text += f"# Model card for {model_name}\n" if 'description' in model_card: readme_text += f"\n{model_card['description']}\n" if 'details' in model_card: readme_text += f"\n## Model Details\n" for k, v in model_card['details'].items(): if isinstance(v, (list, tuple)): readme_text += f"- **{k}:**\n" for vi in v: readme_text += f" - {vi}\n" elif isinstance(v, dict): readme_text += f"- **{k}:**\n" for ki, vi in v.items(): readme_text += f" - {ki}: {vi}\n" else: readme_text += f"- **{k}:** {v}\n" if 'usage' in model_card: readme_text += f"\n## Model Usage\n" readme_text += model_card['usage'] readme_text += '\n' if 'comparison' in model_card: readme_text += f"\n## Model Comparison\n" readme_text += model_card['comparison'] readme_text += '\n' if 'citation' in model_card: readme_text += f"\n## Citation\n" if not isinstance(model_card['citation'], (list, tuple)): citations = [model_card['citation']] else: citations = model_card['citation'] for c in citations: readme_text += f"```bibtex\n{c}\n```\n" return readme_text def _get_safe_alternatives(filename: str) -> Iterable[str]: """Returns potential safetensors alternatives for a given filename. Use case: When downloading a model from the Huggingface Hub, we first look if a .safetensors file exists and if yes, we use it. Main use case is filename "pytorch_model.bin" => check for "model.safetensors" or "pytorch_model.safetensors". """ if filename == HF_WEIGHTS_NAME: yield HF_SAFE_WEIGHTS_NAME if filename == HF_OPEN_CLIP_WEIGHTS_NAME: yield HF_OPEN_CLIP_SAFE_WEIGHTS_NAME if filename not in (HF_WEIGHTS_NAME, HF_OPEN_CLIP_WEIGHTS_NAME) and filename.endswith(".bin"): yield filename[:-4] + ".safetensors"

pytorch-image-models/timm/models/_hub.py/0

""" ConvMixer """ import torch import torch.nn as nn from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD from timm.layers import SelectAdaptivePool2d from ._registry import register_model, generate_default_cfgs from ._builder import build_model_with_cfg from ._manipulate import checkpoint_seq __all__ = ['ConvMixer'] class Residual(nn.Module): def __init__(self, fn): super().__init__() self.fn = fn def forward(self, x): return self.fn(x) + x class ConvMixer(nn.Module): def __init__( self, dim, depth, kernel_size=9, patch_size=7, in_chans=3, num_classes=1000, global_pool='avg', drop_rate=0., act_layer=nn.GELU, **kwargs, ): super().__init__() self.num_classes = num_classes self.num_features = dim self.grad_checkpointing = False self.stem = nn.Sequential( nn.Conv2d(in_chans, dim, kernel_size=patch_size, stride=patch_size), act_layer(), nn.BatchNorm2d(dim) ) self.blocks = nn.Sequential( *[nn.Sequential( Residual(nn.Sequential( nn.Conv2d(dim, dim, kernel_size, groups=dim, padding="same"), act_layer(), nn.BatchNorm2d(dim) )), nn.Conv2d(dim, dim, kernel_size=1), act_layer(), nn.BatchNorm2d(dim) ) for i in range(depth)] ) self.pooling = SelectAdaptivePool2d(pool_type=global_pool, flatten=True) self.head_drop = nn.Dropout(drop_rate) self.head = nn.Linear(dim, num_classes) if num_classes > 0 else nn.Identity() @torch.jit.ignore def group_matcher(self, coarse=False): matcher = dict(stem=r'^stem', blocks=r'^blocks\.(\d+)') return matcher @torch.jit.ignore def set_grad_checkpointing(self, enable=True): self.grad_checkpointing = enable @torch.jit.ignore def get_classifier(self): return self.head def reset_classifier(self, num_classes, global_pool=None): self.num_classes = num_classes if global_pool is not None: self.pooling = SelectAdaptivePool2d(pool_type=global_pool, flatten=True) self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity() def forward_features(self, x): x = self.stem(x) if self.grad_checkpointing and not torch.jit.is_scripting(): x = checkpoint_seq(self.blocks, x) else: x = self.blocks(x) return x def forward_head(self, x, pre_logits: bool = False): x = self.pooling(x) x = self.head_drop(x) return x if pre_logits else self.head(x) def forward(self, x): x = self.forward_features(x) x = self.forward_head(x) return x def _create_convmixer(variant, pretrained=False, **kwargs): if kwargs.get('features_only', None): raise RuntimeError('features_only not implemented for ConvMixer models.') return build_model_with_cfg(ConvMixer, variant, pretrained, **kwargs) def _cfg(url='', **kwargs): return { 'url': url, 'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None, 'crop_pct': .96, 'interpolation': 'bicubic', 'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD, 'classifier': 'head', 'first_conv': 'stem.0', **kwargs } default_cfgs = generate_default_cfgs({ 'convmixer_1536_20.in1k': _cfg(hf_hub_id='timm/'), 'convmixer_768_32.in1k': _cfg(hf_hub_id='timm/'), 'convmixer_1024_20_ks9_p14.in1k': _cfg(hf_hub_id='timm/') }) @register_model def convmixer_1536_20(pretrained=False, **kwargs) -> ConvMixer: model_args = dict(dim=1536, depth=20, kernel_size=9, patch_size=7, **kwargs) return _create_convmixer('convmixer_1536_20', pretrained, **model_args) @register_model def convmixer_768_32(pretrained=False, **kwargs) -> ConvMixer: model_args = dict(dim=768, depth=32, kernel_size=7, patch_size=7, act_layer=nn.ReLU, **kwargs) return _create_convmixer('convmixer_768_32', pretrained, **model_args) @register_model def convmixer_1024_20_ks9_p14(pretrained=False, **kwargs) -> ConvMixer: model_args = dict(dim=1024, depth=20, kernel_size=9, patch_size=14, **kwargs) return _create_convmixer('convmixer_1024_20_ks9_p14', pretrained, **model_args)

pytorch-image-models/timm/models/convmixer.py/0

""" LeViT Paper: `LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference` - https://arxiv.org/abs/2104.01136 @article{graham2021levit, title={LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference}, author={Benjamin Graham and Alaaeldin El-Nouby and Hugo Touvron and Pierre Stock and Armand Joulin and Herv\'e J\'egou and Matthijs Douze}, journal={arXiv preprint arXiv:22104.01136}, year={2021} } Adapted from official impl at https://github.com/facebookresearch/LeViT, original copyright bellow. This version combines both conv/linear models and fixes torchscript compatibility. Modifications and additions for timm hacked together by / Copyright 2021, Ross Wightman """ # Copyright (c) 2015-present, Facebook, Inc. # All rights reserved. # Modified from # https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py # Copyright 2020 Ross Wightman, Apache-2.0 License from collections import OrderedDict from functools import partial from typing import Dict import torch import torch.nn as nn from timm.data import IMAGENET_DEFAULT_STD, IMAGENET_DEFAULT_MEAN from timm.layers import to_ntuple, to_2tuple, get_act_layer, DropPath, trunc_normal_, ndgrid from ._builder import build_model_with_cfg from ._manipulate import checkpoint_seq from ._registry import generate_default_cfgs, register_model __all__ = ['Levit'] class ConvNorm(nn.Module): def __init__( self, in_chs, out_chs, kernel_size=1, stride=1, padding=0, dilation=1, groups=1, bn_weight_init=1): super().__init__() self.linear = nn.Conv2d(in_chs, out_chs, kernel_size, stride, padding, dilation, groups, bias=False) self.bn = nn.BatchNorm2d(out_chs) nn.init.constant_(self.bn.weight, bn_weight_init) @torch.no_grad() def fuse(self): c, bn = self.linear, self.bn w = bn.weight / (bn.running_var + bn.eps) ** 0.5 w = c.weight * w[:, None, None, None] b = bn.bias - bn.running_mean * bn.weight / (bn.running_var + bn.eps) ** 0.5 m = nn.Conv2d( w.size(1), w.size(0), w.shape[2:], stride=self.linear.stride, padding=self.linear.padding, dilation=self.linear.dilation, groups=self.linear.groups) m.weight.data.copy_(w) m.bias.data.copy_(b) return m def forward(self, x): return self.bn(self.linear(x)) class LinearNorm(nn.Module): def __init__(self, in_features, out_features, bn_weight_init=1): super().__init__() self.linear = nn.Linear(in_features, out_features, bias=False) self.bn = nn.BatchNorm1d(out_features) nn.init.constant_(self.bn.weight, bn_weight_init) @torch.no_grad() def fuse(self): l, bn = self.linear, self.bn w = bn.weight / (bn.running_var + bn.eps) ** 0.5 w = l.weight * w[:, None] b = bn.bias - bn.running_mean * bn.weight / (bn.running_var + bn.eps) ** 0.5 m = nn.Linear(w.size(1), w.size(0)) m.weight.data.copy_(w) m.bias.data.copy_(b) return m def forward(self, x): x = self.linear(x) return self.bn(x.flatten(0, 1)).reshape_as(x) class NormLinear(nn.Module): def __init__(self, in_features, out_features, bias=True, std=0.02, drop=0.): super().__init__() self.bn = nn.BatchNorm1d(in_features) self.drop = nn.Dropout(drop) self.linear = nn.Linear(in_features, out_features, bias=bias) trunc_normal_(self.linear.weight, std=std) if self.linear.bias is not None: nn.init.constant_(self.linear.bias, 0) @torch.no_grad() def fuse(self): bn, l = self.bn, self.linear w = bn.weight / (bn.running_var + bn.eps) ** 0.5 b = bn.bias - self.bn.running_mean * self.bn.weight / (bn.running_var + bn.eps) ** 0.5 w = l.weight * w[None, :] if l.bias is None: b = b @ self.linear.weight.T else: b = (l.weight @ b[:, None]).view(-1) + self.linear.bias m = nn.Linear(w.size(1), w.size(0)) m.weight.data.copy_(w) m.bias.data.copy_(b) return m def forward(self, x): return self.linear(self.drop(self.bn(x))) class Stem8(nn.Sequential): def __init__(self, in_chs, out_chs, act_layer): super().__init__() self.stride = 8 self.add_module('conv1', ConvNorm(in_chs, out_chs // 4, 3, stride=2, padding=1)) self.add_module('act1', act_layer()) self.add_module('conv2', ConvNorm(out_chs // 4, out_chs // 2, 3, stride=2, padding=1)) self.add_module('act2', act_layer()) self.add_module('conv3', ConvNorm(out_chs // 2, out_chs, 3, stride=2, padding=1)) class Stem16(nn.Sequential): def __init__(self, in_chs, out_chs, act_layer): super().__init__() self.stride = 16 self.add_module('conv1', ConvNorm(in_chs, out_chs // 8, 3, stride=2, padding=1)) self.add_module('act1', act_layer()) self.add_module('conv2', ConvNorm(out_chs // 8, out_chs // 4, 3, stride=2, padding=1)) self.add_module('act2', act_layer()) self.add_module('conv3', ConvNorm(out_chs // 4, out_chs // 2, 3, stride=2, padding=1)) self.add_module('act3', act_layer()) self.add_module('conv4', ConvNorm(out_chs // 2, out_chs, 3, stride=2, padding=1)) class Downsample(nn.Module): def __init__(self, stride, resolution, use_pool=False): super().__init__() self.stride = stride self.resolution = to_2tuple(resolution) self.pool = nn.AvgPool2d(3, stride=stride, padding=1, count_include_pad=False) if use_pool else None def forward(self, x): B, N, C = x.shape x = x.view(B, self.resolution[0], self.resolution[1], C) if self.pool is not None: x = self.pool(x.permute(0, 3, 1, 2)).permute(0, 2, 3, 1) else: x = x[:, ::self.stride, ::self.stride] return x.reshape(B, -1, C) class Attention(nn.Module): attention_bias_cache: Dict[str, torch.Tensor] def __init__( self, dim, key_dim, num_heads=8, attn_ratio=4., resolution=14, use_conv=False, act_layer=nn.SiLU, ): super().__init__() ln_layer = ConvNorm if use_conv else LinearNorm resolution = to_2tuple(resolution) self.use_conv = use_conv self.num_heads = num_heads self.scale = key_dim ** -0.5 self.key_dim = key_dim self.key_attn_dim = key_dim * num_heads self.val_dim = int(attn_ratio * key_dim) self.val_attn_dim = int(attn_ratio * key_dim) * num_heads self.qkv = ln_layer(dim, self.val_attn_dim + self.key_attn_dim * 2) self.proj = nn.Sequential(OrderedDict([ ('act', act_layer()), ('ln', ln_layer(self.val_attn_dim, dim, bn_weight_init=0)) ])) self.attention_biases = nn.Parameter(torch.zeros(num_heads, resolution[0] * resolution[1])) pos = torch.stack(ndgrid(torch.arange(resolution[0]), torch.arange(resolution[1]))).flatten(1) rel_pos = (pos[..., :, None] - pos[..., None, :]).abs() rel_pos = (rel_pos[0] * resolution[1]) + rel_pos[1] self.register_buffer('attention_bias_idxs', rel_pos, persistent=False) self.attention_bias_cache = {} @torch.no_grad() def train(self, mode=True): super().train(mode) if mode and self.attention_bias_cache: self.attention_bias_cache = {} # clear ab cache def get_attention_biases(self, device: torch.device) -> torch.Tensor: if torch.jit.is_tracing() or self.training: return self.attention_biases[:, self.attention_bias_idxs] else: device_key = str(device) if device_key not in self.attention_bias_cache: self.attention_bias_cache[device_key] = self.attention_biases[:, self.attention_bias_idxs] return self.attention_bias_cache[device_key] def forward(self, x): # x (B,C,H,W) if self.use_conv: B, C, H, W = x.shape q, k, v = self.qkv(x).view( B, self.num_heads, -1, H * W).split([self.key_dim, self.key_dim, self.val_dim], dim=2) attn = (q.transpose(-2, -1) @ k) * self.scale + self.get_attention_biases(x.device) attn = attn.softmax(dim=-1) x = (v @ attn.transpose(-2, -1)).view(B, -1, H, W) else: B, N, C = x.shape q, k, v = self.qkv(x).view( B, N, self.num_heads, -1).split([self.key_dim, self.key_dim, self.val_dim], dim=3) q = q.permute(0, 2, 1, 3) k = k.permute(0, 2, 3, 1) v = v.permute(0, 2, 1, 3) attn = q @ k * self.scale + self.get_attention_biases(x.device) attn = attn.softmax(dim=-1) x = (attn @ v).transpose(1, 2).reshape(B, N, self.val_attn_dim) x = self.proj(x) return x class AttentionDownsample(nn.Module): attention_bias_cache: Dict[str, torch.Tensor] def __init__( self, in_dim, out_dim, key_dim, num_heads=8, attn_ratio=2.0, stride=2, resolution=14, use_conv=False, use_pool=False, act_layer=nn.SiLU, ): super().__init__() resolution = to_2tuple(resolution) self.stride = stride self.resolution = resolution self.num_heads = num_heads self.key_dim = key_dim self.key_attn_dim = key_dim * num_heads self.val_dim = int(attn_ratio * key_dim) self.val_attn_dim = self.val_dim * self.num_heads self.scale = key_dim ** -0.5 self.use_conv = use_conv if self.use_conv: ln_layer = ConvNorm sub_layer = partial( nn.AvgPool2d, kernel_size=3 if use_pool else 1, padding=1 if use_pool else 0, count_include_pad=False) else: ln_layer = LinearNorm sub_layer = partial(Downsample, resolution=resolution, use_pool=use_pool) self.kv = ln_layer(in_dim, self.val_attn_dim + self.key_attn_dim) self.q = nn.Sequential(OrderedDict([ ('down', sub_layer(stride=stride)), ('ln', ln_layer(in_dim, self.key_attn_dim)) ])) self.proj = nn.Sequential(OrderedDict([ ('act', act_layer()), ('ln', ln_layer(self.val_attn_dim, out_dim)) ])) self.attention_biases = nn.Parameter(torch.zeros(num_heads, resolution[0] * resolution[1])) k_pos = torch.stack(ndgrid(torch.arange(resolution[0]), torch.arange(resolution[1]))).flatten(1) q_pos = torch.stack(ndgrid( torch.arange(0, resolution[0], step=stride), torch.arange(0, resolution[1], step=stride) )).flatten(1) rel_pos = (q_pos[..., :, None] - k_pos[..., None, :]).abs() rel_pos = (rel_pos[0] * resolution[1]) + rel_pos[1] self.register_buffer('attention_bias_idxs', rel_pos, persistent=False) self.attention_bias_cache = {} # per-device attention_biases cache @torch.no_grad() def train(self, mode=True): super().train(mode) if mode and self.attention_bias_cache: self.attention_bias_cache = {} # clear ab cache def get_attention_biases(self, device: torch.device) -> torch.Tensor: if torch.jit.is_tracing() or self.training: return self.attention_biases[:, self.attention_bias_idxs] else: device_key = str(device) if device_key not in self.attention_bias_cache: self.attention_bias_cache[device_key] = self.attention_biases[:, self.attention_bias_idxs] return self.attention_bias_cache[device_key] def forward(self, x): if self.use_conv: B, C, H, W = x.shape HH, WW = (H - 1) // self.stride + 1, (W - 1) // self.stride + 1 k, v = self.kv(x).view(B, self.num_heads, -1, H * W).split([self.key_dim, self.val_dim], dim=2) q = self.q(x).view(B, self.num_heads, self.key_dim, -1) attn = (q.transpose(-2, -1) @ k) * self.scale + self.get_attention_biases(x.device) attn = attn.softmax(dim=-1) x = (v @ attn.transpose(-2, -1)).reshape(B, self.val_attn_dim, HH, WW) else: B, N, C = x.shape k, v = self.kv(x).view(B, N, self.num_heads, -1).split([self.key_dim, self.val_dim], dim=3) k = k.permute(0, 2, 3, 1) # BHCN v = v.permute(0, 2, 1, 3) # BHNC q = self.q(x).view(B, -1, self.num_heads, self.key_dim).permute(0, 2, 1, 3) attn = q @ k * self.scale + self.get_attention_biases(x.device) attn = attn.softmax(dim=-1) x = (attn @ v).transpose(1, 2).reshape(B, -1, self.val_attn_dim) x = self.proj(x) return x class LevitMlp(nn.Module): """ MLP for Levit w/ normalization + ability to switch btw conv and linear """ def __init__( self, in_features, hidden_features=None, out_features=None, use_conv=False, act_layer=nn.SiLU, drop=0. ): super().__init__() out_features = out_features or in_features hidden_features = hidden_features or in_features ln_layer = ConvNorm if use_conv else LinearNorm self.ln1 = ln_layer(in_features, hidden_features) self.act = act_layer() self.drop = nn.Dropout(drop) self.ln2 = ln_layer(hidden_features, out_features, bn_weight_init=0) def forward(self, x): x = self.ln1(x) x = self.act(x) x = self.drop(x) x = self.ln2(x) return x class LevitDownsample(nn.Module): def __init__( self, in_dim, out_dim, key_dim, num_heads=8, attn_ratio=4., mlp_ratio=2., act_layer=nn.SiLU, attn_act_layer=None, resolution=14, use_conv=False, use_pool=False, drop_path=0., ): super().__init__() attn_act_layer = attn_act_layer or act_layer self.attn_downsample = AttentionDownsample( in_dim=in_dim, out_dim=out_dim, key_dim=key_dim, num_heads=num_heads, attn_ratio=attn_ratio, act_layer=attn_act_layer, resolution=resolution, use_conv=use_conv, use_pool=use_pool, ) self.mlp = LevitMlp( out_dim, int(out_dim * mlp_ratio), use_conv=use_conv, act_layer=act_layer ) self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity() def forward(self, x): x = self.attn_downsample(x) x = x + self.drop_path(self.mlp(x)) return x class LevitBlock(nn.Module): def __init__( self, dim, key_dim, num_heads=8, attn_ratio=4., mlp_ratio=2., resolution=14, use_conv=False, act_layer=nn.SiLU, attn_act_layer=None, drop_path=0., ): super().__init__() attn_act_layer = attn_act_layer or act_layer self.attn = Attention( dim=dim, key_dim=key_dim, num_heads=num_heads, attn_ratio=attn_ratio, resolution=resolution, use_conv=use_conv, act_layer=attn_act_layer, ) self.drop_path1 = DropPath(drop_path) if drop_path > 0. else nn.Identity() self.mlp = LevitMlp( dim, int(dim * mlp_ratio), use_conv=use_conv, act_layer=act_layer ) self.drop_path2 = DropPath(drop_path) if drop_path > 0. else nn.Identity() def forward(self, x): x = x + self.drop_path1(self.attn(x)) x = x + self.drop_path2(self.mlp(x)) return x class LevitStage(nn.Module): def __init__( self, in_dim, out_dim, key_dim, depth=4, num_heads=8, attn_ratio=4.0, mlp_ratio=4.0, act_layer=nn.SiLU, attn_act_layer=None, resolution=14, downsample='', use_conv=False, drop_path=0., ): super().__init__() resolution = to_2tuple(resolution) if downsample: self.downsample = LevitDownsample( in_dim, out_dim, key_dim=key_dim, num_heads=in_dim // key_dim, attn_ratio=4., mlp_ratio=2., act_layer=act_layer, attn_act_layer=attn_act_layer, resolution=resolution, use_conv=use_conv, drop_path=drop_path, ) resolution = [(r - 1) // 2 + 1 for r in resolution] else: assert in_dim == out_dim self.downsample = nn.Identity() blocks = [] for _ in range(depth): blocks += [LevitBlock( out_dim, key_dim, num_heads=num_heads, attn_ratio=attn_ratio, mlp_ratio=mlp_ratio, act_layer=act_layer, attn_act_layer=attn_act_layer, resolution=resolution, use_conv=use_conv, drop_path=drop_path, )] self.blocks = nn.Sequential(*blocks) def forward(self, x): x = self.downsample(x) x = self.blocks(x) return x class Levit(nn.Module): """ Vision Transformer with support for patch or hybrid CNN input stage NOTE: distillation is defaulted to True since pretrained weights use it, will cause problems w/ train scripts that don't take tuple outputs, """ def __init__( self, img_size=224, in_chans=3, num_classes=1000, embed_dim=(192,), key_dim=64, depth=(12,), num_heads=(3,), attn_ratio=2., mlp_ratio=2., stem_backbone=None, stem_stride=None, stem_type='s16', down_op='subsample', act_layer='hard_swish', attn_act_layer=None, use_conv=False, global_pool='avg', drop_rate=0., drop_path_rate=0.): super().__init__() act_layer = get_act_layer(act_layer) attn_act_layer = get_act_layer(attn_act_layer or act_layer) self.use_conv = use_conv self.num_classes = num_classes self.global_pool = global_pool self.num_features = embed_dim[-1] self.embed_dim = embed_dim self.drop_rate = drop_rate self.grad_checkpointing = False self.feature_info = [] num_stages = len(embed_dim) assert len(depth) == num_stages num_heads = to_ntuple(num_stages)(num_heads) attn_ratio = to_ntuple(num_stages)(attn_ratio) mlp_ratio = to_ntuple(num_stages)(mlp_ratio) if stem_backbone is not None: assert stem_stride >= 2 self.stem = stem_backbone stride = stem_stride else: assert stem_type in ('s16', 's8') if stem_type == 's16': self.stem = Stem16(in_chans, embed_dim[0], act_layer=act_layer) else: self.stem = Stem8(in_chans, embed_dim[0], act_layer=act_layer) stride = self.stem.stride resolution = tuple([i // p for i, p in zip(to_2tuple(img_size), to_2tuple(stride))]) in_dim = embed_dim[0] stages = [] for i in range(num_stages): stage_stride = 2 if i > 0 else 1 stages += [LevitStage( in_dim, embed_dim[i], key_dim, depth=depth[i], num_heads=num_heads[i], attn_ratio=attn_ratio[i], mlp_ratio=mlp_ratio[i], act_layer=act_layer, attn_act_layer=attn_act_layer, resolution=resolution, use_conv=use_conv, downsample=down_op if stage_stride == 2 else '', drop_path=drop_path_rate )] stride *= stage_stride resolution = tuple([(r - 1) // stage_stride + 1 for r in resolution]) self.feature_info += [dict(num_chs=embed_dim[i], reduction=stride, module=f'stages.{i}')] in_dim = embed_dim[i] self.stages = nn.Sequential(*stages) # Classifier head self.head = NormLinear(embed_dim[-1], num_classes, drop=drop_rate) if num_classes > 0 else nn.Identity() @torch.jit.ignore def no_weight_decay(self): return {x for x in self.state_dict().keys() if 'attention_biases' in x} @torch.jit.ignore def group_matcher(self, coarse=False): matcher = dict( stem=r'^cls_token|pos_embed|patch_embed', # stem and embed blocks=[(r'^blocks\.(\d+)', None), (r'^norm', (99999,))] ) return matcher @torch.jit.ignore def set_grad_checkpointing(self, enable=True): self.grad_checkpointing = enable @torch.jit.ignore def get_classifier(self): return self.head def reset_classifier(self, num_classes, global_pool=None, distillation=None): self.num_classes = num_classes if global_pool is not None: self.global_pool = global_pool self.head = NormLinear( self.embed_dim[-1], num_classes, drop=self.drop_rate) if num_classes > 0 else nn.Identity() def forward_features(self, x): x = self.stem(x) if not self.use_conv: x = x.flatten(2).transpose(1, 2) if self.grad_checkpointing and not torch.jit.is_scripting(): x = checkpoint_seq(self.stages, x) else: x = self.stages(x) return x def forward_head(self, x, pre_logits: bool = False): if self.global_pool == 'avg': x = x.mean(dim=(-2, -1)) if self.use_conv else x.mean(dim=1) return x if pre_logits else self.head(x) def forward(self, x): x = self.forward_features(x) x = self.forward_head(x) return x class LevitDistilled(Levit): def __init__(self, *args, **kwargs): super().__init__(*args, **kwargs) self.head_dist = NormLinear(self.num_features, self.num_classes) if self.num_classes > 0 else nn.Identity() self.distilled_training = False # must set this True to train w/ distillation token @torch.jit.ignore def get_classifier(self): return self.head, self.head_dist def reset_classifier(self, num_classes, global_pool=None, distillation=None): self.num_classes = num_classes if global_pool is not None: self.global_pool = global_pool self.head = NormLinear( self.num_features, num_classes, drop=self.drop_rate) if num_classes > 0 else nn.Identity() self.head_dist = NormLinear(self.num_features, num_classes) if num_classes > 0 else nn.Identity() @torch.jit.ignore def set_distilled_training(self, enable=True): self.distilled_training = enable def forward_head(self, x, pre_logits: bool = False): if self.global_pool == 'avg': x = x.mean(dim=(-2, -1)) if self.use_conv else x.mean(dim=1) if pre_logits: return x x, x_dist = self.head(x), self.head_dist(x) if self.distilled_training and self.training and not torch.jit.is_scripting(): # only return separate classification predictions when training in distilled mode return x, x_dist else: # during standard train/finetune, inference average the classifier predictions return (x + x_dist) / 2 def checkpoint_filter_fn(state_dict, model): if 'model' in state_dict: state_dict = state_dict['model'] # filter out attn biases, should not have been persistent state_dict = {k: v for k, v in state_dict.items() if 'attention_bias_idxs' not in k} D = model.state_dict() out_dict = {} for ka, kb, va, vb in zip(D.keys(), state_dict.keys(), D.values(), state_dict.values()): if va.ndim == 4 and vb.ndim == 2: vb = vb[:, :, None, None] if va.shape != vb.shape: # head or first-conv shapes may change for fine-tune assert 'head' in ka or 'stem.conv1.linear' in ka out_dict[ka] = vb return out_dict model_cfgs = dict( levit_128s=dict( embed_dim=(128, 256, 384), key_dim=16, num_heads=(4, 6, 8), depth=(2, 3, 4)), levit_128=dict( embed_dim=(128, 256, 384), key_dim=16, num_heads=(4, 8, 12), depth=(4, 4, 4)), levit_192=dict( embed_dim=(192, 288, 384), key_dim=32, num_heads=(3, 5, 6), depth=(4, 4, 4)), levit_256=dict( embed_dim=(256, 384, 512), key_dim=32, num_heads=(4, 6, 8), depth=(4, 4, 4)), levit_384=dict( embed_dim=(384, 512, 768), key_dim=32, num_heads=(6, 9, 12), depth=(4, 4, 4)), # stride-8 stem experiments levit_384_s8=dict( embed_dim=(384, 512, 768), key_dim=32, num_heads=(6, 9, 12), depth=(4, 4, 4), act_layer='silu', stem_type='s8'), levit_512_s8=dict( embed_dim=(512, 640, 896), key_dim=64, num_heads=(8, 10, 14), depth=(4, 4, 4), act_layer='silu', stem_type='s8'), # wider experiments levit_512=dict( embed_dim=(512, 768, 1024), key_dim=64, num_heads=(8, 12, 16), depth=(4, 4, 4), act_layer='silu'), # deeper experiments levit_256d=dict( embed_dim=(256, 384, 512), key_dim=32, num_heads=(4, 6, 8), depth=(4, 8, 6), act_layer='silu'), levit_512d=dict( embed_dim=(512, 640, 768), key_dim=64, num_heads=(8, 10, 12), depth=(4, 8, 6), act_layer='silu'), ) def create_levit(variant, cfg_variant=None, pretrained=False, distilled=True, **kwargs): is_conv = '_conv' in variant out_indices = kwargs.pop('out_indices', (0, 1, 2)) if kwargs.get('features_only', None): if not is_conv: raise RuntimeError('features_only not implemented for LeVit in non-convolutional mode.') if cfg_variant is None: if variant in model_cfgs: cfg_variant = variant elif is_conv: cfg_variant = variant.replace('_conv', '') model_cfg = dict(model_cfgs[cfg_variant], **kwargs) model = build_model_with_cfg( LevitDistilled if distilled else Levit, variant, pretrained, pretrained_filter_fn=checkpoint_filter_fn, feature_cfg=dict(flatten_sequential=True, out_indices=out_indices), **model_cfg, ) return model def _cfg(url='', **kwargs): return { 'url': url, 'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None, 'crop_pct': .9, 'interpolation': 'bicubic', 'fixed_input_size': True, 'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD, 'first_conv': 'stem.conv1.linear', 'classifier': ('head.linear', 'head_dist.linear'), **kwargs } default_cfgs = generate_default_cfgs({ # weights in nn.Linear mode 'levit_128s.fb_dist_in1k': _cfg( hf_hub_id='timm/', ), 'levit_128.fb_dist_in1k': _cfg( hf_hub_id='timm/', ), 'levit_192.fb_dist_in1k': _cfg( hf_hub_id='timm/', ), 'levit_256.fb_dist_in1k': _cfg( hf_hub_id='timm/', ), 'levit_384.fb_dist_in1k': _cfg( hf_hub_id='timm/', ), # weights in nn.Conv2d mode 'levit_conv_128s.fb_dist_in1k': _cfg( hf_hub_id='timm/', pool_size=(4, 4), ), 'levit_conv_128.fb_dist_in1k': _cfg( hf_hub_id='timm/', pool_size=(4, 4), ), 'levit_conv_192.fb_dist_in1k': _cfg( hf_hub_id='timm/', pool_size=(4, 4), ), 'levit_conv_256.fb_dist_in1k': _cfg( hf_hub_id='timm/', pool_size=(4, 4), ), 'levit_conv_384.fb_dist_in1k': _cfg( hf_hub_id='timm/', pool_size=(4, 4), ), 'levit_384_s8.untrained': _cfg(classifier='head.linear'), 'levit_512_s8.untrained': _cfg(classifier='head.linear'), 'levit_512.untrained': _cfg(classifier='head.linear'), 'levit_256d.untrained': _cfg(classifier='head.linear'), 'levit_512d.untrained': _cfg(classifier='head.linear'), 'levit_conv_384_s8.untrained': _cfg(classifier='head.linear'), 'levit_conv_512_s8.untrained': _cfg(classifier='head.linear'), 'levit_conv_512.untrained': _cfg(classifier='head.linear'), 'levit_conv_256d.untrained': _cfg(classifier='head.linear'), 'levit_conv_512d.untrained': _cfg(classifier='head.linear'), }) @register_model def levit_128s(pretrained=False, **kwargs) -> Levit: return create_levit('levit_128s', pretrained=pretrained, **kwargs) @register_model def levit_128(pretrained=False, **kwargs) -> Levit: return create_levit('levit_128', pretrained=pretrained, **kwargs) @register_model def levit_192(pretrained=False, **kwargs) -> Levit: return create_levit('levit_192', pretrained=pretrained, **kwargs) @register_model def levit_256(pretrained=False, **kwargs) -> Levit: return create_levit('levit_256', pretrained=pretrained, **kwargs) @register_model def levit_384(pretrained=False, **kwargs) -> Levit: return create_levit('levit_384', pretrained=pretrained, **kwargs) @register_model def levit_384_s8(pretrained=False, **kwargs) -> Levit: return create_levit('levit_384_s8', pretrained=pretrained, **kwargs) @register_model def levit_512_s8(pretrained=False, **kwargs) -> Levit: return create_levit('levit_512_s8', pretrained=pretrained, distilled=False, **kwargs) @register_model def levit_512(pretrained=False, **kwargs) -> Levit: return create_levit('levit_512', pretrained=pretrained, distilled=False, **kwargs) @register_model def levit_256d(pretrained=False, **kwargs) -> Levit: return create_levit('levit_256d', pretrained=pretrained, distilled=False, **kwargs) @register_model def levit_512d(pretrained=False, **kwargs) -> Levit: return create_levit('levit_512d', pretrained=pretrained, distilled=False, **kwargs) @register_model def levit_conv_128s(pretrained=False, **kwargs) -> Levit: return create_levit('levit_conv_128s', pretrained=pretrained, use_conv=True, **kwargs) @register_model def levit_conv_128(pretrained=False, **kwargs) -> Levit: return create_levit('levit_conv_128', pretrained=pretrained, use_conv=True, **kwargs) @register_model def levit_conv_192(pretrained=False, **kwargs) -> Levit: return create_levit('levit_conv_192', pretrained=pretrained, use_conv=True, **kwargs) @register_model def levit_conv_256(pretrained=False, **kwargs) -> Levit: return create_levit('levit_conv_256', pretrained=pretrained, use_conv=True, **kwargs) @register_model def levit_conv_384(pretrained=False, **kwargs) -> Levit: return create_levit('levit_conv_384', pretrained=pretrained, use_conv=True, **kwargs) @register_model def levit_conv_384_s8(pretrained=False, **kwargs) -> Levit: return create_levit('levit_conv_384_s8', pretrained=pretrained, use_conv=True, **kwargs) @register_model def levit_conv_512_s8(pretrained=False, **kwargs) -> Levit: return create_levit('levit_conv_512_s8', pretrained=pretrained, use_conv=True, distilled=False, **kwargs) @register_model def levit_conv_512(pretrained=False, **kwargs) -> Levit: return create_levit('levit_conv_512', pretrained=pretrained, use_conv=True, distilled=False, **kwargs) @register_model def levit_conv_256d(pretrained=False, **kwargs) -> Levit: return create_levit('levit_conv_256d', pretrained=pretrained, use_conv=True, distilled=False, **kwargs) @register_model def levit_conv_512d(pretrained=False, **kwargs) -> Levit: return create_levit('levit_conv_512d', pretrained=pretrained, use_conv=True, distilled=False, **kwargs)

pytorch-image-models/timm/models/levit.py/0

""" RepViT Paper: `RepViT: Revisiting Mobile CNN From ViT Perspective` - https://arxiv.org/abs/2307.09283 @misc{wang2023repvit, title={RepViT: Revisiting Mobile CNN From ViT Perspective}, author={Ao Wang and Hui Chen and Zijia Lin and Hengjun Pu and Guiguang Ding}, year={2023}, eprint={2307.09283}, archivePrefix={arXiv}, primaryClass={cs.CV} } Adapted from official impl at https://github.com/jameslahm/RepViT """ __all__ = ['RepVit'] import torch.nn as nn from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD from ._registry import register_model, generate_default_cfgs from ._builder import build_model_with_cfg from timm.layers import SqueezeExcite, trunc_normal_, to_ntuple, to_2tuple from ._manipulate import checkpoint_seq import torch class ConvNorm(nn.Sequential): def __init__(self, in_dim, out_dim, ks=1, stride=1, pad=0, dilation=1, groups=1, bn_weight_init=1): super().__init__() self.add_module('c', nn.Conv2d(in_dim, out_dim, ks, stride, pad, dilation, groups, bias=False)) self.add_module('bn', nn.BatchNorm2d(out_dim)) nn.init.constant_(self.bn.weight, bn_weight_init) nn.init.constant_(self.bn.bias, 0) @torch.no_grad() def fuse(self): c, bn = self._modules.values() w = bn.weight / (bn.running_var + bn.eps) ** 0.5 w = c.weight * w[:, None, None, None] b = bn.bias - bn.running_mean * bn.weight / (bn.running_var + bn.eps) ** 0.5 m = nn.Conv2d( w.size(1) * self.c.groups, w.size(0), w.shape[2:], stride=self.c.stride, padding=self.c.padding, dilation=self.c.dilation, groups=self.c.groups, device=c.weight.device, ) m.weight.data.copy_(w) m.bias.data.copy_(b) return m class NormLinear(nn.Sequential): def __init__(self, in_dim, out_dim, bias=True, std=0.02): super().__init__() self.add_module('bn', nn.BatchNorm1d(in_dim)) self.add_module('l', nn.Linear(in_dim, out_dim, bias=bias)) trunc_normal_(self.l.weight, std=std) if bias: nn.init.constant_(self.l.bias, 0) @torch.no_grad() def fuse(self): bn, l = self._modules.values() w = bn.weight / (bn.running_var + bn.eps) ** 0.5 b = bn.bias - self.bn.running_mean * self.bn.weight / (bn.running_var + bn.eps) ** 0.5 w = l.weight * w[None, :] if l.bias is None: b = b @ self.l.weight.T else: b = (l.weight @ b[:, None]).view(-1) + self.l.bias m = nn.Linear(w.size(1), w.size(0), device=l.weight.device) m.weight.data.copy_(w) m.bias.data.copy_(b) return m class RepVggDw(nn.Module): def __init__(self, ed, kernel_size, legacy=False): super().__init__() self.conv = ConvNorm(ed, ed, kernel_size, 1, (kernel_size - 1) // 2, groups=ed) if legacy: self.conv1 = ConvNorm(ed, ed, 1, 1, 0, groups=ed) # Make torchscript happy. self.bn = nn.Identity() else: self.conv1 = nn.Conv2d(ed, ed, 1, 1, 0, groups=ed) self.bn = nn.BatchNorm2d(ed) self.dim = ed self.legacy = legacy def forward(self, x): return self.bn(self.conv(x) + self.conv1(x) + x) @torch.no_grad() def fuse(self): conv = self.conv.fuse() if self.legacy: conv1 = self.conv1.fuse() else: conv1 = self.conv1 conv_w = conv.weight conv_b = conv.bias conv1_w = conv1.weight conv1_b = conv1.bias conv1_w = nn.functional.pad(conv1_w, [1, 1, 1, 1]) identity = nn.functional.pad( torch.ones(conv1_w.shape[0], conv1_w.shape[1], 1, 1, device=conv1_w.device), [1, 1, 1, 1] ) final_conv_w = conv_w + conv1_w + identity final_conv_b = conv_b + conv1_b conv.weight.data.copy_(final_conv_w) conv.bias.data.copy_(final_conv_b) if not self.legacy: bn = self.bn w = bn.weight / (bn.running_var + bn.eps) ** 0.5 w = conv.weight * w[:, None, None, None] b = bn.bias + (conv.bias - bn.running_mean) * bn.weight / (bn.running_var + bn.eps) ** 0.5 conv.weight.data.copy_(w) conv.bias.data.copy_(b) return conv class RepVitMlp(nn.Module): def __init__(self, in_dim, hidden_dim, act_layer): super().__init__() self.conv1 = ConvNorm(in_dim, hidden_dim, 1, 1, 0) self.act = act_layer() self.conv2 = ConvNorm(hidden_dim, in_dim, 1, 1, 0, bn_weight_init=0) def forward(self, x): return self.conv2(self.act(self.conv1(x))) class RepViTBlock(nn.Module): def __init__(self, in_dim, mlp_ratio, kernel_size, use_se, act_layer, legacy=False): super(RepViTBlock, self).__init__() self.token_mixer = RepVggDw(in_dim, kernel_size, legacy) self.se = SqueezeExcite(in_dim, 0.25) if use_se else nn.Identity() self.channel_mixer = RepVitMlp(in_dim, in_dim * mlp_ratio, act_layer) def forward(self, x): x = self.token_mixer(x) x = self.se(x) identity = x x = self.channel_mixer(x) return identity + x class RepVitStem(nn.Module): def __init__(self, in_chs, out_chs, act_layer): super().__init__() self.conv1 = ConvNorm(in_chs, out_chs // 2, 3, 2, 1) self.act1 = act_layer() self.conv2 = ConvNorm(out_chs // 2, out_chs, 3, 2, 1) self.stride = 4 def forward(self, x): return self.conv2(self.act1(self.conv1(x))) class RepVitDownsample(nn.Module): def __init__(self, in_dim, mlp_ratio, out_dim, kernel_size, act_layer, legacy=False): super().__init__() self.pre_block = RepViTBlock(in_dim, mlp_ratio, kernel_size, use_se=False, act_layer=act_layer, legacy=legacy) self.spatial_downsample = ConvNorm(in_dim, in_dim, kernel_size, 2, (kernel_size - 1) // 2, groups=in_dim) self.channel_downsample = ConvNorm(in_dim, out_dim, 1, 1) self.ffn = RepVitMlp(out_dim, out_dim * mlp_ratio, act_layer) def forward(self, x): x = self.pre_block(x) x = self.spatial_downsample(x) x = self.channel_downsample(x) identity = x x = self.ffn(x) return x + identity class RepVitClassifier(nn.Module): def __init__(self, dim, num_classes, distillation=False, drop=0.0): super().__init__() self.head_drop = nn.Dropout(drop) self.head = NormLinear(dim, num_classes) if num_classes > 0 else nn.Identity() self.distillation = distillation self.distilled_training = False self.num_classes = num_classes if distillation: self.head_dist = NormLinear(dim, num_classes) if num_classes > 0 else nn.Identity() def forward(self, x): x = self.head_drop(x) if self.distillation: x1, x2 = self.head(x), self.head_dist(x) if self.training and self.distilled_training and not torch.jit.is_scripting(): return x1, x2 else: return (x1 + x2) / 2 else: x = self.head(x) return x @torch.no_grad() def fuse(self): if not self.num_classes > 0: return nn.Identity() head = self.head.fuse() if self.distillation: head_dist = self.head_dist.fuse() head.weight += head_dist.weight head.bias += head_dist.bias head.weight /= 2 head.bias /= 2 return head else: return head class RepVitStage(nn.Module): def __init__(self, in_dim, out_dim, depth, mlp_ratio, act_layer, kernel_size=3, downsample=True, legacy=False): super().__init__() if downsample: self.downsample = RepVitDownsample(in_dim, mlp_ratio, out_dim, kernel_size, act_layer, legacy) else: assert in_dim == out_dim self.downsample = nn.Identity() blocks = [] use_se = True for _ in range(depth): blocks.append(RepViTBlock(out_dim, mlp_ratio, kernel_size, use_se, act_layer, legacy)) use_se = not use_se self.blocks = nn.Sequential(*blocks) def forward(self, x): x = self.downsample(x) x = self.blocks(x) return x class RepVit(nn.Module): def __init__( self, in_chans=3, img_size=224, embed_dim=(48,), depth=(2,), mlp_ratio=2, global_pool='avg', kernel_size=3, num_classes=1000, act_layer=nn.GELU, distillation=True, drop_rate=0.0, legacy=False, ): super(RepVit, self).__init__() self.grad_checkpointing = False self.global_pool = global_pool self.embed_dim = embed_dim self.num_classes = num_classes in_dim = embed_dim[0] self.stem = RepVitStem(in_chans, in_dim, act_layer) stride = self.stem.stride resolution = tuple([i // p for i, p in zip(to_2tuple(img_size), to_2tuple(stride))]) num_stages = len(embed_dim) mlp_ratios = to_ntuple(num_stages)(mlp_ratio) self.feature_info = [] stages = [] for i in range(num_stages): downsample = True if i != 0 else False stages.append( RepVitStage( in_dim, embed_dim[i], depth[i], mlp_ratio=mlp_ratios[i], act_layer=act_layer, kernel_size=kernel_size, downsample=downsample, legacy=legacy, ) ) stage_stride = 2 if downsample else 1 stride *= stage_stride resolution = tuple([(r - 1) // stage_stride + 1 for r in resolution]) self.feature_info += [dict(num_chs=embed_dim[i], reduction=stride, module=f'stages.{i}')] in_dim = embed_dim[i] self.stages = nn.Sequential(*stages) self.num_features = embed_dim[-1] self.head_drop = nn.Dropout(drop_rate) self.head = RepVitClassifier(embed_dim[-1], num_classes, distillation) @torch.jit.ignore def group_matcher(self, coarse=False): matcher = dict(stem=r'^stem', blocks=[(r'^blocks\.(\d+)', None), (r'^norm', (99999,))]) # stem and embed return matcher @torch.jit.ignore def set_grad_checkpointing(self, enable=True): self.grad_checkpointing = enable @torch.jit.ignore def get_classifier(self): return self.head def reset_classifier(self, num_classes, global_pool=None, distillation=False): self.num_classes = num_classes if global_pool is not None: self.global_pool = global_pool self.head = ( RepVitClassifier(self.embed_dim[-1], num_classes, distillation) if num_classes > 0 else nn.Identity() ) @torch.jit.ignore def set_distilled_training(self, enable=True): self.head.distilled_training = enable def forward_features(self, x): x = self.stem(x) if self.grad_checkpointing and not torch.jit.is_scripting(): x = checkpoint_seq(self.stages, x) else: x = self.stages(x) return x def forward_head(self, x, pre_logits: bool = False): if self.global_pool == 'avg': x = x.mean((2, 3), keepdim=False) x = self.head_drop(x) return self.head(x) def forward(self, x): x = self.forward_features(x) x = self.forward_head(x) return x @torch.no_grad() def fuse(self): def fuse_children(net): for child_name, child in net.named_children(): if hasattr(child, 'fuse'): fused = child.fuse() setattr(net, child_name, fused) fuse_children(fused) else: fuse_children(child) fuse_children(self) def _cfg(url='', **kwargs): return { 'url': url, 'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': (7, 7), 'crop_pct': 0.95, 'interpolation': 'bicubic', 'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD, 'first_conv': 'stem.conv1.c', 'classifier': ('head.head.l', 'head.head_dist.l'), **kwargs, } default_cfgs = generate_default_cfgs( { 'repvit_m1.dist_in1k': _cfg( hf_hub_id='timm/', ), 'repvit_m2.dist_in1k': _cfg( hf_hub_id='timm/', ), 'repvit_m3.dist_in1k': _cfg( hf_hub_id='timm/', ), 'repvit_m0_9.dist_300e_in1k': _cfg( hf_hub_id='timm/', ), 'repvit_m0_9.dist_450e_in1k': _cfg( hf_hub_id='timm/', ), 'repvit_m1_0.dist_300e_in1k': _cfg( hf_hub_id='timm/', ), 'repvit_m1_0.dist_450e_in1k': _cfg( hf_hub_id='timm/', ), 'repvit_m1_1.dist_300e_in1k': _cfg( hf_hub_id='timm/', ), 'repvit_m1_1.dist_450e_in1k': _cfg( hf_hub_id='timm/', ), 'repvit_m1_5.dist_300e_in1k': _cfg( hf_hub_id='timm/', ), 'repvit_m1_5.dist_450e_in1k': _cfg( hf_hub_id='timm/', ), 'repvit_m2_3.dist_300e_in1k': _cfg( hf_hub_id='timm/', ), 'repvit_m2_3.dist_450e_in1k': _cfg( hf_hub_id='timm/', ), } ) def _create_repvit(variant, pretrained=False, **kwargs): out_indices = kwargs.pop('out_indices', (0, 1, 2, 3)) model = build_model_with_cfg( RepVit, variant, pretrained, feature_cfg=dict(flatten_sequential=True, out_indices=out_indices), **kwargs, ) return model @register_model def repvit_m1(pretrained=False, **kwargs): """ Constructs a RepViT-M1 model """ model_args = dict(embed_dim=(48, 96, 192, 384), depth=(2, 2, 14, 2), legacy=True) return _create_repvit('repvit_m1', pretrained=pretrained, **dict(model_args, **kwargs)) @register_model def repvit_m2(pretrained=False, **kwargs): """ Constructs a RepViT-M2 model """ model_args = dict(embed_dim=(64, 128, 256, 512), depth=(2, 2, 12, 2), legacy=True) return _create_repvit('repvit_m2', pretrained=pretrained, **dict(model_args, **kwargs)) @register_model def repvit_m3(pretrained=False, **kwargs): """ Constructs a RepViT-M3 model """ model_args = dict(embed_dim=(64, 128, 256, 512), depth=(4, 4, 18, 2), legacy=True) return _create_repvit('repvit_m3', pretrained=pretrained, **dict(model_args, **kwargs)) @register_model def repvit_m0_9(pretrained=False, **kwargs): """ Constructs a RepViT-M0.9 model """ model_args = dict(embed_dim=(48, 96, 192, 384), depth=(2, 2, 14, 2)) return _create_repvit('repvit_m0_9', pretrained=pretrained, **dict(model_args, **kwargs)) @register_model def repvit_m1_0(pretrained=False, **kwargs): """ Constructs a RepViT-M1.0 model """ model_args = dict(embed_dim=(56, 112, 224, 448), depth=(2, 2, 14, 2)) return _create_repvit('repvit_m1_0', pretrained=pretrained, **dict(model_args, **kwargs)) @register_model def repvit_m1_1(pretrained=False, **kwargs): """ Constructs a RepViT-M1.1 model """ model_args = dict(embed_dim=(64, 128, 256, 512), depth=(2, 2, 12, 2)) return _create_repvit('repvit_m1_1', pretrained=pretrained, **dict(model_args, **kwargs)) @register_model def repvit_m1_5(pretrained=False, **kwargs): """ Constructs a RepViT-M1.5 model """ model_args = dict(embed_dim=(64, 128, 256, 512), depth=(4, 4, 24, 4)) return _create_repvit('repvit_m1_5', pretrained=pretrained, **dict(model_args, **kwargs)) @register_model def repvit_m2_3(pretrained=False, **kwargs): """ Constructs a RepViT-M2.3 model """ model_args = dict(embed_dim=(80, 160, 320, 640), depth=(6, 6, 34, 2)) return _create_repvit('repvit_m2_3', pretrained=pretrained, **dict(model_args, **kwargs))

pytorch-image-models/timm/models/repvit.py/0

""" Twins A PyTorch impl of : `Twins: Revisiting the Design of Spatial Attention in Vision Transformers` - https://arxiv.org/pdf/2104.13840.pdf Code/weights from https://github.com/Meituan-AutoML/Twins, original copyright/license info below """ # -------------------------------------------------------- # Twins # Copyright (c) 2021 Meituan # Licensed under The Apache 2.0 License [see LICENSE for details] # Written by Xinjie Li, Xiangxiang Chu # -------------------------------------------------------- import math from functools import partial from typing import Tuple import torch import torch.nn as nn import torch.nn.functional as F from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD from timm.layers import Mlp, DropPath, to_2tuple, trunc_normal_, use_fused_attn from ._builder import build_model_with_cfg from ._features_fx import register_notrace_module from ._registry import register_model, generate_default_cfgs from .vision_transformer import Attention __all__ = ['Twins'] # model_registry will add each entrypoint fn to this Size_ = Tuple[int, int] @register_notrace_module # reason: FX can't symbolically trace control flow in forward method class LocallyGroupedAttn(nn.Module): """ LSA: self attention within a group """ fused_attn: torch.jit.Final[bool] def __init__(self, dim, num_heads=8, attn_drop=0., proj_drop=0., ws=1): assert ws != 1 super(LocallyGroupedAttn, self).__init__() assert dim % num_heads == 0, f"dim {dim} should be divided by num_heads {num_heads}." self.dim = dim self.num_heads = num_heads head_dim = dim // num_heads self.scale = head_dim ** -0.5 self.fused_attn = use_fused_attn() self.qkv = nn.Linear(dim, dim * 3, bias=True) self.attn_drop = nn.Dropout(attn_drop) self.proj = nn.Linear(dim, dim) self.proj_drop = nn.Dropout(proj_drop) self.ws = ws def forward(self, x, size: Size_): # There are two implementations for this function, zero padding or mask. We don't observe obvious difference for # both. You can choose any one, we recommend forward_padding because it's neat. However, # the masking implementation is more reasonable and accurate. B, N, C = x.shape H, W = size x = x.view(B, H, W, C) pad_l = pad_t = 0 pad_r = (self.ws - W % self.ws) % self.ws pad_b = (self.ws - H % self.ws) % self.ws x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b)) _, Hp, Wp, _ = x.shape _h, _w = Hp // self.ws, Wp // self.ws x = x.reshape(B, _h, self.ws, _w, self.ws, C).transpose(2, 3) qkv = self.qkv(x).reshape( B, _h * _w, self.ws * self.ws, 3, self.num_heads, C // self.num_heads).permute(3, 0, 1, 4, 2, 5) q, k, v = qkv.unbind(0) if self.fused_attn: x = F.scaled_dot_product_attention( q, k, v, dropout_p=self.attn_drop.p if self.training else 0., ) else: q = q * self.scale attn = q @ k.transpose(-2, -1) attn = attn.softmax(dim=-1) attn = self.attn_drop(attn) x = attn @ v x = x.transpose(2, 3).reshape(B, _h, _w, self.ws, self.ws, C) x = x.transpose(2, 3).reshape(B, _h * self.ws, _w * self.ws, C) if pad_r > 0 or pad_b > 0: x = x[:, :H, :W, :].contiguous() x = x.reshape(B, N, C) x = self.proj(x) x = self.proj_drop(x) return x # def forward_mask(self, x, size: Size_): # B, N, C = x.shape # H, W = size # x = x.view(B, H, W, C) # pad_l = pad_t = 0 # pad_r = (self.ws - W % self.ws) % self.ws # pad_b = (self.ws - H % self.ws) % self.ws # x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b)) # _, Hp, Wp, _ = x.shape # _h, _w = Hp // self.ws, Wp // self.ws # mask = torch.zeros((1, Hp, Wp), device=x.device) # mask[:, -pad_b:, :].fill_(1) # mask[:, :, -pad_r:].fill_(1) # # x = x.reshape(B, _h, self.ws, _w, self.ws, C).transpose(2, 3) # B, _h, _w, ws, ws, C # mask = mask.reshape(1, _h, self.ws, _w, self.ws).transpose(2, 3).reshape(1, _h * _w, self.ws * self.ws) # attn_mask = mask.unsqueeze(2) - mask.unsqueeze(3) # 1, _h*_w, ws*ws, ws*ws # attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-1000.0)).masked_fill(attn_mask == 0, float(0.0)) # qkv = self.qkv(x).reshape( # B, _h * _w, self.ws * self.ws, 3, self.num_heads, C // self.num_heads).permute(3, 0, 1, 4, 2, 5) # # n_h, B, _w*_h, nhead, ws*ws, dim # q, k, v = qkv[0], qkv[1], qkv[2] # B, _h*_w, n_head, ws*ws, dim_head # attn = (q @ k.transpose(-2, -1)) * self.scale # B, _h*_w, n_head, ws*ws, ws*ws # attn = attn + attn_mask.unsqueeze(2) # attn = attn.softmax(dim=-1) # attn = self.attn_drop(attn) # attn @v -> B, _h*_w, n_head, ws*ws, dim_head # attn = (attn @ v).transpose(2, 3).reshape(B, _h, _w, self.ws, self.ws, C) # x = attn.transpose(2, 3).reshape(B, _h * self.ws, _w * self.ws, C) # if pad_r > 0 or pad_b > 0: # x = x[:, :H, :W, :].contiguous() # x = x.reshape(B, N, C) # x = self.proj(x) # x = self.proj_drop(x) # return x class GlobalSubSampleAttn(nn.Module): """ GSA: using a key to summarize the information for a group to be efficient. """ fused_attn: torch.jit.Final[bool] def __init__(self, dim, num_heads=8, attn_drop=0., proj_drop=0., sr_ratio=1): super().__init__() assert dim % num_heads == 0, f"dim {dim} should be divided by num_heads {num_heads}." self.dim = dim self.num_heads = num_heads head_dim = dim // num_heads self.scale = head_dim ** -0.5 self.fused_attn = use_fused_attn() self.q = nn.Linear(dim, dim, bias=True) self.kv = nn.Linear(dim, dim * 2, bias=True) self.attn_drop = nn.Dropout(attn_drop) self.proj = nn.Linear(dim, dim) self.proj_drop = nn.Dropout(proj_drop) self.sr_ratio = sr_ratio if sr_ratio > 1: self.sr = nn.Conv2d(dim, dim, kernel_size=sr_ratio, stride=sr_ratio) self.norm = nn.LayerNorm(dim) else: self.sr = None self.norm = None def forward(self, x, size: Size_): B, N, C = x.shape q = self.q(x).reshape(B, N, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3) if self.sr is not None: x = x.permute(0, 2, 1).reshape(B, C, *size) x = self.sr(x).reshape(B, C, -1).permute(0, 2, 1) x = self.norm(x) kv = self.kv(x).reshape(B, -1, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4) k, v = kv.unbind(0) if self.fused_attn: x = torch.nn.functional.scaled_dot_product_attention( q, k, v, dropout_p=self.attn_drop.p if self.training else 0., ) else: q = q * self.scale attn = q @ k.transpose(-2, -1) attn = attn.softmax(dim=-1) attn = self.attn_drop(attn) x = attn @ v x = x.transpose(1, 2).reshape(B, N, C) x = self.proj(x) x = self.proj_drop(x) return x class Block(nn.Module): def __init__( self, dim, num_heads, mlp_ratio=4., proj_drop=0., attn_drop=0., drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, sr_ratio=1, ws=None, ): super().__init__() self.norm1 = norm_layer(dim) if ws is None: self.attn = Attention(dim, num_heads, False, None, attn_drop, proj_drop) elif ws == 1: self.attn = GlobalSubSampleAttn(dim, num_heads, attn_drop, proj_drop, sr_ratio) else: self.attn = LocallyGroupedAttn(dim, num_heads, attn_drop, proj_drop, ws) self.drop_path1 = DropPath(drop_path) if drop_path > 0. else nn.Identity() self.norm2 = norm_layer(dim) self.mlp = Mlp( in_features=dim, hidden_features=int(dim * mlp_ratio), act_layer=act_layer, drop=proj_drop, ) self.drop_path2 = DropPath(drop_path) if drop_path > 0. else nn.Identity() def forward(self, x, size: Size_): x = x + self.drop_path1(self.attn(self.norm1(x), size)) x = x + self.drop_path2(self.mlp(self.norm2(x))) return x class PosConv(nn.Module): # PEG from https://arxiv.org/abs/2102.10882 def __init__(self, in_chans, embed_dim=768, stride=1): super(PosConv, self).__init__() self.proj = nn.Sequential( nn.Conv2d(in_chans, embed_dim, 3, stride, 1, bias=True, groups=embed_dim), ) self.stride = stride def forward(self, x, size: Size_): B, N, C = x.shape cnn_feat_token = x.transpose(1, 2).view(B, C, *size) x = self.proj(cnn_feat_token) if self.stride == 1: x += cnn_feat_token x = x.flatten(2).transpose(1, 2) return x def no_weight_decay(self): return ['proj.%d.weight' % i for i in range(4)] class PatchEmbed(nn.Module): """ Image to Patch Embedding """ def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768): super().__init__() img_size = to_2tuple(img_size) patch_size = to_2tuple(patch_size) self.img_size = img_size self.patch_size = patch_size assert img_size[0] % patch_size[0] == 0 and img_size[1] % patch_size[1] == 0, \ f"img_size {img_size} should be divided by patch_size {patch_size}." self.H, self.W = img_size[0] // patch_size[0], img_size[1] // patch_size[1] self.num_patches = self.H * self.W self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size) self.norm = nn.LayerNorm(embed_dim) def forward(self, x) -> Tuple[torch.Tensor, Size_]: B, C, H, W = x.shape x = self.proj(x).flatten(2).transpose(1, 2) x = self.norm(x) out_size = (H // self.patch_size[0], W // self.patch_size[1]) return x, out_size class Twins(nn.Module): """ Twins Vision Transfomer (Revisiting Spatial Attention) Adapted from PVT (PyramidVisionTransformer) class at https://github.com/whai362/PVT.git """ def __init__( self, img_size=224, patch_size=4, in_chans=3, num_classes=1000, global_pool='avg', embed_dims=(64, 128, 256, 512), num_heads=(1, 2, 4, 8), mlp_ratios=(4, 4, 4, 4), depths=(3, 4, 6, 3), sr_ratios=(8, 4, 2, 1), wss=None, drop_rate=0., pos_drop_rate=0., proj_drop_rate=0., attn_drop_rate=0., drop_path_rate=0., norm_layer=partial(nn.LayerNorm, eps=1e-6), block_cls=Block, ): super().__init__() self.num_classes = num_classes self.global_pool = global_pool self.depths = depths self.embed_dims = embed_dims self.num_features = embed_dims[-1] self.grad_checkpointing = False img_size = to_2tuple(img_size) prev_chs = in_chans self.patch_embeds = nn.ModuleList() self.pos_drops = nn.ModuleList() for i in range(len(depths)): self.patch_embeds.append(PatchEmbed(img_size, patch_size, prev_chs, embed_dims[i])) self.pos_drops.append(nn.Dropout(p=pos_drop_rate)) prev_chs = embed_dims[i] img_size = tuple(t // patch_size for t in img_size) patch_size = 2 self.blocks = nn.ModuleList() dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] # stochastic depth decay rule cur = 0 for k in range(len(depths)): _block = nn.ModuleList([block_cls( dim=embed_dims[k], num_heads=num_heads[k], mlp_ratio=mlp_ratios[k], proj_drop=proj_drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer, sr_ratio=sr_ratios[k], ws=1 if wss is None or i % 2 == 1 else wss[k]) for i in range(depths[k])], ) self.blocks.append(_block) cur += depths[k] self.pos_block = nn.ModuleList([PosConv(embed_dim, embed_dim) for embed_dim in embed_dims]) self.norm = norm_layer(self.num_features) # classification head self.head_drop = nn.Dropout(drop_rate) self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity() # init weights self.apply(self._init_weights) @torch.jit.ignore def no_weight_decay(self): return set(['pos_block.' + n for n, p in self.pos_block.named_parameters()]) @torch.jit.ignore def group_matcher(self, coarse=False): matcher = dict( stem=r'^patch_embeds.0', # stem and embed blocks=[ (r'^(?:blocks|patch_embeds|pos_block)\.(\d+)', None), ('^norm', (99999,)) ] if coarse else [ (r'^blocks\.(\d+)\.(\d+)', None), (r'^(?:patch_embeds|pos_block)\.(\d+)', (0,)), (r'^norm', (99999,)) ] ) return matcher @torch.jit.ignore def set_grad_checkpointing(self, enable=True): assert not enable, 'gradient checkpointing not supported' @torch.jit.ignore def get_classifier(self): return self.head def reset_classifier(self, num_classes, global_pool=None): self.num_classes = num_classes if global_pool is not None: assert global_pool in ('', 'avg') self.global_pool = global_pool self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity() def _init_weights(self, m): if isinstance(m, nn.Linear): trunc_normal_(m.weight, std=.02) if isinstance(m, nn.Linear) and m.bias is not None: nn.init.constant_(m.bias, 0) elif isinstance(m, nn.LayerNorm): nn.init.constant_(m.bias, 0) nn.init.constant_(m.weight, 1.0) elif isinstance(m, nn.Conv2d): fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels fan_out //= m.groups m.weight.data.normal_(0, math.sqrt(2.0 / fan_out)) if m.bias is not None: m.bias.data.zero_() def forward_features(self, x): B = x.shape[0] for i, (embed, drop, blocks, pos_blk) in enumerate( zip(self.patch_embeds, self.pos_drops, self.blocks, self.pos_block)): x, size = embed(x) x = drop(x) for j, blk in enumerate(blocks): x = blk(x, size) if j == 0: x = pos_blk(x, size) # PEG here if i < len(self.depths) - 1: x = x.reshape(B, *size, -1).permute(0, 3, 1, 2).contiguous() x = self.norm(x) return x def forward_head(self, x, pre_logits: bool = False): if self.global_pool == 'avg': x = x.mean(dim=1) x = self.head_drop(x) return x if pre_logits else self.head(x) def forward(self, x): x = self.forward_features(x) x = self.forward_head(x) return x def _create_twins(variant, pretrained=False, **kwargs): if kwargs.get('features_only', None): raise RuntimeError('features_only not implemented for Vision Transformer models.') model = build_model_with_cfg(Twins, variant, pretrained, **kwargs) return model def _cfg(url='', **kwargs): return { 'url': url, 'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None, 'crop_pct': .9, 'interpolation': 'bicubic', 'fixed_input_size': True, 'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD, 'first_conv': 'patch_embeds.0.proj', 'classifier': 'head', **kwargs } default_cfgs = generate_default_cfgs({ 'twins_pcpvt_small.in1k': _cfg(hf_hub_id='timm/'), 'twins_pcpvt_base.in1k': _cfg(hf_hub_id='timm/'), 'twins_pcpvt_large.in1k': _cfg(hf_hub_id='timm/'), 'twins_svt_small.in1k': _cfg(hf_hub_id='timm/'), 'twins_svt_base.in1k': _cfg(hf_hub_id='timm/'), 'twins_svt_large.in1k': _cfg(hf_hub_id='timm/'), }) @register_model def twins_pcpvt_small(pretrained=False, **kwargs) -> Twins: model_args = dict( patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1]) return _create_twins('twins_pcpvt_small', pretrained=pretrained, **dict(model_args, **kwargs)) @register_model def twins_pcpvt_base(pretrained=False, **kwargs) -> Twins: model_args = dict( patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], depths=[3, 4, 18, 3], sr_ratios=[8, 4, 2, 1]) return _create_twins('twins_pcpvt_base', pretrained=pretrained, **dict(model_args, **kwargs)) @register_model def twins_pcpvt_large(pretrained=False, **kwargs) -> Twins: model_args = dict( patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], depths=[3, 8, 27, 3], sr_ratios=[8, 4, 2, 1]) return _create_twins('twins_pcpvt_large', pretrained=pretrained, **dict(model_args, **kwargs)) @register_model def twins_svt_small(pretrained=False, **kwargs) -> Twins: model_args = dict( patch_size=4, embed_dims=[64, 128, 256, 512], num_heads=[2, 4, 8, 16], mlp_ratios=[4, 4, 4, 4], depths=[2, 2, 10, 4], wss=[7, 7, 7, 7], sr_ratios=[8, 4, 2, 1]) return _create_twins('twins_svt_small', pretrained=pretrained, **dict(model_args, **kwargs)) @register_model def twins_svt_base(pretrained=False, **kwargs) -> Twins: model_args = dict( patch_size=4, embed_dims=[96, 192, 384, 768], num_heads=[3, 6, 12, 24], mlp_ratios=[4, 4, 4, 4], depths=[2, 2, 18, 2], wss=[7, 7, 7, 7], sr_ratios=[8, 4, 2, 1]) return _create_twins('twins_svt_base', pretrained=pretrained, **dict(model_args, **kwargs)) @register_model def twins_svt_large(pretrained=False, **kwargs) -> Twins: model_args = dict( patch_size=4, embed_dims=[128, 256, 512, 1024], num_heads=[4, 8, 16, 32], mlp_ratios=[4, 4, 4, 4], depths=[2, 2, 18, 2], wss=[7, 7, 7, 7], sr_ratios=[8, 4, 2, 1]) return _create_twins('twins_svt_large', pretrained=pretrained, **dict(model_args, **kwargs))

pytorch-image-models/timm/models/twins.py/0

""" AdamP Optimizer Implementation copied from https://github.com/clovaai/AdamP/blob/master/adamp/adamp.py Paper: `Slowing Down the Weight Norm Increase in Momentum-based Optimizers` - https://arxiv.org/abs/2006.08217 Code: https://github.com/clovaai/AdamP Copyright (c) 2020-present NAVER Corp. MIT license """ import torch import torch.nn.functional as F from torch.optim.optimizer import Optimizer import math def _channel_view(x) -> torch.Tensor: return x.reshape(x.size(0), -1) def _layer_view(x) -> torch.Tensor: return x.reshape(1, -1) def projection(p, grad, perturb, delta: float, wd_ratio: float, eps: float): wd = 1. expand_size = (-1,) + (1,) * (len(p.shape) - 1) for view_func in [_channel_view, _layer_view]: param_view = view_func(p) grad_view = view_func(grad) cosine_sim = F.cosine_similarity(grad_view, param_view, dim=1, eps=eps).abs_() # FIXME this is a problem for PyTorch XLA if cosine_sim.max() < delta / math.sqrt(param_view.size(1)): p_n = p / param_view.norm(p=2, dim=1).add_(eps).reshape(expand_size) perturb -= p_n * view_func(p_n * perturb).sum(dim=1).reshape(expand_size) wd = wd_ratio return perturb, wd return perturb, wd class AdamP(Optimizer): def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-8, weight_decay=0, delta=0.1, wd_ratio=0.1, nesterov=False): defaults = dict( lr=lr, betas=betas, eps=eps, weight_decay=weight_decay, delta=delta, wd_ratio=wd_ratio, nesterov=nesterov) super(AdamP, self).__init__(params, defaults) @torch.no_grad() def step(self, closure=None): loss = None if closure is not None: with torch.enable_grad(): loss = closure() for group in self.param_groups: for p in group['params']: if p.grad is None: continue grad = p.grad beta1, beta2 = group['betas'] nesterov = group['nesterov'] state = self.state[p] # State initialization if len(state) == 0: state['step'] = 0 state['exp_avg'] = torch.zeros_like(p) state['exp_avg_sq'] = torch.zeros_like(p) # Adam exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq'] state['step'] += 1 bias_correction1 = 1 - beta1 ** state['step'] bias_correction2 = 1 - beta2 ** state['step'] exp_avg.mul_(beta1).add_(grad, alpha=1 - beta1) exp_avg_sq.mul_(beta2).addcmul_(grad, grad, value=1 - beta2) denom = (exp_avg_sq.sqrt() / math.sqrt(bias_correction2)).add_(group['eps']) step_size = group['lr'] / bias_correction1 if nesterov: perturb = (beta1 * exp_avg + (1 - beta1) * grad) / denom else: perturb = exp_avg / denom # Projection wd_ratio = 1. if len(p.shape) > 1: perturb, wd_ratio = projection(p, grad, perturb, group['delta'], group['wd_ratio'], group['eps']) # Weight decay if group['weight_decay'] > 0: p.mul_(1. - group['lr'] * group['weight_decay'] * wd_ratio) # Step p.add_(perturb, alpha=-step_size) return loss

pytorch-image-models/timm/optim/adamp.py/0

from .cosine_lr import CosineLRScheduler from .multistep_lr import MultiStepLRScheduler from .plateau_lr import PlateauLRScheduler from .poly_lr import PolyLRScheduler from .step_lr import StepLRScheduler from .tanh_lr import TanhLRScheduler from .scheduler_factory import create_scheduler, create_scheduler_v2, scheduler_kwargs

pytorch-image-models/timm/scheduler/__init__.py/0

""" JIT scripting/tracing utils Hacked together by / Copyright 2020 Ross Wightman """ import os import torch def set_jit_legacy(): """ Set JIT executor to legacy w/ support for op fusion This is hopefully a temporary need in 1.5/1.5.1/1.6 to restore performance due to changes in the JIT exectutor. These API are not supported so could change. """ # assert hasattr(torch._C, '_jit_set_profiling_executor'), "Old JIT behavior doesn't exist!" torch._C._jit_set_profiling_executor(False) torch._C._jit_set_profiling_mode(False) torch._C._jit_override_can_fuse_on_gpu(True) #torch._C._jit_set_texpr_fuser_enabled(True) def set_jit_fuser(fuser): if fuser == "te": # default fuser should be == 'te' torch._C._jit_set_profiling_executor(True) torch._C._jit_set_profiling_mode(True) torch._C._jit_override_can_fuse_on_cpu(False) torch._C._jit_override_can_fuse_on_gpu(True) torch._C._jit_set_texpr_fuser_enabled(True) try: torch._C._jit_set_nvfuser_enabled(False) except Exception: pass elif fuser == "old" or fuser == "legacy": torch._C._jit_set_profiling_executor(False) torch._C._jit_set_profiling_mode(False) torch._C._jit_override_can_fuse_on_gpu(True) torch._C._jit_set_texpr_fuser_enabled(False) try: torch._C._jit_set_nvfuser_enabled(False) except Exception: pass elif fuser == "nvfuser" or fuser == "nvf": os.environ['PYTORCH_NVFUSER_DISABLE_FALLBACK'] = '1' #os.environ['PYTORCH_NVFUSER_DISABLE_FMA'] = '1' #os.environ['PYTORCH_NVFUSER_JIT_OPT_LEVEL'] = '0' torch._C._jit_set_texpr_fuser_enabled(False) torch._C._jit_set_profiling_executor(True) torch._C._jit_set_profiling_mode(True) torch._C._jit_can_fuse_on_cpu() torch._C._jit_can_fuse_on_gpu() torch._C._jit_override_can_fuse_on_cpu(False) torch._C._jit_override_can_fuse_on_gpu(False) torch._C._jit_set_nvfuser_guard_mode(True) torch._C._jit_set_nvfuser_enabled(True) else: assert False, f"Invalid jit fuser ({fuser})"

pytorch-image-models/timm/utils/jit.py/0

/// Inspired by https://github.com/hatoo/oha/blob/bb989ea3cd77727e7743e7daa60a19894bb5e901/src/monitor.rs use crate::generation::{Decode, Message, Prefill}; use crossterm::event::{KeyCode, KeyEvent, KeyModifiers}; use text_generation_client::ClientError; use tokio::sync::mpsc; use tui::backend::Backend; use tui::layout::{Alignment, Constraint, Direction, Layout}; use tui::style::{Color, Modifier, Style}; use tui::text::{Line, Span}; use tui::widgets::{ Axis, BarChart, Block, Borders, Chart, Dataset, Gauge, GraphType, Paragraph, Tabs, }; use tui::{symbols, Frame}; /// TUI powered App pub(crate) struct App { pub(crate) running: bool, pub(crate) data: Data, completed_runs: Vec<usize>, completed_batch: usize, current_batch: usize, current_tab: usize, touched_tab: bool, zoom: bool, is_error: bool, tokenizer_name: String, sequence_length: u32, decode_length: u32, n_run: usize, receiver: mpsc::Receiver<Result<Message, ClientError>>, } impl App { pub(crate) fn new( receiver: mpsc::Receiver<Result<Message, ClientError>>, tokenizer_name: String, sequence_length: u32, decode_length: u32, n_run: usize, batch_size: Vec<u32>, ) -> Self { let current_tab = 0; let completed_runs: Vec<usize> = (0..batch_size.len()).map(|_| 0).collect(); let completed_batch = 0; let current_batch = 0; let is_error = false; let data = Data::new(n_run, batch_size); Self { running: true, data, completed_runs, completed_batch, current_batch, current_tab, touched_tab: false, zoom: false, is_error, tokenizer_name, sequence_length, decode_length, n_run, receiver, } } /// Handle crossterm key events pub(crate) fn handle_key_event(&mut self, key_event: KeyEvent) { match key_event { // Increase and wrap tab KeyEvent { code: KeyCode::Right, .. } | KeyEvent { code: KeyCode::Tab, .. } => { self.touched_tab = true; self.current_tab = (self.current_tab + 1) % self.data.batch_size.len(); } // Decrease and wrap tab KeyEvent { code: KeyCode::Left, .. } => { self.touched_tab = true; if self.current_tab > 0 { self.current_tab -= 1; } else { self.current_tab = self.data.batch_size.len() - 1; } } // Zoom on throughput/latency fig KeyEvent { code: KeyCode::Char('+'), .. } => { self.zoom = true; } // Unzoom on throughput/latency fig KeyEvent { code: KeyCode::Char('-'), .. } => { self.zoom = false; } // Quit KeyEvent { code: KeyCode::Char('q'), .. } | KeyEvent { code: KeyCode::Char('c'), modifiers: KeyModifiers::CONTROL, .. } => { self.running = false; } _ => (), } } /// Get all pending messages from generation task pub(crate) fn tick(&mut self) { while let Ok(message) = self.receiver.try_recv() { match message { Ok(message) => match message { Message::Prefill(step) => self.data.push_prefill(step, self.current_batch), Message::Decode(step) => self.data.push_decode(step, self.current_batch), Message::EndRun => { self.completed_runs[self.current_batch] += 1; } Message::EndBatch => { self.data.end_batch(self.current_batch); self.completed_batch += 1; if self.current_batch < self.data.batch_size.len() - 1 { // Only go to next tab if the user never touched the tab keys if !self.touched_tab { self.current_tab += 1; } self.current_batch += 1; } } Message::Warmup => {} }, Err(_) => self.is_error = true, } } } /// Render frame pub fn render<B: Backend>(&mut self, f: &mut Frame<'_, B>) { let batch_progress = (self.completed_batch as f64 / self.data.batch_size.len() as f64).clamp(0.0, 1.0); let run_progress = (self.completed_runs[self.current_batch] as f64 / self.n_run as f64).clamp(0.0, 1.0); // Vertical layout let row5 = Layout::default() .direction(Direction::Vertical) .constraints( [ Constraint::Length(1), Constraint::Length(3), Constraint::Length(3), Constraint::Length(13), Constraint::Min(10), ] .as_ref(), ) .split(f.size()); // Top row horizontal layout let top = Layout::default() .direction(Direction::Horizontal) .constraints([Constraint::Percentage(50), Constraint::Percentage(50)].as_ref()) .split(row5[2]); // Mid row horizontal layout let mid = Layout::default() .direction(Direction::Horizontal) .constraints( [ Constraint::Percentage(25), Constraint::Percentage(25), Constraint::Percentage(25), Constraint::Percentage(25), ] .as_ref(), ) .split(row5[3]); // Left mid row vertical layout let prefill_text = Layout::default() .direction(Direction::Vertical) .constraints([Constraint::Length(8), Constraint::Length(5)].as_ref()) .split(mid[0]); // Right mid row vertical layout let decode_text = Layout::default() .direction(Direction::Vertical) .constraints([Constraint::Length(8), Constraint::Length(5)].as_ref()) .split(mid[2]); let decode_text_latency = Layout::default() .direction(Direction::Horizontal) .constraints([Constraint::Percentage(50), Constraint::Percentage(50)].as_ref()) .split(decode_text[0]); // Bottom row horizontal layout let bottom = Layout::default() .direction(Direction::Horizontal) .constraints([Constraint::Percentage(50), Constraint::Percentage(50)].as_ref()) .split(row5[4]); // Title let title = Block::default() .borders(Borders::NONE) .title(format!( "Model: {} | Sequence Length: {} | Decode Length: {}", self.tokenizer_name, self.sequence_length, self.decode_length )) .style( Style::default() .add_modifier(Modifier::BOLD) .fg(Color::White), ); f.render_widget(title, row5[0]); // Helper let helper = Block::default() .borders(Borders::NONE) .title("<- | tab | ->: change batch tab | q / CTRL + c: quit | +/-: zoom") .title_alignment(Alignment::Right) .style(Style::default().fg(Color::White)); f.render_widget(helper, row5[0]); // Batch tabs let titles = self .data .batch_size .iter() .map(|b| { Line::from(vec![Span::styled( format!("Batch: {b}"), Style::default().fg(Color::White), )]) }) .collect(); let tabs = Tabs::new(titles) .block(Block::default().borders(Borders::ALL).title("Tabs")) .select(self.current_tab) .style(Style::default().fg(Color::LightCyan)) .highlight_style( Style::default() .add_modifier(Modifier::BOLD) .bg(Color::Black), ); f.render_widget(tabs, row5[1]); // Total progress bar let color = if self.is_error { Color::Red } else { Color::LightGreen }; let batch_gauge = progress_gauge( "Total Progress", format!("{} / {}", self.completed_batch, self.data.batch_size.len()), batch_progress, color, ); f.render_widget(batch_gauge, top[0]); // Batch progress Bar let color = if self.is_error { Color::Red } else { Color::LightBlue }; let run_gauge = progress_gauge( "Batch Progress", format!( "{} / {}", self.completed_runs[self.current_batch], self.n_run ), run_progress, color, ); f.render_widget(run_gauge, top[1]); // Prefill text infos let prefill_latency_block = latency_paragraph( &mut self.data.prefill_latencies[self.current_tab], "Prefill", ); let prefill_throughput_block = throughput_paragraph(&self.data.prefill_throughputs[self.current_tab], "Prefill"); f.render_widget(prefill_latency_block, prefill_text[0]); f.render_widget(prefill_throughput_block, prefill_text[1]); // Prefill latency histogram let histo_width = 7; let bins = if mid[1].width < 2 { 0 } else { (mid[1].width as usize - 2) / (histo_width + 1) } .max(2); let histo_data = latency_histogram_data(&self.data.prefill_latencies[self.current_tab], bins); let histo_data_str: Vec<(&str, u64)> = histo_data.iter().map(|(l, v)| (l.as_str(), *v)).collect(); let prefill_histogram = latency_histogram(&histo_data_str, "Prefill").bar_width(histo_width as u16); f.render_widget(prefill_histogram, mid[1]); // Decode text info let decode_latency_block = latency_paragraph( &mut self.data.decode_latencies[self.current_tab], "Decode Total", ); let decode_token_latency_block = latency_paragraph( &mut self.data.decode_token_latencies[self.current_tab], "Decode Token", ); let decode_throughput_block = throughput_paragraph(&self.data.decode_throughputs[self.current_tab], "Decode"); f.render_widget(decode_latency_block, decode_text_latency[0]); f.render_widget(decode_token_latency_block, decode_text_latency[1]); f.render_widget(decode_throughput_block, decode_text[1]); // Decode latency histogram let histo_data = latency_histogram_data(&self.data.decode_latencies[self.current_tab], bins); let histo_data_str: Vec<(&str, u64)> = histo_data.iter().map(|(l, v)| (l.as_str(), *v)).collect(); let decode_histogram = latency_histogram(&histo_data_str, "Decode").bar_width(histo_width as u16); f.render_widget(decode_histogram, mid[3]); // Prefill latency/throughput chart let prefill_latency_throughput_chart = latency_throughput_chart( &self.data.prefill_batch_latency_throughput, &self.data.batch_size, self.zoom, "Prefill", ); f.render_widget(prefill_latency_throughput_chart, bottom[0]); // Decode latency/throughput chart let decode_latency_throughput_chart = latency_throughput_chart( &self.data.decode_batch_latency_throughput, &self.data.batch_size, self.zoom, "Decode", ); f.render_widget(decode_latency_throughput_chart, bottom[1]); } } /// App internal data struct pub(crate) struct Data { pub(crate) batch_size: Vec<u32>, pub(crate) prefill_latencies: Vec<Vec<f64>>, pub(crate) prefill_throughputs: Vec<Vec<f64>>, pub(crate) decode_latencies: Vec<Vec<f64>>, pub(crate) decode_token_latencies: Vec<Vec<f64>>, pub(crate) decode_throughputs: Vec<Vec<f64>>, pub(crate) prefill_batch_latency_throughput: Vec<(f64, f64)>, pub(crate) decode_batch_latency_throughput: Vec<(f64, f64)>, } impl Data { fn new(n_run: usize, batch_size: Vec<u32>) -> Self { let prefill_latencies: Vec<Vec<f64>> = (0..batch_size.len()) .map(|_| Vec::with_capacity(n_run)) .collect(); let prefill_throughputs: Vec<Vec<f64>> = prefill_latencies.clone(); let decode_latencies: Vec<Vec<f64>> = prefill_latencies.clone(); let decode_token_latencies: Vec<Vec<f64>> = decode_latencies.clone(); let decode_throughputs: Vec<Vec<f64>> = prefill_throughputs.clone(); let prefill_batch_latency_throughput: Vec<(f64, f64)> = Vec::with_capacity(batch_size.len()); let decode_batch_latency_throughput: Vec<(f64, f64)> = prefill_batch_latency_throughput.clone(); Self { batch_size, prefill_latencies, prefill_throughputs, decode_latencies, decode_token_latencies, decode_throughputs, prefill_batch_latency_throughput, decode_batch_latency_throughput, } } fn push_prefill(&mut self, prefill: Prefill, batch_idx: usize) { let latency = prefill.latency.as_micros() as f64 / 1000.0; self.prefill_latencies[batch_idx].push(latency); self.prefill_throughputs[batch_idx].push(prefill.throughput); } fn push_decode(&mut self, decode: Decode, batch_idx: usize) { let latency = decode.latency.as_micros() as f64 / 1000.0; let token_latency = decode.token_latency.as_micros() as f64 / 1000.0; self.decode_latencies[batch_idx].push(latency); self.decode_token_latencies[batch_idx].push(token_latency); self.decode_throughputs[batch_idx].push(decode.throughput); } fn end_batch(&mut self, batch_idx: usize) { self.prefill_batch_latency_throughput.push(( self.prefill_latencies[batch_idx].iter().sum::<f64>() / self.prefill_latencies[batch_idx].len() as f64, self.prefill_throughputs[batch_idx].iter().sum::<f64>() / self.prefill_throughputs[batch_idx].len() as f64, )); self.decode_batch_latency_throughput.push(( self.decode_latencies[batch_idx].iter().sum::<f64>() / self.decode_latencies[batch_idx].len() as f64, self.decode_throughputs[batch_idx].iter().sum::<f64>() / self.decode_throughputs[batch_idx].len() as f64, )); } } /// Progress bar fn progress_gauge(title: &str, label: String, progress: f64, color: Color) -> Gauge { Gauge::default() .block(Block::default().title(title).borders(Borders::ALL)) .gauge_style(Style::default().fg(color)) .label(Span::raw(label)) .ratio(progress) } /// Throughput paragraph fn throughput_paragraph<'a>(throughput: &Vec<f64>, name: &'static str) -> Paragraph<'a> { // Throughput average/high/low texts let throughput_texts = statis_spans(throughput, "tokens/secs"); // Throughput block Paragraph::new(throughput_texts).block( Block::default() .title(Span::raw(format!("{name} Throughput"))) .borders(Borders::ALL), ) } /// Latency paragraph fn latency_paragraph<'a>(latency: &mut Vec<f64>, name: &'static str) -> Paragraph<'a> { // Latency average/high/low texts let mut latency_texts = statis_spans(latency, "ms"); // Sort latency for percentiles float_ord::sort(latency); let latency_percentiles = crate::utils::percentiles(latency, &[50, 90, 99]); // Latency p50/p90/p99 texts let colors = [Color::LightGreen, Color::LightYellow, Color::LightRed]; for (i, (name, value)) in latency_percentiles.iter().enumerate() { let span = Line::from(vec![Span::styled( format!("{name}: {value:.2} ms"), Style::default().fg(colors[i]), )]); latency_texts.push(span); } Paragraph::new(latency_texts).block( Block::default() .title(Span::raw(format!("{name} Latency"))) .borders(Borders::ALL), ) } /// Average/High/Low spans fn statis_spans<'a>(data: &Vec<f64>, unit: &'static str) -> Vec<Line<'a>> { vec![ Line::from(vec![Span::styled( format!( "Average: {:.2} {unit}", data.iter().sum::<f64>() / data.len() as f64 ), Style::default().fg(Color::LightBlue), )]), Line::from(vec![Span::styled( format!( "Lowest: {:.2} {unit}", data.iter() .min_by(|a, b| a.total_cmp(b)) .unwrap_or(&std::f64::NAN) ), Style::default().fg(Color::Reset), )]), Line::from(vec![Span::styled( format!( "Highest: {:.2} {unit}", data.iter() .max_by(|a, b| a.total_cmp(b)) .unwrap_or(&std::f64::NAN) ), Style::default().fg(Color::Reset), )]), ] } /// Latency histogram data fn latency_histogram_data(latency: &[f64], bins: usize) -> Vec<(String, u64)> { let histo_data: Vec<(String, u64)> = { let histo = crate::utils::histogram(latency, bins); histo .into_iter() .map(|(label, v)| (format!("{label:.2}"), v as u64)) .collect() }; histo_data } /// Latency Histogram fn latency_histogram<'a>( histo_data_str: &'a Vec<(&'a str, u64)>, name: &'static str, ) -> BarChart<'a> { BarChart::default() .block( Block::default() .title(format!("{name} latency histogram")) .style(Style::default().fg(Color::LightYellow).bg(Color::Reset)) .borders(Borders::ALL), ) .data(histo_data_str.as_slice()) } /// Latency/Throughput chart fn latency_throughput_chart<'a>( latency_throughput: &'a Vec<(f64, f64)>, batch_sizes: &'a [u32], zoom: bool, name: &'static str, ) -> Chart<'a> { let latency_iter = latency_throughput.iter().map(|(l, _)| l); let throughput_iter = latency_throughput.iter().map(|(_, t)| t); // Get extreme values let min_latency: f64 = *latency_iter .clone() .min_by(|a, b| a.total_cmp(b)) .unwrap_or(&std::f64::NAN); let max_latency: f64 = *latency_iter .max_by(|a, b| a.total_cmp(b)) .unwrap_or(&std::f64::NAN); let min_throughput: f64 = *throughput_iter .clone() .min_by(|a, b| a.total_cmp(b)) .unwrap_or(&std::f64::NAN); let max_throughput: f64 = *throughput_iter .max_by(|a, b| a.total_cmp(b)) .unwrap_or(&std::f64::NAN); // Char min max values let min_x = if zoom { ((min_latency - 0.05 * min_latency) / 100.0).floor() * 100.0 } else { 0.0 }; let max_x = ((max_latency + 0.05 * max_latency) / 100.0).ceil() * 100.0; let step_x = (max_x - min_x) / 4.0; // Chart min max values let min_y = if zoom { ((min_throughput - 0.05 * min_throughput) / 100.0).floor() * 100.0 } else { 0.0 }; let max_y = ((max_throughput + 0.05 * max_throughput) / 100.0).ceil() * 100.0; let step_y = (max_y - min_y) / 4.0; // Labels let mut x_labels = vec![Span::styled( format!("{min_x:.2}"), Style::default() .add_modifier(Modifier::BOLD) .fg(Color::Gray) .bg(Color::Reset), )]; for i in 0..3 { x_labels.push(Span::styled( format!("{:.2}", min_x + ((i + 1) as f64 * step_x)), Style::default().fg(Color::Gray).bg(Color::Reset), )); } x_labels.push(Span::styled( format!("{max_x:.2}"), Style::default() .add_modifier(Modifier::BOLD) .fg(Color::Gray) .bg(Color::Reset), )); // Labels let mut y_labels = vec![Span::styled( format!("{min_y:.2}"), Style::default() .add_modifier(Modifier::BOLD) .fg(Color::Gray) .bg(Color::Reset), )]; for i in 0..3 { y_labels.push(Span::styled( format!("{:.2}", min_y + ((i + 1) as f64 * step_y)), Style::default().fg(Color::Gray).bg(Color::Reset), )); } y_labels.push(Span::styled( format!("{max_y:.2}"), Style::default() .add_modifier(Modifier::BOLD) .fg(Color::Gray) .bg(Color::Reset), )); // Chart dataset let colors = color_vec(); let datasets: Vec<Dataset> = (0..latency_throughput.len()) .map(|i| { let color_idx = i % colors.len(); Dataset::default() .name(batch_sizes[i].to_string()) .marker(symbols::Marker::Block) .style(Style::default().fg(colors[color_idx])) .graph_type(GraphType::Scatter) .data(&latency_throughput[i..(i + 1)]) }) .collect(); // Chart Chart::new(datasets) .style(Style::default().fg(Color::Cyan).bg(Color::Reset)) .block( Block::default() .title(Span::styled( format!("{name} throughput over latency"), Style::default().fg(Color::Gray).bg(Color::Reset), )) .borders(Borders::ALL), ) .x_axis( Axis::default() .title("ms") .style(Style::default().fg(Color::Gray).bg(Color::Reset)) .labels(x_labels) .bounds([min_x, max_x]), ) .y_axis( Axis::default() .title("tokens/secs") .style(Style::default().fg(Color::Gray).bg(Color::Reset)) .labels(y_labels) .bounds([min_y, max_y]), ) } // Colors for latency/throughput chart fn color_vec() -> Vec<Color> { vec![ Color::Red, Color::Green, Color::Yellow, Color::Blue, Color::Magenta, Color::Cyan, Color::Gray, Color::DarkGray, Color::LightRed, Color::LightGreen, Color::LightYellow, Color::LightBlue, Color::LightMagenta, Color::LightCyan, ] }

text-generation-inference/benchmark/src/app.rs/0

import pytest from text_generation.types import Parameters, Request from text_generation.errors import ValidationError def test_parameters_validation(): # Test best_of Parameters(best_of=1) with pytest.raises(ValidationError): Parameters(best_of=0) with pytest.raises(ValidationError): Parameters(best_of=-1) Parameters(best_of=2, do_sample=True) with pytest.raises(ValidationError): Parameters(best_of=2) with pytest.raises(ValidationError): Parameters(best_of=2, seed=1) # Test repetition_penalty Parameters(repetition_penalty=1) with pytest.raises(ValidationError): Parameters(repetition_penalty=0) with pytest.raises(ValidationError): Parameters(repetition_penalty=-1) # Test seed Parameters(seed=1) with pytest.raises(ValidationError): Parameters(seed=-1) # Test temperature Parameters(temperature=1) with pytest.raises(ValidationError): Parameters(temperature=0) with pytest.raises(ValidationError): Parameters(temperature=-1) # Test top_k Parameters(top_k=1) with pytest.raises(ValidationError): Parameters(top_k=0) with pytest.raises(ValidationError): Parameters(top_k=-1) # Test top_p Parameters(top_p=0.5) with pytest.raises(ValidationError): Parameters(top_p=0) with pytest.raises(ValidationError): Parameters(top_p=-1) with pytest.raises(ValidationError): Parameters(top_p=1) # Test truncate Parameters(truncate=1) with pytest.raises(ValidationError): Parameters(truncate=0) with pytest.raises(ValidationError): Parameters(truncate=-1) # Test typical_p Parameters(typical_p=0.5) with pytest.raises(ValidationError): Parameters(typical_p=0) with pytest.raises(ValidationError): Parameters(typical_p=-1) with pytest.raises(ValidationError): Parameters(typical_p=1) def test_request_validation(): Request(inputs="test") with pytest.raises(ValidationError): Request(inputs="") Request(inputs="test", stream=True) Request(inputs="test", parameters=Parameters(best_of=2, do_sample=True)) with pytest.raises(ValidationError): Request( inputs="test", parameters=Parameters(best_of=2, do_sample=True), stream=True )

text-generation-inference/clients/python/tests/test_types.py/0

{ "details": { "best_of_sequences": null, "finish_reason": "length", "generated_tokens": 10, "prefill": [ { "id": 1, "logprob": null, "text": "<s>" }, { "id": 1724, "logprob": -7.703125, "text": "What" }, { "id": 338, "logprob": -1.4765625, "text": "is" }, { "id": 21784, "logprob": -9.390625, "text": "Deep" }, { "id": 29257, "logprob": -1.8583984, "text": "Learning" }, { "id": 29973, "logprob": -0.7548828, "text": "?" } ], "seed": null, "tokens": [ { "id": 13, "logprob": -1.9306641, "special": false, "text": "\n" }, { "id": 5618, "logprob": -2.4550781, "special": false, "text": "What" }, { "id": 338, "logprob": -0.5732422, "special": false, "text": " is" }, { "id": 278, "logprob": -1.5761719, "special": false, "text": " the" }, { "id": 4328, "logprob": -1.5888672, "special": false, "text": " difference" }, { "id": 1546, "logprob": -0.026504517, "special": false, "text": " between" }, { "id": 21784, "logprob": -1.4287109, "special": false, "text": " Deep" }, { "id": 29257, "logprob": -0.15856934, "special": false, "text": " Learning" }, { "id": 322, "logprob": -0.17456055, "special": false, "text": " and" }, { "id": 6189, "logprob": -0.62646484, "special": false, "text": " Machine" } ], "top_tokens": null }, "generated_text": "\nWhat is the difference between Deep Learning and Machine" }

text-generation-inference/integration-tests/models/__snapshots__/test_flash_awq/test_flash_llama_awq.json/0

{ "details": { "best_of_sequences": null, "finish_reason": "length", "generated_tokens": 10, "prefill": [ { "id": 1, "logprob": null, "text": "<s>" }, { "id": 1724, "logprob": -10.734375, "text": "What" }, { "id": 338, "logprob": -1.5488281, "text": "is" }, { "id": 21784, "logprob": -9.2890625, "text": "Deep" }, { "id": 29257, "logprob": -1.2753906, "text": "Learning" }, { "id": 29973, "logprob": -0.48046875, "text": "?" } ], "seed": null, "tokens": [ { "id": 13, "logprob": -1.1845703, "special": false, "text": "\n" }, { "id": 2772, "logprob": -0.5727539, "special": false, "text": "De" }, { "id": 1022, "logprob": -0.000108122826, "special": false, "text": "ep" }, { "id": 6509, "logprob": -0.1239624, "special": false, "text": " learning" }, { "id": 338, "logprob": -0.044433594, "special": false, "text": " is" }, { "id": 263, "logprob": -0.01852417, "special": false, "text": " a" }, { "id": 11306, "logprob": -0.45922852, "special": false, "text": " subset" }, { "id": 310, "logprob": -0.0002104044, "special": false, "text": " of" }, { "id": 4933, "logprob": -0.004787445, "special": false, "text": " machine" }, { "id": 6509, "logprob": -0.00026226044, "special": false, "text": " learning" } ] }, "generated_text": "\nDeep learning is a subset of machine learning" }

text-generation-inference/integration-tests/models/__snapshots__/test_flash_medusa/test_flash_medusa_simple.json/0

{ "generated_text": "\n return sum(L) / len(L)\n\n\ndef geometric_mean(L", "details": { "best_of_sequences": null, "finish_reason": "length", "generated_tokens": 20, "seed": null, "prefill": [ { "id": 589, "text": "def", "logprob": null }, { "id": 3226, "text": " ge", "logprob": -9.0234375 }, { "id": 21017, "text": "ometric", "logprob": -9.0859375 }, { "id": 81, "text": "_", "logprob": -0.25878906 }, { "id": 6009, "text": "mean", "logprob": -2.2109375 }, { "id": 26, "text": "(", "logprob": -0.30371094 }, { "id": 62, "text": "L", "logprob": -5.6054688 }, { "id": 44, "text": ":", "logprob": -3.0722656 }, { "id": 1682, "text": " List", "logprob": -0.6879883 }, { "id": 77, "text": "[", "logprob": -0.38500977 }, { "id": 1808, "text": "float", "logprob": -0.984375 }, { "id": 10794, "text": "]):", "logprob": -2.5351562 } ], "tokens": [ { "id": 284, "text": "\n ", "logprob": -1.1738281, "special": false }, { "id": 442, "text": " return", "logprob": -0.95947266, "special": false }, { "id": 3632, "text": " sum", "logprob": -1.4199219, "special": false }, { "id": 26, "text": "(", "logprob": -0.085876465, "special": false }, { "id": 62, "text": "L", "logprob": -0.09875488, "special": false }, { "id": 27, "text": ")", "logprob": -0.30517578, "special": false }, { "id": 517, "text": " /", "logprob": -0.42089844, "special": false }, { "id": 2069, "text": " len", "logprob": -0.042053223, "special": false }, { "id": 26, "text": "(", "logprob": -0.0011806488, "special": false }, { "id": 62, "text": "L", "logprob": -0.0005259514, "special": false }, { "id": 27, "text": ")", "logprob": -0.0017633438, "special": false }, { "id": 478, "text": "\n\n", "logprob": -0.69189453, "special": false }, { "id": 203, "text": "\n", "logprob": -0.041870117, "special": false }, { "id": 589, "text": "def", "logprob": -0.27856445, "special": false }, { "id": 3226, "text": " ge", "logprob": -1.7255859, "special": false }, { "id": 21017, "text": "ometric", "logprob": -0.011291504, "special": false }, { "id": 81, "text": "_", "logprob": -0.008430481, "special": false }, { "id": 6009, "text": "mean", "logprob": -0.025787354, "special": false }, { "id": 26, "text": "(", "logprob": -0.073913574, "special": false }, { "id": 62, "text": "L", "logprob": -0.09967041, "special": false } ] } }

text-generation-inference/integration-tests/models/__snapshots__/test_flash_starcoder_gptq/test_flash_starcoder_gptq.json/0

import pytest @pytest.fixture(scope="module") def bloom_560_handle(launcher): with launcher("bigscience/bloom-560m") as handle: yield handle @pytest.fixture(scope="module") async def bloom_560(bloom_560_handle): await bloom_560_handle.health(240) return bloom_560_handle.client @pytest.mark.asyncio async def test_bloom_560m(bloom_560, response_snapshot): response = await bloom_560.generate( "Pour déguster un ortolan, il faut tout d'abord", max_new_tokens=10, top_p=0.9, decoder_input_details=True, seed=0, ) assert response.details.generated_tokens == 10 assert response == response_snapshot @pytest.mark.asyncio async def test_bloom_560m_all_params(bloom_560, response_snapshot): response = await bloom_560.generate( "Pour déguster un ortolan, il faut tout d'abord", max_new_tokens=10, repetition_penalty=1.2, return_full_text=True, stop_sequences=["test"], temperature=0.5, top_p=0.9, top_k=10, truncate=5, typical_p=0.9, watermark=True, decoder_input_details=True, seed=0, ) assert response.details.generated_tokens == 10 assert response == response_snapshot @pytest.mark.asyncio async def test_bloom_560m_load(bloom_560, generate_load, response_snapshot): responses = await generate_load( bloom_560, "Pour déguster un ortolan, il faut tout d'abord", max_new_tokens=10, n=4, ) assert len(responses) == 4 assert all([r.generated_text == responses[0].generated_text for r in responses]) assert responses == response_snapshot

text-generation-inference/integration-tests/models/test_bloom_560m.py/0

import pytest @pytest.fixture(scope="module") def mpt_sharded_handle(launcher): with launcher("mosaicml/mpt-7b", num_shard=2) as handle: yield handle @pytest.fixture(scope="module") async def mpt_sharded(mpt_sharded_handle): await mpt_sharded_handle.health(300) return mpt_sharded_handle.client @pytest.mark.asyncio async def test_mpt(mpt_sharded, response_snapshot): response = await mpt_sharded.generate( "What is Deep Learning?", max_new_tokens=17, decoder_input_details=True, ) assert response.details.generated_tokens == 17 assert ( response.generated_text == " - Deep Learning\nDeep Learning is a subfield of machine learning that uses artificial neural" ) assert response == response_snapshot @pytest.mark.asyncio async def test_mpt_load(mpt_sharded, generate_load, response_snapshot): responses = await generate_load( mpt_sharded, "What is Deep Learning?", max_new_tokens=17, n=4, ) assert len(responses) == 4 assert all([r.generated_text == responses[0].generated_text for r in responses]) assert ( responses[0].generated_text == " - Deep Learning\nDeep Learning is a subfield of machine learning that uses artificial neural" ) assert responses == response_snapshot

text-generation-inference/integration-tests/models/test_mpt.py/0

import { get_options, run } from "./common.js"; const reference_latency_ms = 22; const host = __ENV.HOST || '127.0.0.1:8000'; const max_new_tokens = 50; function generate_payload(gpt){ const input = gpt["conversations"][0]["value"]; return {"prompt": input, "temperature": 0.5, "ignore_eos": true} } export const options = get_options(reference_latency_ms); export default function(){ run(host, generate_payload, max_new_tokens); }

text-generation-inference/load_tests/vllm.js/0

use axum::http::HeaderValue; use clap::Parser; use hf_hub::api::tokio::{Api, ApiBuilder, ApiRepo}; use hf_hub::{Repo, RepoType}; use opentelemetry::sdk::propagation::TraceContextPropagator; use opentelemetry::sdk::trace; use opentelemetry::sdk::trace::Sampler; use opentelemetry::sdk::Resource; use opentelemetry::{global, KeyValue}; use opentelemetry_otlp::WithExportConfig; use std::fs::File; use std::io::BufReader; use std::net::{IpAddr, Ipv4Addr, SocketAddr}; use std::path::Path; use text_generation_client::{ClientError, ShardedClient}; use text_generation_router::{server, HubModelInfo, HubTokenizerConfig}; use thiserror::Error; use tokenizers::Tokenizer; use tower_http::cors::AllowOrigin; use tracing_subscriber::layer::SubscriberExt; use tracing_subscriber::util::SubscriberInitExt; use tracing_subscriber::{EnvFilter, Layer}; /// App Configuration #[derive(Parser, Debug)] #[clap(author, version, about, long_about = None)] struct Args { #[clap(default_value = "128", long, env)] max_concurrent_requests: usize, #[clap(default_value = "2", long, env)] max_best_of: usize, #[clap(default_value = "4", long, env)] max_stop_sequences: usize, #[clap(default_value = "5", long, env)] max_top_n_tokens: u32, #[clap(default_value = "1024", long, env)] max_input_length: usize, #[clap(default_value = "2048", long, env)] max_total_tokens: usize, #[clap(default_value = "1.2", long, env)] waiting_served_ratio: f32, #[clap(default_value = "4096", long, env)] max_batch_prefill_tokens: u32, #[clap(long, env)] max_batch_total_tokens: Option<u32>, #[clap(default_value = "20", long, env)] max_waiting_tokens: usize, #[clap(default_value = "0.0.0.0", long, env)] hostname: String, #[clap(default_value = "3000", long, short, env)] port: u16, #[clap(default_value = "/tmp/text-generation-server-0", long, env)] master_shard_uds_path: String, #[clap(default_value = "bigscience/bloom", long, env)] tokenizer_name: String, #[clap(long, env)] tokenizer_config_path: Option<String>, #[clap(long, env)] revision: Option<String>, #[clap(default_value = "2", long, env)] validation_workers: usize, #[clap(long, env)] json_output: bool, #[clap(long, env)] otlp_endpoint: Option<String>, #[clap(long, env)] cors_allow_origin: Option<Vec<String>>, #[clap(long, env)] ngrok: bool, #[clap(long, env)] ngrok_authtoken: Option<String>, #[clap(long, env)] ngrok_edge: Option<String>, #[clap(long, env, default_value_t = false)] messages_api_enabled: bool, } #[tokio::main] async fn main() -> Result<(), RouterError> { // Get args let args = Args::parse(); // Pattern match configuration let Args { max_concurrent_requests, max_best_of, max_stop_sequences, max_top_n_tokens, max_input_length, max_total_tokens, waiting_served_ratio, max_batch_prefill_tokens, max_batch_total_tokens, max_waiting_tokens, hostname, port, master_shard_uds_path, tokenizer_name, tokenizer_config_path, revision, validation_workers, json_output, otlp_endpoint, cors_allow_origin, ngrok, ngrok_authtoken, ngrok_edge, messages_api_enabled, } = args; // Launch Tokio runtime init_logging(otlp_endpoint, json_output); // Validate args if max_input_length >= max_total_tokens { return Err(RouterError::ArgumentValidation( "`max_input_length` must be < `max_total_tokens`".to_string(), )); } if max_input_length as u32 > max_batch_prefill_tokens { return Err(RouterError::ArgumentValidation(format!("`max_batch_prefill_tokens` must be >= `max_input_length`. Given: {max_batch_prefill_tokens} and {max_input_length}"))); } if validation_workers == 0 { return Err(RouterError::ArgumentValidation( "`validation_workers` must be > 0".to_string(), )); } if let Some(ref max_batch_total_tokens) = max_batch_total_tokens { if max_batch_prefill_tokens > *max_batch_total_tokens { return Err(RouterError::ArgumentValidation(format!("`max_batch_prefill_tokens` must be <= `max_batch_total_tokens`. Given: {max_batch_prefill_tokens} and {max_batch_total_tokens}"))); } if max_total_tokens as u32 > *max_batch_total_tokens { return Err(RouterError::ArgumentValidation(format!("`max_total_tokens` must be <= `max_batch_total_tokens`. Given: {max_total_tokens} and {max_batch_total_tokens}"))); } } // CORS allowed origins // map to go inside the option and then map to parse from String to HeaderValue // Finally, convert to AllowOrigin let cors_allow_origin: Option<AllowOrigin> = cors_allow_origin.map(|cors_allow_origin| { AllowOrigin::list( cors_allow_origin .iter() .map(|origin| origin.parse::<HeaderValue>().unwrap()), ) }); // Parse Huggingface hub token let authorization_token = std::env::var("HUGGING_FACE_HUB_TOKEN").ok(); // Tokenizer instance // This will only be used to validate payloads let local_path = Path::new(&tokenizer_name); let local_model = local_path.exists() && local_path.is_dir(); // Shared API builder initialization let api_builder = || { let mut builder = ApiBuilder::new() .with_progress(false) .with_token(authorization_token); if let Ok(cache_dir) = std::env::var("HUGGINGFACE_HUB_CACHE") { builder = builder.with_cache_dir(cache_dir.into()); } builder }; // Decide if we need to use the API based on the revision and local path let use_api = revision.is_some() || !local_path.exists() || !local_path.is_dir(); // Initialize API if needed let api = if use_api { tracing::info!("Using the Hugging Face API"); match api_builder().build() { Ok(api) => Some(api), Err(_) => { tracing::warn!("Unable to build the Hugging Face API"); None } } } else { None }; // Load tokenizer and model info let (tokenizer, model_info) = if local_model { let tokenizer = Tokenizer::from_file(local_path.join("tokenizer.json")).ok(); let model_info = HubModelInfo { model_id: tokenizer_name.to_string(), sha: None, pipeline_tag: None, }; (tokenizer, model_info) } else if let Some(api) = api.clone() { let api_repo = api.repo(Repo::with_revision( tokenizer_name.to_string(), RepoType::Model, revision.clone().unwrap_or_else(|| "main".to_string()), )); let tokenizer = match api_repo.get("tokenizer.json").await { Ok(tokenizer_filename) => Tokenizer::from_file(tokenizer_filename).ok(), Err(_) => get_base_tokenizer(&api, &api_repo).await, }; let model_info = get_model_info(&api_repo).await.unwrap_or_else(|| { tracing::warn!("Could not retrieve model info from the Hugging Face hub."); HubModelInfo { model_id: tokenizer_name.to_string(), sha: None, pipeline_tag: None, } }); (tokenizer, model_info) } else { // No API and no local model return Err(RouterError::ArgumentValidation( "No local model found and no revision specified".to_string(), )); }; // Load tokenizer config if found locally, or check if we can get it from the API if needed let tokenizer_config = if let Some(path) = tokenizer_config_path { tracing::info!("Using local tokenizer config from user specified path"); HubTokenizerConfig::from_file(&std::path::PathBuf::from(path)) } else if local_model { tracing::info!("Using local tokenizer config"); HubTokenizerConfig::from_file(&local_path.join("tokenizer_config.json")) } else { match api { Some(api) => { tracing::info!("Using the Hugging Face API to retrieve tokenizer config"); let repo = Repo::with_revision( tokenizer_name.to_string(), RepoType::Model, revision.unwrap_or("main".to_string()), ); get_tokenizer_config(&api.repo(repo)) .await .unwrap_or_else(|| { tracing::warn!( "Could not retrieve tokenizer config from the Hugging Face hub." ); HubTokenizerConfig::default() }) } None => { tracing::warn!("Could not find tokenizer config locally and no API specified"); HubTokenizerConfig::default() } } }; if tokenizer.is_none() { tracing::warn!("Could not find a fast tokenizer implementation for {tokenizer_name}"); tracing::warn!("Rust input length validation and truncation is disabled"); } // if pipeline-tag == text-generation we default to return_full_text = true let compat_return_full_text = match &model_info.pipeline_tag { None => { tracing::warn!("no pipeline tag found for model {tokenizer_name}"); false } Some(pipeline_tag) => pipeline_tag.as_str() == "text-generation", }; // Instantiate sharded client from the master unix socket let mut sharded_client = ShardedClient::connect_uds(master_shard_uds_path) .await .map_err(RouterError::Connection)?; // Clear the cache; useful if the webserver rebooted sharded_client .clear_cache(None) .await .map_err(RouterError::Cache)?; // Get info from the shard let shard_info = sharded_client.info().await.map_err(RouterError::Info)?; // Warmup model tracing::info!("Warming up model"); let max_supported_batch_total_tokens = match sharded_client .warmup( max_input_length as u32, max_batch_prefill_tokens, max_total_tokens as u32, ) .await .map_err(RouterError::Warmup)? { // Older models do not support automatic max-batch-total-tokens None => { let max_batch_total_tokens = max_batch_total_tokens .unwrap_or(16000.max((max_total_tokens as u32).max(max_batch_prefill_tokens))); tracing::warn!("Model does not support automatic max batch total tokens"); max_batch_total_tokens } // Flash attention models return their max supported total tokens Some(max_supported_batch_total_tokens) => { // Warn if user added his own max-batch-total-tokens as we will ignore it if max_batch_total_tokens.is_some() { tracing::warn!( "`--max-batch-total-tokens` is deprecated for Flash \ Attention models." ); tracing::warn!( "Inferred max batch total tokens: {max_supported_batch_total_tokens}" ); } if max_total_tokens as u32 > max_supported_batch_total_tokens { return Err(RouterError::ArgumentValidation(format!("`max_total_tokens` must be <= `max_batch_total_tokens`. Given: {max_total_tokens} and {max_supported_batch_total_tokens}"))); } max_supported_batch_total_tokens } }; tracing::info!("Setting max batch total tokens to {max_supported_batch_total_tokens}"); tracing::info!("Connected"); let addr = match hostname.parse() { Ok(ip) => SocketAddr::new(ip, port), Err(_) => { tracing::warn!("Invalid hostname, defaulting to 0.0.0.0"); SocketAddr::new(IpAddr::V4(Ipv4Addr::new(0, 0, 0, 0)), port) } }; // Run server server::run( model_info, shard_info, compat_return_full_text, max_concurrent_requests, max_best_of, max_stop_sequences, max_top_n_tokens, max_input_length, max_total_tokens, waiting_served_ratio, max_batch_prefill_tokens, max_supported_batch_total_tokens, max_waiting_tokens, sharded_client, tokenizer, validation_workers, addr, cors_allow_origin, ngrok, ngrok_authtoken, ngrok_edge, tokenizer_config, messages_api_enabled, ) .await?; Ok(()) } /// Init logging using env variables LOG_LEVEL and LOG_FORMAT: /// - otlp_endpoint is an optional URL to an Open Telemetry collector /// - LOG_LEVEL may be TRACE, DEBUG, INFO, WARN or ERROR (default to INFO) /// - LOG_FORMAT may be TEXT or JSON (default to TEXT) fn init_logging(otlp_endpoint: Option<String>, json_output: bool) { let mut layers = Vec::new(); // STDOUT/STDERR layer let fmt_layer = tracing_subscriber::fmt::layer() .with_file(true) .with_line_number(true); let fmt_layer = match json_output { true => fmt_layer.json().flatten_event(true).boxed(), false => fmt_layer.boxed(), }; layers.push(fmt_layer); // OpenTelemetry tracing layer if let Some(otlp_endpoint) = otlp_endpoint { global::set_text_map_propagator(TraceContextPropagator::new()); let tracer = opentelemetry_otlp::new_pipeline() .tracing() .with_exporter( opentelemetry_otlp::new_exporter() .tonic() .with_endpoint(otlp_endpoint), ) .with_trace_config( trace::config() .with_resource(Resource::new(vec![KeyValue::new( "service.name", "text-generation-inference.router", )])) .with_sampler(Sampler::AlwaysOn), ) .install_batch(opentelemetry::runtime::Tokio); if let Ok(tracer) = tracer { layers.push(tracing_opentelemetry::layer().with_tracer(tracer).boxed()); init_tracing_opentelemetry::init_propagator().unwrap(); }; } // Filter events with LOG_LEVEL let env_filter = EnvFilter::try_from_env("LOG_LEVEL").unwrap_or_else(|_| EnvFilter::new("info")); tracing_subscriber::registry() .with(env_filter) .with(layers) .init(); } /// get model info from the Huggingface Hub pub async fn get_model_info(api: &ApiRepo) -> Option<HubModelInfo> { let response = api.info_request().send().await.ok()?; if response.status().is_success() { let hub_model_info: HubModelInfo = serde_json::from_str(&response.text().await.ok()?).ok()?; if let Some(sha) = &hub_model_info.sha { tracing::info!( "Serving revision {sha} of model {}", hub_model_info.model_id ); } Some(hub_model_info) } else { None } } /// get base tokenizer pub async fn get_base_tokenizer(api: &Api, api_repo: &ApiRepo) -> Option<Tokenizer> { let config_filename = api_repo.get("config.json").await.ok()?; // Open the file in read-only mode with buffer. let file = File::open(config_filename).ok()?; let reader = BufReader::new(file); // Read the JSON contents of the file as an instance of `User`. let config: serde_json::Value = serde_json::from_reader(reader).ok()?; if let Some(serde_json::Value::String(base_model_id)) = config.get("base_model_name_or_path") { let api_base_repo = api.repo(Repo::with_revision( base_model_id.to_string(), RepoType::Model, "main".to_string(), )); let tokenizer_filename = api_base_repo.get("tokenizer.json").await.ok()?; Tokenizer::from_file(tokenizer_filename).ok() } else { None } } /// get tokenizer_config from the Huggingface Hub pub async fn get_tokenizer_config(api_repo: &ApiRepo) -> Option<HubTokenizerConfig> { let tokenizer_config_filename = api_repo.get("tokenizer_config.json").await.ok()?; // Open the file in read-only mode with buffer. let file = File::open(tokenizer_config_filename).ok()?; let reader = BufReader::new(file); // Read the JSON contents of the file as an instance of 'HubTokenizerConfig'. let tokenizer_config: HubTokenizerConfig = serde_json::from_reader(reader) .map_err(|e| { tracing::warn!("Unable to parse tokenizer config: {}", e); e }) .ok()?; Some(tokenizer_config) } #[derive(Debug, Error)] enum RouterError { #[error("Argument validation error: {0}")] ArgumentValidation(String), #[error("Unable to connect to the Python model shards: {0}")] Connection(ClientError), #[error("Unable to clear the Python model shards cache: {0}")] Cache(ClientError), #[error("Unable to get the Python model shards info: {0}")] Info(ClientError), #[error("Unable to warmup the Python model shards: {0}")] Warmup(ClientError), #[error("Tokio runtime failed to start: {0}")] Tokio(#[from] std::io::Error), #[error("Axum webserver failed: {0}")] Axum(#[from] axum::BoxError), }

text-generation-inference/router/src/main.rs/0

from setuptools import setup from torch.utils.cpp_extension import BuildExtension, CUDAExtension import torch extra_compile_args = ["-std=c++17"] if not torch.version.hip: extra_compile_args.append("-arch=compute_80") setup( name="custom_kernels", ext_modules=[ CUDAExtension( name="custom_kernels.fused_bloom_attention_cuda", sources=["custom_kernels/fused_bloom_attention_cuda.cu"], extra_compile_args=extra_compile_args, ), CUDAExtension( name="custom_kernels.fused_attention_cuda", sources=["custom_kernels/fused_attention_cuda.cu"], extra_compile_args=extra_compile_args, ), ], cmdclass={"build_ext": BuildExtension}, )

text-generation-inference/server/custom_kernels/setup.py/0

#ifndef _config_h #define _config_h #define MAX_Q_GEMM_ROWS 50 #define MAX_Q_GEMM_WEIGHTS 4 // must be <= MAX_Q_GEMM_ROWS #define QMODE_2BIT 1 #define QMODE_3BIT 1 #define QMODE_4BIT 1 #define QMODE_5BIT 1 #define QMODE_6BIT 0 #define QMODE_8BIT 0 #endif

text-generation-inference/server/exllamav2_kernels/exllamav2_kernels/config.h/0

#ifndef _qdq_util_cuh #define _qdq_util_cuh union half2_uint32 { uint32_t as_uint32; half2 as_half2; __device__ half2_uint32(uint32_t val) : as_uint32(val) {} __device__ half2_uint32(half2 val) : as_half2(val) {} __device__ half2_uint32() : as_uint32(0) {} }; union half_uint16 { uint16_t as_uint16; half as_half; __device__ half_uint16(uint16_t val) : as_uint16(val) {} __device__ half_uint16(half val) : as_half(val) {} __device__ half_uint16() : as_uint16(0) {} }; // Max_scale premultiplied by 1/256 __forceinline__ __device__ half dq_scale(const int qs, const half max_scale) { int qs_i = qs + 1; half qs_h = __int2half_rn(qs_i * qs_i); qs_h = __hmul(qs_h, max_scale); return qs_h; } __forceinline__ __device__ half dq(const int q, const int qzero, const half scale) { return __hmul(__int2half_rn(q - qzero), scale); } __forceinline__ __device__ half dq_ns(const int q, const int qzero) { //return __hsub(__int2half_rn(q), __int2half_rn(qzero)); return __int2half_rn(q - qzero); } __forceinline__ __device__ int exb(const uint32_t q, const int shift, const int mask) { return (int)((q >> shift) & mask); } __forceinline__ __device__ int exb(const uint32_t q1, const uint32_t q0, const int shift, const int mask) { return (int)(__funnelshift_rc(q0, q1, shift) & mask); } #endif

text-generation-inference/server/exllamav2_kernels/exllamav2_kernels/cuda/quant/qdq_util.cuh/0

import os import requests import tempfile import pytest import huggingface_hub.constants from huggingface_hub import hf_api import text_generation_server.utils.hub from text_generation_server.utils.hub import ( weight_hub_files, download_weights, weight_files, EntryNotFoundError, LocalEntryNotFoundError, RevisionNotFoundError, ) @pytest.fixture() def offline(): current_value = text_generation_server.utils.hub.HF_HUB_OFFLINE text_generation_server.utils.hub.HF_HUB_OFFLINE = True yield "offline" text_generation_server.utils.hub.HF_HUB_OFFLINE = current_value @pytest.fixture() def fresh_cache(): with tempfile.TemporaryDirectory() as d: current_value = huggingface_hub.constants.HUGGINGFACE_HUB_CACHE huggingface_hub.constants.HUGGINGFACE_HUB_CACHE = d text_generation_server.utils.hub.HUGGINGFACE_HUB_CACHE = d os.environ["HUGGINGFACE_HUB_CACHE"] = d yield huggingface_hub.constants.HUGGINGFACE_HUB_CACHE = current_value os.environ["HUGGINGFACE_HUB_CACHE"] = current_value text_generation_server.utils.hub.HUGGINGFACE_HUB_CACHE = current_value @pytest.fixture() def prefetched(): model_id = "bert-base-uncased" huggingface_hub.snapshot_download( repo_id=model_id, revision="main", local_files_only=False, repo_type="model", allow_patterns=["*.safetensors"], ) yield model_id def test_weight_hub_files_offline_error(offline, fresh_cache): # If the model is not prefetched then it will raise an error with pytest.raises(EntryNotFoundError): weight_hub_files("gpt2") def test_weight_hub_files_offline_ok(prefetched, offline): # If the model is prefetched then we should be able to get the weight files from local cache filenames = weight_hub_files(prefetched) root = None assert len(filenames) == 1 for f in filenames: curroot, filename = os.path.split(f) if root is None: root = curroot else: assert root == curroot assert filename == "model.safetensors" def test_weight_hub_files(): filenames = weight_hub_files("bigscience/bloom-560m") assert filenames == ["model.safetensors"] def test_weight_hub_files_llm(): filenames = weight_hub_files("bigscience/bloom") assert filenames == [f"model_{i:05d}-of-00072.safetensors" for i in range(1, 73)] def test_weight_hub_files_empty(): with pytest.raises(EntryNotFoundError): weight_hub_files("bigscience/bloom", extension=".errors") def test_download_weights(): model_id = "bigscience/bloom-560m" filenames = weight_hub_files(model_id) files = download_weights(filenames, model_id) local_files = weight_files("bigscience/bloom-560m") assert files == local_files def test_weight_files_revision_error(): with pytest.raises(RevisionNotFoundError): weight_files("bigscience/bloom-560m", revision="error") def test_weight_files_not_cached_error(fresh_cache): with pytest.raises(LocalEntryNotFoundError): weight_files("bert-base-uncased")

text-generation-inference/server/tests/utils/test_hub.py/0

# coding=utf-8 # Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved. # # This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX # and OPT implementations in this library. It has been modified from its # original forms to accommodate minor architectural differences compared # to GPT-NeoX and OPT used by the Meta AI team that trained the model. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import torch import torch.distributed import numpy as np from torch import nn from transformers.activations import ACT2FN from transformers.configuration_utils import PretrainedConfig from typing import Optional, List, Tuple from loguru import logger from text_generation_server.utils import paged_attention, flash_attn from text_generation_server.utils.layers import ( FastLinear, FastRMSNorm, TensorParallelRowLinear, TensorParallelColumnLinear, TensorParallelEmbedding, PositionRotaryEmbedding, TensorParallelHead, get_linear, ) HAS_MEGABLOCKS = True try: import stk import megablocks.ops as ops except ImportError: logger.warning("Mixtral: megablocks is not installed") HAS_MEGABLOCKS = False class MixtralConfig(PretrainedConfig): model_type = "mixtral" def __init__( self, vocab_size=32000, hidden_size=4096, intermediate_size=14336, num_hidden_layers=32, num_attention_heads=32, num_key_value_heads=8, hidden_act="silu", max_position_embeddings=4096 * 32, initializer_range=0.02, rms_norm_eps=1e-05, use_cache=True, pad_token_id=None, bos_token_id=1, eos_token_id=2, pretraining_tp=1, tie_word_embeddings=False, rope_theta=10000.0, sliding_window=None, num_experts_per_tok=2, num_local_experts=8, **kwargs, ): self.vocab_size = vocab_size self.max_position_embeddings = max_position_embeddings self.hidden_size = hidden_size self.intermediate_size = intermediate_size self.num_hidden_layers = num_hidden_layers self.num_attention_heads = num_attention_heads self.sliding_window = sliding_window # for backward compatibility if num_key_value_heads is None: num_key_value_heads = num_attention_heads self.num_key_value_heads = num_key_value_heads self.hidden_act = hidden_act self.initializer_range = initializer_range self.rms_norm_eps = rms_norm_eps self.pretraining_tp = pretraining_tp self.use_cache = use_cache self.rope_theta = rope_theta self.num_experts_per_tok = num_experts_per_tok self.num_local_experts = num_local_experts super().__init__( pad_token_id=pad_token_id, bos_token_id=bos_token_id, eos_token_id=eos_token_id, tie_word_embeddings=tie_word_embeddings, **kwargs, ) def promote_scalar(x: torch.Tensor) -> torch.Tensor: return x.view(1) if len(x.size()) == 0 else x def load_attention(config, prefix, weights): if config.num_attention_heads != config.num_key_value_heads: return _load_gqa(config, prefix, weights) else: return TensorParallelColumnLinear.load_multi( config, prefixes=[f"{prefix}.q_proj", f"{prefix}.k_proj", f"{prefix}.v_proj"], dim=0, weights=weights, bias=False, ) def _load_gqa(config, prefix: str, weights): assert config.hidden_size % config.num_attention_heads == 0 assert config.num_attention_heads % weights.process_group.size() == 0 weight = weights.get_multi_weights_col( prefixes=[f"{prefix}.q_proj", f"{prefix}.k_proj", f"{prefix}.v_proj"], quantize=config.quantize, dim=0, ) if config.quantize not in ["gptq", "awq"]: weight = weight.to(dtype=weights.dtype).to(device=weights.device) head_size = config.hidden_size // config.num_attention_heads num_heads = config.num_attention_heads // weights.process_group.size() num_key_value_heads = config.num_key_value_heads // weights.process_group.size() assert list(weight.shape) == [ (num_heads + 2 * num_key_value_heads) * head_size, config.hidden_size, ], f"{list(weight.shape)} != {[(num_heads + 2 * config.num_key_value_heads) * head_size, config.hidden_size]}" return TensorParallelColumnLinear( get_linear(weight, bias=None, quantize=config.quantize) ) def _load_experts(config, prefix, mat, weights): if config.quantize is not None: raise NotImplementedError("Mixtral does not support weight quantization yet.") assert mat in ["w1", "w2", "w3"] world_size = weights.process_group.size() rank = weights.process_group.rank() assert ( config.intermediate_size % world_size == 0 ), f"The chosen size {config.intermediate_size} is not compatible with sharding on {world_size} shards" block_size = config.intermediate_size // world_size start = rank * block_size stop = (rank + 1) * block_size tensor = torch.empty( (config.num_local_experts * block_size, config.hidden_size), dtype=weights.dtype, device=weights.device, ) for i in range(config.num_local_experts): slice_ = weights._get_slice(f"{prefix}.{i}.{mat}.weight") if mat == "w2": expert_slice = slice_[:, start:stop].t().contiguous() else: expert_slice = slice_[start:stop] tensor[i * block_size : (i + 1) * block_size] = expert_slice.to( dtype=weights.dtype ).to(device=weights.device) return tensor class MixtralAttention(torch.nn.Module): def __init__( self, prefix: str, config, weights, ): super().__init__() self.max_past = ( config.sliding_window if config.sliding_window is not None else -1 ) self.num_heads = config.num_attention_heads self.hidden_size = config.hidden_size self.head_size = self.hidden_size // self.num_heads self.rotary_emb = PositionRotaryEmbedding.static( config=config, dim=self.head_size, base=config.rope_theta, device=weights.device, ) self.softmax_scale = self.head_size**-0.5 if self.num_heads % weights.process_group.size() != 0: raise ValueError( f"`num_heads` must be divisible by `num_shards` (got `num_heads`: {self.num_heads} " f"and `num_shards`: {weights.process_group.size()}" ) self.num_heads = self.num_heads // weights.process_group.size() self.num_key_value_heads = ( config.num_key_value_heads // weights.process_group.size() ) self.query_key_value = load_attention(config, prefix, weights) self.o_proj = TensorParallelRowLinear.load( config, prefix=f"{prefix}.o_proj", weights=weights, bias=False, ) self.num_groups = self.num_heads // self.num_key_value_heads self.kv_head_mapping = torch.arange( 0, self.num_key_value_heads, dtype=torch.int32, device=weights.device ).repeat_interleave(self.num_groups) def forward( self, hidden_states, cos, sin, cu_seqlen_prefill, kv_cache, block_tables, slots, input_lengths, max_s, prefill_cache_indices, ): qkv = self.query_key_value(hidden_states) query, kv = qkv.split( [ self.head_size * self.num_heads, 2 * self.head_size * self.num_key_value_heads, ], dim=1, ) query = query.view(-1, self.num_heads, self.head_size) kv = kv.view(-1, 2, self.num_key_value_heads, self.head_size) self.rotary_emb(query, torch.select(kv, dim=1, index=0), cos, sin) if prefill_cache_indices is not None: kv_to_cache = kv[prefill_cache_indices] else: kv_to_cache = kv paged_attention.reshape_and_cache( kv_to_cache[:, 0], kv_to_cache[:, 1], kv_cache[0], kv_cache[1], slots ) # output tensor attn_output = torch.empty_like(query) # Prefill if cu_seqlen_prefill is not None: # flash attention flash_attn.attention( query, torch.select(kv, dim=1, index=0), torch.select(kv, dim=1, index=1), attn_output, cu_seqlen_prefill, max_s, self.softmax_scale, window_size_left=self.max_past, ) # Decode else: paged_attention.attention( attn_output, query, kv_cache[0], kv_cache[1], self.kv_head_mapping, self.softmax_scale, block_tables, input_lengths, max_s, ) return self.o_proj(attn_output.view(-1, self.num_heads * self.head_size)) @torch.jit.script def select_experts(gate_logits: torch.Tensor, top_k: int): # all_probs: (sequence_length, n_experts) and upcast for softmax all_probs = torch.nn.functional.softmax(gate_logits, dim=1, dtype=torch.float) # weights, selected_experts: (sequence_length, top-k) weights, selected_experts = torch.topk(all_probs, top_k, dim=-1) weights /= weights.sum(dim=-1, keepdim=True) weights = weights.view(-1) selected_experts = selected_experts.view(-1) return selected_experts, weights @torch.jit.script def round_up(x: torch.Tensor, value: int): return torch.div(x + (value - 1), value, rounding_mode="trunc") * value class BlockSparseMoE(nn.Module): """ Built on the paper and library Megablocks as described in https://arxiv.org/abs/2211.15841. This implementation is strictly equivalent to standard MoE with full capacity (no dropped tokens). It's faster since it formulates MoE operations in terms of block-sparse operations to accomodate imbalanced assignments of tokens to experts, whereas standard MoE either (1) drop tokens at the cost of reduced performance or (2) set capacity factor to number of experts and thus waste computation and memory on padding. """ def __init__(self, prefix, config: MixtralConfig, weights): super().__init__() self.hidden_dim = config.hidden_size self.ffn_dim = config.intermediate_size // weights.process_group.size() self.num_experts = config.num_local_experts self.top_k = config.num_experts_per_tok act = config.hidden_act if "gelu" in act: self.act = lambda x: torch.nn.functional.gelu( x, approximate="tanh" if act in ["gelu_fast", "gelu_pytorch_tanh"] else "none", ) elif "silu" in act: self.act = torch.nn.functional.silu else: self.act = ACT2FN[act] # gating self.gate = FastLinear.load(config, f"{prefix}.gate", weights, bias=False) # merged expert weights, all of size (n_experts * ffn_dim, hidden_dim) self.w1 = _load_experts(config, f"{prefix}.experts", "w1", weights) self.w2 = _load_experts(config, f"{prefix}.experts", "w2", weights) self.w3 = _load_experts(config, f"{prefix}.experts", "w3", weights) self.offsets = None self.offsets_block_rows = 0 self.process_group = weights.process_group # Calculate the number of bits needed to represent the expert indices # so that we can pass it to radix sort. self.sort_end_bit = max(int(np.ceil(np.log2(self.num_experts))), 1) self.blocking = 128 self.quantize_scatter_num_bits = -1 def topology(self, x: torch.Tensor, padded_bins: torch.Tensor): padded_tokens, _ = x.size() assert padded_tokens % self.blocking == 0 assert self.ffn_dim % self.blocking == 0 # Offsets for the sparse matrix. All rows have the # same number of nonzero blocks dictated by the # dimensionality of a single expert. block_rows = padded_tokens // self.blocking blocks_per_row = self.ffn_dim // self.blocking if self.offsets is None or block_rows > self.offsets_block_rows: self.offsets = torch.arange( 0, block_rows * blocks_per_row + 1, blocks_per_row, dtype=torch.int32, device=x.device, ) self.offsets_block_rows = block_rows offsets = self.offsets else: offsets = self.offsets[: block_rows + 1] # Indices for the sparse matrix. The indices for # the intermediate matrix are dynamic depending # on the mapping of tokens to experts. column_indices = ops.topology( padded_bins, self.blocking, block_rows, blocks_per_row ) # For now, use meta init to save the device memory. data = torch.empty( column_indices.numel(), self.blocking, self.blocking, dtype=x.dtype, device="meta", ) shape = (padded_tokens, self.ffn_dim * self.num_experts) row_indices = stk.ops.row_indices(shape, data, offsets, column_indices) return stk.Matrix( shape, data, row_indices, column_indices, offsets, False, False, False, ) def indices_and_padded_bins(self, selected_experts: torch.Tensor): # Sort the expert ids to produce the scatter/gather # indices for the permutation. # selected_experts = selected_experts.int() # returns bin_ids == num of experts for this sequence ? == unique selected experts? # and indices == how to sort tokens? bin_ids, indices = ops.sort(selected_experts, self.sort_end_bit) # bin_ids => [0, 0, 0, 2, 2, ...] => [num_tokens * top_k] # indices => [14, 32, 33, ...] => [num_tokens * top_k] # Histogram the expert ids to identify the number of # tokens routed to each expert. tokens_per_expert = ops.histogram(selected_experts, self.num_experts) # tokens_per_expert => [3, 0, 2, ...] => [num_experts] # Round the token counts up to the block size used in # the matrix muliplications. Caculate the starting # position of each bin. # List of size num_experts padded_tokens_per_expert = round_up(tokens_per_expert, self.blocking) # padded_tokens_per_expert => [128, O, 128, ...] # Cumulative selected experts per token padded_bins = ops.inclusive_cumsum(padded_tokens_per_expert, 0) padded_bins = promote_scalar(padded_bins) # padded_bins => [128, 128, 256, ...] # Calculate the bin bounds for the sorted tokens. bins = ops.inclusive_cumsum(tokens_per_expert, 0) bins = promote_scalar(bins) # bins => [3, 3, 5, ...] return indices, bin_ids, bins, padded_bins, tokens_per_expert def sparse_forward(self, x: torch.Tensor) -> torch.Tensor: """ x: (sequence_length, model_dim) gate_logits: (sequence_length, n_experts) """ # optional reshape input_shape = x.shape x = x.view(-1, input_shape[-1]) # gate_logits: (sequence_length, n_experts) gate_logits = self.gate(x) selected_experts, weights = select_experts(gate_logits, self.top_k) ( indices, bin_ids, bins, padded_bins, _, ) = self.indices_and_padded_bins(selected_experts) # Permute tokens and pad to prepare expert computation # (top_k * sequence_length + padding, model_dim) x = ops.padded_gather(x, indices, bin_ids, bins, padded_bins, self.top_k) # Create the sparse matrix topology with torch.no_grad(): topo = self.topology(x, padded_bins) # Perform the expert computation # First Dense x Dense -> Sparse for w1 and w3, # (top_k * sequence_length + padding, ffn_dim * n_experts) x = stk.Matrix( topo.size(), self.act(stk.ops.sdd(x, self.w1.t(), topo).data) * stk.ops.sdd(x, self.w3.t(), topo).data, topo.row_indices, topo.column_indices, topo.offsets, topo.column_indices_t, topo.offsets_t, topo.block_offsets_t, ) # Then Sparse x Dense -> Dense for w2 # (top_k * sequence_length + padding, model_dim) x = stk.ops.dsd(x, self.w2) # Permute back and remove padding # (sequence_length, model_dim) x = ops.padded_scatter( x, indices, bin_ids, weights, bins, padded_bins, self.top_k, self.quantize_scatter_num_bits, ).view(*input_shape) if self.process_group.size() > 1: torch.distributed.all_reduce(x, group=self.process_group) return x.view(*input_shape) def dense_forward(self, x: torch.Tensor) -> torch.Tensor: """ x: (sequence_length, model_dim) gate_logits: (sequence_length, n_experts) """ # optional reshape input_shape = x.shape x = x.view(-1, input_shape[-1]) # gate_logits: (sequence_length, n_experts) gate_logits = self.gate(x) # all_probs: (sequence_length, n_experts) and upcast for softmax all_probs = torch.nn.functional.softmax(gate_logits, dim=1, dtype=torch.float) if self.top_k < self.num_experts: _, not_selected_experts = torch.topk( all_probs, self.num_experts - self.top_k, largest=False, sorted=False, dim=1, ) # Mask not selected experts all_probs.scatter_(1, not_selected_experts, 0) # Re-normalize weights = all_probs / all_probs.sum(dim=1, keepdim=True) # Expand to [num_experts, sequence_length, model_dim] x = x.view(1, -1, input_shape[-1]).expand(self.num_experts, -1, input_shape[-1]) # Permute to [num_experts, model_dim, ffn_dim] w1 = self.w1.view(self.num_experts, self.ffn_dim, self.hidden_dim).permute( 0, 2, 1 ) w3 = self.w3.view(self.num_experts, self.ffn_dim, self.hidden_dim).permute( 0, 2, 1 ) inter = self.act(torch.bmm(x, w1)) * torch.bmm(x, w3) out = torch.bmm( inter, self.w2.view(self.num_experts, self.ffn_dim, self.hidden_dim) ) # Mask not selected experts out *= weights.t().view(self.num_experts, -1, 1) # Sum experts out = out.sum(0) # Reduce sum if self.process_group.size() > 1: torch.distributed.all_reduce(out, group=self.process_group) return out def forward(self, x: torch.Tensor) -> torch.Tensor: if len(x) > 256 and HAS_MEGABLOCKS: return self.sparse_forward(x) # This is faster when there is not a lot of tokens return self.dense_forward(x) class DenseMoE(nn.Module): def __init__(self, prefix, config: MixtralConfig, weights): super().__init__() self.hidden_dim = config.hidden_size self.ffn_dim = config.intermediate_size // weights.process_group.size() self.num_experts = config.num_local_experts self.top_k = config.num_experts_per_tok act = config.hidden_act if "gelu" in act: self.act = lambda x: torch.nn.functional.gelu( x, approximate="tanh" if act in ["gelu_fast", "gelu_pytorch_tanh"] else "none", ) elif "silu" in act: self.act = torch.nn.functional.silu else: self.act = ACT2FN[act] # gating self.gate = FastLinear.load(config, f"{prefix}.gate", weights, bias=False) self.w1 = [ TensorParallelColumnLinear.load( config, prefix=f"{prefix}.experts.{i}.w1", weights=weights, bias=False ) for i in range(self.num_experts) ] self.w3 = [ TensorParallelColumnLinear.load( config, prefix=f"{prefix}.experts.{i}.w3", weights=weights, bias=False ) for i in range(self.num_experts) ] self.w2 = [ TensorParallelRowLinear.load( config, prefix=f"{prefix}.experts.{i}.w2", weights=weights, bias=False ) for i in range(self.num_experts) ] self.process_group = weights.process_group def forward(self, x: torch.Tensor) -> torch.Tensor: """ x: (sequence_length, model_dim) gate_logits: (sequence_length, n_experts) """ # optional reshape input_shape = x.shape x = x.view(-1, input_shape[-1]) # gate_logits: (sequence_length, n_experts) gate_logits = self.gate(x) # all_probs: (sequence_length, n_experts) and upcast for softmax all_probs = torch.nn.functional.softmax(gate_logits, dim=1, dtype=torch.float) if self.top_k < self.num_experts: _, not_selected_experts = torch.topk( all_probs, self.num_experts - self.top_k, largest=False, sorted=False, dim=1, ) # Mask not selected experts all_probs.scatter_(1, not_selected_experts, 0) # Re-normalize weights = all_probs / all_probs.sum(dim=1, keepdim=True) # Final output tensor out = x.new_zeros(x.shape[0], self.hidden_dim) for i in range(self.num_experts): h = self.act(self.w1[i](x)) * self.w3[i](x) h = self.w2[i](h, reduce=False) # Add expert output to out with masking out += h * weights[:, i].view(-1, 1) # Reduce sum if self.process_group.size() > 1: torch.distributed.all_reduce(out, group=self.process_group) return out class MixtralLayer(nn.Module): def __init__(self, layer_id, config, weights): super().__init__() prefix = f"model.layers.{layer_id}" self.self_attn = MixtralAttention( prefix=f"{prefix}.self_attn", config=config, weights=weights ) moe_cls = BlockSparseMoE if config.quantize is None else DenseMoE self.moe = moe_cls(f"{prefix}.block_sparse_moe", config, weights) self.input_layernorm = FastRMSNorm.load( prefix=f"{prefix}.input_layernorm", weights=weights, eps=config.rms_norm_eps ) self.post_attention_layernorm = FastRMSNorm.load( prefix=f"{prefix}.post_attention_layernorm", weights=weights, eps=config.rms_norm_eps, ) def forward( self, hidden_states, residual, cos, sin, cu_seqlen_prefill, kv_cache, block_tables, slots, input_lengths, max_s, prefill_cache_indices, ): normed_hidden_states, res = self.input_layernorm(hidden_states, residual) # Self Attention attn_output = self.self_attn( normed_hidden_states, cos, sin, cu_seqlen_prefill, kv_cache, block_tables, slots, input_lengths, max_s, prefill_cache_indices, ) # faster post attention rms norm normed_attn_res_output, attn_res = self.post_attention_layernorm( attn_output, res ) moe_output = self.moe(normed_attn_res_output) return moe_output, attn_res class MixtralModel(torch.nn.Module): def __init__(self, config, weights): super().__init__() self.embed_tokens = TensorParallelEmbedding( prefix="model.embed_tokens", weights=weights ) self.layers = nn.ModuleList( [ MixtralLayer( layer_id, config, weights, ) for layer_id in range(config.num_hidden_layers) ] ) self.norm = FastRMSNorm.load( prefix="model.norm", weights=weights, eps=config.rms_norm_eps ) self.head_size = self.layers[0].self_attn.head_size self.num_heads = self.layers[0].self_attn.num_heads self.num_key_value_heads = self.layers[0].self_attn.num_key_value_heads def forward( self, input_ids: torch.Tensor, position_ids: torch.Tensor, cu_seqlen_prefill: Optional[torch.Tensor], kv_cache: List[Tuple[torch.Tensor, torch.Tensor]], block_tables: torch.Tensor, slots: torch.Tensor, input_lengths: torch.Tensor, max_s: int, true_max_s: int, prefill_cache_indices: Optional[torch.Tensor], ) -> torch.Tensor: hidden_states = self.embed_tokens(input_ids) # Get rotary cos and sin for this forward # Avoid to index in each layer cos, sin = self.layers[0].self_attn.rotary_emb.get_cos_sin( position_ids, true_max_s, hidden_states.dtype ) residual = None for i, layer in enumerate(self.layers): hidden_states, residual = layer( hidden_states, residual, cos, sin, cu_seqlen_prefill, kv_cache[i], block_tables, slots, input_lengths, max_s, prefill_cache_indices, ) hidden_states, _ = self.norm(hidden_states, residual) return hidden_states class FlashMixtralForCausalLM(torch.nn.Module): def __init__(self, config, weights): super().__init__() self.model = MixtralModel(config, weights) self.lm_head = TensorParallelHead.load( config, prefix="lm_head", weights=weights, ) self.max_past = config.sliding_window def forward( self, input_ids: torch.Tensor, position_ids: torch.Tensor, cu_seqlen_prefill: Optional[torch.Tensor], kv_cache: List[Tuple[torch.Tensor, torch.Tensor]], block_tables: torch.Tensor, slots: torch.Tensor, input_lengths: torch.Tensor, max_s: int, prefill_cache_indices: Optional[torch.Tensor], lm_head_indices: Optional[torch.Tensor] = None, ) -> torch.Tensor: true_max_s = max_s if prefill_cache_indices is not None: # Slots also need to be sliced as it has the same size as the whole kv tensor slots = slots[prefill_cache_indices] elif self.max_past is not None: # Clamp in decode mode as paged attention requires clamped values whereas the flash attention # kernel requires the true values max_s = min(self.max_past, max_s) input_lengths = torch.clamp(input_lengths, max=self.max_past) hidden_states = self.model( input_ids, position_ids, cu_seqlen_prefill, kv_cache, block_tables, slots, input_lengths, max_s, true_max_s, prefill_cache_indices, ) if lm_head_indices is not None: hidden_states = hidden_states[lm_head_indices] logits = self.lm_head(hidden_states) return logits

text-generation-inference/server/text_generation_server/models/custom_modeling/flash_mixtral_modeling.py/0

import math import time import itertools import torch import torch.distributed import numpy as np from dataclasses import dataclass from opentelemetry import trace from transformers import PreTrainedTokenizerBase from typing import Optional, Tuple, List, Type, Dict from text_generation_server.models import Model from text_generation_server.utils.tokens import batch_top_tokens from text_generation_server.utils.speculate import get_speculate from text_generation_server.models.types import ( Batch, Tokens, Generation, GeneratedText, ) from text_generation_server.models.cache_manager import ( get_cache_manager, set_cache_manager, BLOCK_SIZE, ) from text_generation_server.pb import generate_pb2 from text_generation_server.utils import StoppingCriteria, HeterogeneousNextTokenChooser from text_generation_server.utils.dist import MEMORY_FRACTION tracer = trace.get_tracer(__name__) @dataclass class FlashCausalLMBatch(Batch): batch_id: int requests: List[generate_pb2.Request] # request id -> idx in list mapping requests_idx_mapping: Dict[int, int] # Decoder values input_ids: torch.Tensor position_ids: torch.Tensor speculative_ids: torch.Tensor # Flash Attention values # tensor of length b containing the cumulative sequence lengths of the sequences in the batch, only used in prefill cu_seqlen_prefill: Optional[torch.Tensor] # Paged Attention values # Set when creating the batch # CPU tensor of length b indicating the start of each sequence in slots start_slots: torch.Tensor # tensor of indices of the currently used slots, length = \sum_{i=0}^{b} s_i in prefill, length = b in decode slot_indices: torch.Tensor # List of tuple of ints representing the number of blocks and slots needed by each sequence needed_blocks_slots: Optional[List[Tuple[int, int]]] # Set in prefill by the CacheManager # list of length b of list of length s_i // block_size block_tables: Optional[List[List[int]]] # tensor of size [b, max_seqlen // block_size] holding the paged attention block tables for all sequences block_tables_tensor: Optional[torch.Tensor] # tensor of length \sum_{i=0}^{b} max_s_i holding the paged attention slots for all sequences slots: Optional[torch.Tensor] max_seqlen: int # Prefill metadata tensors to efficiently compute logprobs prefill_head_indices: Optional[torch.Tensor] prefill_next_token_indices: Optional[torch.tensor] prefill_cu_outlens: Optional[List[int]] # All tokens all_input_ids: List[List[int]] all_input_ids_tensor: torch.Tensor # Lengths of all generations present in the batch input_lengths: List[int] input_lengths_tensor: torch.Tensor prefix_offsets: List[Optional[int]] read_offsets: List[Optional[int]] # Generation helpers next_token_chooser: HeterogeneousNextTokenChooser stopping_criterias: List[StoppingCriteria] top_n_tokens: List[int] top_n_tokens_tensor: torch.Tensor # Number of blocks in this batch blocks: int # Maximum number of blocks max_blocks: int def to_pb(self) -> generate_pb2.CachedBatch: return generate_pb2.CachedBatch( id=self.batch_id, request_ids=[r.id for r in self.requests], size=len(self), max_tokens=self.blocks * BLOCK_SIZE, ) @classmethod def from_pb( cls, pb: generate_pb2.Batch, tokenizer: PreTrainedTokenizerBase, dtype: torch.dtype, device: torch.device, ) -> "FlashCausalLMBatch": batch_inputs = [] max_truncation = 0 for r in pb.requests: batch_inputs.append(r.inputs) max_truncation = max(max_truncation, r.truncate) batch_tokenized_inputs = tokenizer( batch_inputs, truncation=True, max_length=max_truncation )["input_ids"] position_ids = [] speculative_ids = [] cu_seqlen_prefill = [0] needed_blocks_slots = [] start_slots = [] slot_indices = [] input_lengths = [] prefix_offsets = [] read_offsets = [] all_input_ids = [] requests_idx_mapping = {} all_prefill_logprobs = True no_prefill_logprobs = True prefill_head_indices = [] prefill_next_token_indices = [] prefill_cu_outlens = [0] next_token_chooser_parameters = [] stopping_criterias = [] top_n_tokens = [] # Cumulative length cumulative_length = 0 cumulative_max_length = 0 prefill_out_cumulative_length = 0 blocks = 0 max_seqlen = 0 max_length = 0 max_blocks = 0 # Parse batch for i, (r, tokenized_input) in enumerate( zip(pb.requests, batch_tokenized_inputs) ): # request id -> idx in list mapping requests_idx_mapping[r.id] = i tokenized_input = tokenized_input[-r.truncate :] input_length = len(tokenized_input) input_lengths.append(input_length) prefix_offsets.append(input_length - 5) read_offsets.append(input_length) all_input_ids.append(tokenized_input) # Position ids request_position_ids = torch.arange(0, input_length, dtype=torch.int32) position_ids.append(request_position_ids) # Add cumulative lengths of all previous inputs cu_seqlen_prefill.append(cumulative_length + input_length) next_token_chooser_parameters.append(r.parameters) stopping_criteria = StoppingCriteria.from_pb( r.stopping_parameters, tokenizer ) max_new_tokens = stopping_criteria.max_new_tokens stopping_criterias.append(stopping_criteria) top_n_tokens.append(r.top_n_tokens) # Paged attention # Remove one as the first token des not have a past speculative_length = get_speculate() total_tokens = input_length + max_new_tokens - 1 + speculative_length needed_blocks = math.ceil(total_tokens / BLOCK_SIZE) blocks += needed_blocks needed_blocks_slots.append((needed_blocks, total_tokens)) start_slots.append(cumulative_max_length) request_slot_indices = torch.arange( cumulative_max_length, cumulative_max_length + input_length, dtype=torch.int64, ) slot_indices.append(request_slot_indices) all_prefill_logprobs = all_prefill_logprobs and r.prefill_logprobs no_prefill_logprobs = no_prefill_logprobs and not r.prefill_logprobs if r.prefill_logprobs: prefill_head_indices.append(request_position_ids + cumulative_length) prefill_next_token_indices.append( prefill_out_cumulative_length + input_length - 1 ) prefill_cu_outlens.append(prefill_out_cumulative_length + input_length) prefill_out_cumulative_length += input_length else: prefill_head_indices.append( torch.tensor( [cumulative_length + input_length - 1], dtype=torch.int32 ) ) prefill_next_token_indices.append(prefill_out_cumulative_length) prefill_cu_outlens.append(prefill_out_cumulative_length + 1) prefill_out_cumulative_length += 1 # Update cumulative_length += input_length cumulative_max_length += total_tokens max_seqlen = max(max_seqlen, input_length) max_blocks = max(max_blocks, needed_blocks) max_length = max( max_length, input_length + max_new_tokens + speculative_length ) next_token_chooser = HeterogeneousNextTokenChooser.from_pb( next_token_chooser_parameters, dtype, device ) start_slots = torch.tensor(start_slots, dtype=torch.int64) # Padded all_input_ids_tensor all_input_ids_tensor = np.zeros( (len(all_input_ids), max_length), dtype=np.int64 ) for i, input_ids in enumerate(all_input_ids): all_input_ids_tensor[i, : len(input_ids)] = input_ids # Create tensors on device all_input_ids_tensor = torch.tensor( all_input_ids_tensor, dtype=torch.int64, device=device ) if len(pb.requests) > 1: input_ids = np.concatenate(all_input_ids, dtype=np.int64) position_ids = torch.cat(position_ids) slot_indices = torch.cat(slot_indices) else: input_ids = all_input_ids[0] position_ids = position_ids[0] slot_indices = slot_indices[0] cu_seqlen_prefill = torch.tensor( cu_seqlen_prefill, device=device, dtype=torch.int32 ) position_ids = position_ids.to(device) slot_indices = slot_indices.to(device) input_ids = torch.tensor(input_ids, dtype=torch.int64, device=device) input_lengths_tensor = torch.tensor( input_lengths, dtype=torch.int32, device=device ) if all_prefill_logprobs: prefill_head_indices = None prefill_next_token_indices = cu_seqlen_prefill[1:] - 1 elif no_prefill_logprobs: prefill_head_indices = cu_seqlen_prefill[1:] - 1 prefill_next_token_indices = None else: prefill_head_indices = torch.tensor( torch.cat(prefill_head_indices), dtype=torch.int64, device=device ) prefill_next_token_indices = torch.tensor( prefill_next_token_indices, dtype=torch.int64, device=device ) top_n_tokens_tensor = torch.tensor( top_n_tokens, device=device, dtype=torch.int64 ) return cls( batch_id=pb.id, requests=pb.requests, requests_idx_mapping=requests_idx_mapping, input_ids=input_ids, position_ids=position_ids, cu_seqlen_prefill=cu_seqlen_prefill, start_slots=start_slots, slot_indices=slot_indices, needed_blocks_slots=needed_blocks_slots, block_tables=None, block_tables_tensor=None, slots=None, max_seqlen=max_seqlen, prefill_head_indices=prefill_head_indices, prefill_next_token_indices=prefill_next_token_indices, prefill_cu_outlens=prefill_cu_outlens, input_lengths=input_lengths, input_lengths_tensor=input_lengths_tensor, prefix_offsets=prefix_offsets, read_offsets=read_offsets, all_input_ids=all_input_ids, all_input_ids_tensor=all_input_ids_tensor, next_token_chooser=next_token_chooser, stopping_criterias=stopping_criterias, top_n_tokens=top_n_tokens, top_n_tokens_tensor=top_n_tokens_tensor, blocks=blocks, max_blocks=max_blocks, speculative_ids=None, ) @tracer.start_as_current_span("filter") def filter(self, request_ids: List[int]) -> "FlashCausalLMBatch": if len(request_ids) == 0: raise ValueError("Batch must have at least one request") # We assume that if len(requests) == len(self) then the requests are the same if len(request_ids) == len(self): return self device = self.input_ids.device # New values after filtering requests_idx_mapping = {} # Used to index into tensors indices = [] # slots to keep after filtering slot_filtering_indices = torch.zeros( self.slots.shape[0], dtype=torch.bool, device=device ) # Create on CPU to only move to GPU once instead of at every copy slot_indices = torch.empty(len(request_ids), dtype=torch.int64) max_seqlen = 0 requests = [] start_slots = [] block_tables = [] all_input_ids = [] input_lengths = [] prefix_offsets = [] read_offsets = [] stopping_criterias = [] top_n_tokens = [] blocks = 0 max_blocks = 0 # Cumulative length cumulative_max_length = 0 for i, request_id in enumerate(request_ids): idx = self.requests_idx_mapping[request_id] indices.append(idx) requests_idx_mapping[request_id] = i requests.append(self.requests[idx]) # Get length request_input_length = self.input_lengths[idx] max_seqlen = max(max_seqlen, request_input_length) all_input_ids.append(self.all_input_ids[idx]) input_lengths.append(request_input_length) prefix_offsets.append(self.prefix_offsets[idx]) read_offsets.append(self.read_offsets[idx]) stopping_criteria = self.stopping_criterias[idx] stopping_criterias.append(stopping_criteria) top_n_tokens.append(self.top_n_tokens[idx]) remaining_tokens = ( stopping_criteria.max_new_tokens - stopping_criteria.current_tokens ) request_block_table = self.block_tables[idx] blocks += len(request_block_table) block_tables.append(request_block_table) start_slots.append(cumulative_max_length) # Copy to tensor (CPU) slot_indices[i] = cumulative_max_length + request_input_length - 1 # Set slice slot_filtering_indices[ self.start_slots[idx] : self.start_slots[idx] + request_input_length + remaining_tokens - 1 ] = True cumulative_max_length += request_input_length + remaining_tokens - 1 max_blocks = max(max_blocks, len(request_block_table)) block_indices_to_free = [] # Iterate on all requests for i, r in enumerate(self.requests): # Filter requests that are not part of the new batch if r.id not in requests_idx_mapping.keys(): block_indices_to_free.extend(self.block_tables[i]) # Free blocks get_cache_manager().free(block_indices_to_free) # Needed to avoid dropping blocks when the batches will go out of scope self.block_tables = None # Index into tensors input_ids = self.input_ids[indices] position_ids = self.position_ids[indices] all_input_ids_tensor = self.all_input_ids_tensor[indices] block_tables_tensor = self.block_tables_tensor[indices] input_lengths_tensor = self.input_lengths_tensor[indices] slots = self.slots[slot_filtering_indices] next_token_chooser = self.next_token_chooser.filter(indices) top_n_tokens_tensor = self.top_n_tokens_tensor[indices] speculative_ids = ( self.speculative_ids[indices] if self.speculative_ids is not None else None ) start_slots = torch.tensor(start_slots, dtype=torch.int64) # Move to GPU now that we have the whole tensor slot_indices = slot_indices.to(device) return type(self)( batch_id=self.batch_id, requests=requests, requests_idx_mapping=requests_idx_mapping, input_ids=input_ids, position_ids=position_ids, cu_seqlen_prefill=None, start_slots=start_slots, slot_indices=slot_indices, needed_blocks_slots=None, block_tables=block_tables, block_tables_tensor=block_tables_tensor, slots=slots, max_seqlen=max_seqlen, prefill_head_indices=None, prefill_next_token_indices=None, prefill_cu_outlens=None, input_lengths=input_lengths, input_lengths_tensor=input_lengths_tensor, prefix_offsets=prefix_offsets, read_offsets=read_offsets, all_input_ids=all_input_ids, all_input_ids_tensor=all_input_ids_tensor, next_token_chooser=next_token_chooser, stopping_criterias=stopping_criterias, top_n_tokens=top_n_tokens, top_n_tokens_tensor=top_n_tokens_tensor, blocks=blocks, max_blocks=max_blocks, speculative_ids=speculative_ids, ) @classmethod @tracer.start_as_current_span("concatenate") def concatenate(cls, batches: List["FlashCausalLMBatch"]) -> "FlashCausalLMBatch": # Batch attributes requests = [] requests_idx_mapping = {} blocks = 0 total_batch_size = 0 total_slots = 0 max_blocks = 0 max_length = 0 max_seqlen = 0 for b in batches: total_batch_size += len(b) total_slots += len(b.slots) blocks += b.blocks speculative_length = ( b.speculative_ids.shape[1] if b.speculative_ids is not None else 0 ) max_blocks = max(max_blocks, b.max_blocks) max_seqlen = max(max_seqlen, b.max_seqlen) max_length = max( max_length, max( input_length + stopping_criteria.max_new_tokens + speculative_length - stopping_criteria.current_tokens for input_length, stopping_criteria in zip( b.input_lengths, b.stopping_criterias ) ), ) input_ids = batches[0].input_ids.new_empty(total_batch_size) position_ids = batches[0].position_ids.new_empty(total_batch_size) slots = batches[0].slots.new_empty(total_slots) slot_indices = batches[0].slot_indices.new_empty(total_batch_size) input_lengths_tensor = batches[0].input_lengths_tensor.new_empty( total_batch_size ) block_tables_tensor = batches[0].block_tables_tensor.new_zeros( (total_batch_size, max_blocks) ) all_input_ids_tensor = batches[0].all_input_ids_tensor.new_zeros( (total_batch_size, max_length) ) top_n_tokens_tensor = batches[0].top_n_tokens_tensor.new_zeros( total_batch_size, ) start_slots = [] block_tables = [] all_input_ids = [] input_lengths = [] prefix_offsets = [] read_offsets = [] next_token_chooser_parameters = [] stopping_criterias = [] top_n_tokens = [] # Cumulative length cumulative_batch_size = 0 cumulative_slots = 0 for i, batch in enumerate(batches): requests.extend(batch.requests) if i == 0: requests_idx_mapping = batch.requests_idx_mapping else: # We need to offset the mapping for each batch by the cumulative batch size for k, v in batch.requests_idx_mapping.items(): requests_idx_mapping[k] = v + cumulative_batch_size start_index = cumulative_batch_size end_index = cumulative_batch_size + len(batch) slots_start_index = cumulative_slots slots_end_index = cumulative_slots + len(batch.slots) # Copy tensors (GPU) input_ids[start_index:end_index] = batch.input_ids position_ids[start_index:end_index] = batch.position_ids slot_indices[start_index:end_index] = batch.slot_indices + cumulative_slots input_lengths_tensor[start_index:end_index] = batch.input_lengths_tensor top_n_tokens_tensor[start_index:end_index] = batch.top_n_tokens_tensor slots[slots_start_index:slots_end_index] = batch.slots all_input_ids_tensor[ start_index:end_index, : batch.all_input_ids_tensor.shape[1] ] = batch.all_input_ids_tensor[:, :max_length] block_tables_tensor[ start_index:end_index, : batch.block_tables_tensor.shape[1] ] = batch.block_tables_tensor[:, :max_blocks] start_slots.append(batch.start_slots + cumulative_slots) block_tables.extend(batch.block_tables) all_input_ids.extend(batch.all_input_ids) input_lengths.extend(batch.input_lengths) prefix_offsets.extend(batch.prefix_offsets) read_offsets.extend(batch.read_offsets) next_token_chooser_parameters.extend([r.parameters for r in batch.requests]) stopping_criterias.extend(batch.stopping_criterias) top_n_tokens.extend(batch.top_n_tokens) # Update cumulative_batch_size += len(batch) cumulative_slots += len(batch.slots) start_slots = torch.concat(start_slots) next_token_chooser = HeterogeneousNextTokenChooser.from_pb( next_token_chooser_parameters, dtype=batches[0].next_token_chooser.dtype, device=batches[0].next_token_chooser.device, ) speculative_ids = ( torch.cat([b.speculative_ids for b in batches], dim=0) if batches[0].speculative_ids is not None else None ) # Needed to avoid dropping blocks when the batches will go out of scope for b in batches: b.block_tables = None del b return cls( batch_id=batches[0].batch_id, requests=requests, requests_idx_mapping=requests_idx_mapping, input_ids=input_ids, position_ids=position_ids, cu_seqlen_prefill=None, start_slots=start_slots, slot_indices=slot_indices, needed_blocks_slots=None, block_tables=block_tables, block_tables_tensor=block_tables_tensor, slots=slots, max_seqlen=max_seqlen, prefill_head_indices=None, prefill_next_token_indices=None, prefill_cu_outlens=None, input_lengths=input_lengths, input_lengths_tensor=input_lengths_tensor, prefix_offsets=prefix_offsets, read_offsets=read_offsets, all_input_ids=all_input_ids, all_input_ids_tensor=all_input_ids_tensor, next_token_chooser=next_token_chooser, stopping_criterias=stopping_criterias, top_n_tokens=top_n_tokens, top_n_tokens_tensor=top_n_tokens_tensor, blocks=blocks, max_blocks=max_blocks, speculative_ids=speculative_ids, ) def __del__(self): if self.block_tables is not None and self.block_tables: # Free blocks get_cache_manager().free( list(itertools.chain.from_iterable(self.block_tables)) ) def __len__(self): return len(self.requests) class FlashCausalLM(Model): def __init__( self, model: torch.nn.Module, tokenizer: PreTrainedTokenizerBase, num_layers: int, num_kv_heads: int, head_size: int, dtype: torch.dtype, device: torch.device, rank: int = 0, world_size: int = 1, sliding_window: Optional[int] = None, ): self.num_layers = num_layers self.num_kv_heads = num_kv_heads self.head_size = head_size super(FlashCausalLM, self).__init__( model=model, tokenizer=tokenizer, requires_padding=False, dtype=dtype, device=device, rank=rank, world_size=world_size, sliding_window=sliding_window, ) @property def batch_type(self) -> Type[FlashCausalLMBatch]: return FlashCausalLMBatch def warmup(self, batch: FlashCausalLMBatch): torch.cuda.empty_cache() try: cache_manager = set_cache_manager( batch.blocks, self.num_layers, self.num_kv_heads, self.head_size, self.sliding_window is not None, self.dtype, self.device, ) _, batch, _ = self.generate_token(batch) except torch.cuda.OutOfMemoryError as e: raise RuntimeError( f"Not enough memory to handle {len(batch.input_ids)} prefill tokens. " f"You need to decrease `--max-batch-prefill-tokens`" ) from e torch.cuda.synchronize(self.device) # Inspired by the original implementation in [vllm](https://github.com/vllm-project/vllm) # Calculate the number of blocks that can be allocated with the free memory dtype_size = torch.tensor([], dtype=self.dtype).element_size() cache_block_size = BLOCK_SIZE * self.num_kv_heads * self.head_size total_cache_size = self.num_layers * cache_block_size * 2 * dtype_size total_free_memory, _ = torch.cuda.mem_get_info(self.device) total_gpu_memory = torch.cuda.get_device_properties(self.device).total_memory free_memory = max( 0, total_free_memory - (1 - MEMORY_FRACTION) * total_gpu_memory ) num_blocks = ( int(free_memory // total_cache_size) # Add batch.blocks as we allocated it above, so it is included in the peak memory. + cache_manager.num_blocks ) del batch del cache_manager set_cache_manager( num_blocks, self.num_layers, self.num_kv_heads, self.head_size, self.sliding_window is not None, self.dtype, self.device, ) return int(num_blocks * BLOCK_SIZE) def forward(self, batch: FlashCausalLMBatch) -> Tuple[torch.Tensor, torch.Tensor]: # Model Forward if batch.speculative_ids is not None: input_ids = batch.input_ids position_ids = batch.position_ids cu_seqlen_prefill = batch.cu_seqlen_prefill kv_cache = get_cache_manager().kv_cache block_tables = batch.block_tables_tensor slots = batch.slots[batch.slot_indices] input_lengths = batch.input_lengths_tensor max_s = batch.max_seqlen lm_head_indices = batch.prefill_head_indices speculative_ids = batch.speculative_ids B, speculative_length = speculative_ids.shape new_length = speculative_length + 1 new_input_ids = torch.cat( [input_ids.unsqueeze(-1), speculative_ids], dim=1 ).reshape(-1) arange = torch.arange(new_length, device=position_ids.device).unsqueeze(0) arange_int = arange.to(dtype=torch.int32) new_position_ids = ( position_ids.unsqueeze(-1).expand(B, new_length) + arange ).view(-1) slots = (slots.unsqueeze(-1).expand(B, new_length) + arange_int).view(-1) input_lengths = ( input_lengths.unsqueeze(-1).expand(B, new_length) + arange_int ).view(-1) # Add Copy the block tables for all members block_tables = ( block_tables.unsqueeze(1) .expand(B, new_length, -1) .reshape(B * new_length, -1) .contiguous() ) max_s = max_s + speculative_length input_ids = new_input_ids position_ids = new_position_ids else: input_ids = batch.input_ids position_ids = batch.position_ids cu_seqlen_prefill = batch.cu_seqlen_prefill kv_cache = get_cache_manager().kv_cache block_tables = batch.block_tables_tensor slots = batch.slots[batch.slot_indices] input_lengths = batch.input_lengths_tensor max_s = batch.max_seqlen lm_head_indices = batch.prefill_head_indices return self.model.forward( input_ids=input_ids, position_ids=position_ids, cu_seqlen_prefill=cu_seqlen_prefill, kv_cache=kv_cache, block_tables=block_tables, slots=slots, input_lengths=input_lengths, max_s=max_s, lm_head_indices=lm_head_indices, ) @tracer.start_as_current_span("generate_token") def generate_token( self, batch: FlashCausalLMBatch ) -> Tuple[List[Generation], Optional[FlashCausalLMBatch], Tuple[int, int]]: start = time.time_ns() prefill = batch.cu_seqlen_prefill is not None prefill_logprobs = batch.prefill_next_token_indices is not None if batch.needed_blocks_slots: # Allocate blocks to this batch block_tables, block_tables_tensor, slots = get_cache_manager().allocate( batch.needed_blocks_slots, batch.blocks, batch.max_blocks, batch.input_ids.device, ) batch.needed_blocks_slots = None batch.block_tables = block_tables batch.block_tables_tensor = block_tables_tensor batch.slots = slots try: out = self.forward(batch) except Exception as e: del batch raise e if isinstance(out, tuple): out, speculative_logits = out else: speculative_logits = None if prefill: next_token_logits = ( out[batch.prefill_next_token_indices] if prefill_logprobs else out ) if speculative_logits is not None: speculative_logits = ( speculative_logits[batch.prefill_next_token_indices] if prefill_logprobs else speculative_logits ) else: next_token_logits = out speculate = get_speculate() ( next_input_ids, next_token_logprobs, logprobs, accepted_ids, speculative_ids, ) = batch.next_token_chooser( batch.all_input_ids_tensor[:, : batch.max_seqlen], next_token_logits, speculate, batch.speculative_ids, speculative_logits, ) batch_top_token_ids, batch_top_token_logprobs = batch_top_tokens( batch.top_n_tokens, batch.top_n_tokens_tensor, logprobs, accepted_ids ) if prefill: if len(batch) > 1 and prefill_logprobs: # We create the prefill_tokens_indices tensor that will be used to gather prefill logprobs # When batch == 1, we will just use the batch.input_ids values directly prefill_tokens_indices = batch.input_ids.new_zeros(len(out)) next_position_ids = batch.position_ids.new_empty(len(batch)) batch.slot_indices = batch.slot_indices[batch.cu_seqlen_prefill[1:] - 1] # We do not need cu_seqlen_prefill anymore batch.cu_seqlen_prefill = None else: prefill_logprobs = None next_position_ids = batch.position_ids # Cumulative length cumulative_length = 0 # Results generations: List[Generation] = [] stopped = True # Zipped iterator iterator = zip(batch.input_lengths, batch.all_input_ids, accepted_ids) # We do two for loops as the first one can run completely asynchronously from the GPU while for the second # one, we need to first do a GPU <-> CPU sync # It is faster if we delay this sync for the maximum amount of time # For each member of the batch index = 0 for i, (input_length, all_input_ids, n_accepted_ids) in enumerate(iterator): # Indexing metadata start_index = cumulative_length end_index = cumulative_length + input_length if prefill: # Indexing metadata out_start_index = batch.prefill_cu_outlens[i] out_end_index = batch.prefill_cu_outlens[i + 1] out_length = out_end_index - out_start_index # Initialize position_ids # In decode, we do not need this as we can just increment position ids next_position_ids[i] = batch.position_ids[end_index - 1] # Used to gather prefill logprobs # Copy batch.input_ids to prefill_token_indices if prefill_logprobs: if len(batch) > 1: prefill_tokens_indices[ out_start_index : out_end_index - 1 ] = batch.input_ids[start_index + 1 : start_index + out_length] else: # Set prefill_tokens_indices to the correct slice prefill_tokens_indices = batch.input_ids[ start_index + 1 : start_index + out_length ] for j in range(n_accepted_ids): batch.all_input_ids_tensor[i, input_length + j] = next_input_ids[index] index += 1 cumulative_length += input_length batch.input_ids = next_input_ids[accepted_ids.cumsum(dim=-1) - 1] batch.speculative_ids = speculative_ids batch.position_ids = next_position_ids + accepted_ids batch.input_lengths_tensor += accepted_ids batch.slot_indices += accepted_ids if prefill and prefill_logprobs: # Get prefill logprobs prefill_logprobs_tensor = torch.log_softmax(out, -1) prefill_logprobs = torch.gather( prefill_logprobs_tensor, 1, prefill_tokens_indices.view(-1, 1) ) # GPU <-> CPU sync prefill_logprobs = prefill_logprobs.view(-1).tolist() # GPU <-> CPU sync next_token_logprobs = next_token_logprobs.tolist() next_token_ids = next_input_ids.tolist() accepted_ids = accepted_ids.tolist() start_decode = time.time_ns() # Zipped iterator iterator = zip( batch.requests, batch.input_lengths, batch.prefix_offsets, batch.read_offsets, batch.stopping_criterias, batch.all_input_ids, batch.next_token_chooser.do_sample, batch.next_token_chooser.seeds, batch.top_n_tokens, accepted_ids, batch_top_token_ids, batch_top_token_logprobs, ) # For each member of the batch index = 0 for i, ( request, input_length, prefix_offset, read_offset, stopping_criteria, all_input_ids, do_sample, seed, top_n_tokens, n_accepted_ids, top_token_ids, top_token_logprobs, ) in enumerate(iterator): # Append next token to all tokens next_token_texts = [] left = 0 current_stopped = False for j in range(index, index + n_accepted_ids): # Generated token next_token_id = next_token_ids[j] all_input_ids.append(next_token_id) next_token_text, prefix_offset, read_offset = self.decode_token( all_input_ids, prefix_offset, read_offset, ) next_token_texts.append(next_token_text) stop, reason = stopping_criteria( next_token_id, next_token_text, ) if stop: left = index + n_accepted_ids - j - 1 current_stopped = True break else: current_stopped = False stopped = stopped and current_stopped _next_token_ids = next_token_ids[index : index + n_accepted_ids - left] _next_token_logprobs = next_token_logprobs[ index : index + n_accepted_ids - left ] index += n_accepted_ids # Shard generations # All generations will be appended in the rust sharded client if i % self.world_size == self.rank: if stop: # Decode generated tokens output_text, _, _ = self.decode_token( all_input_ids, prefix_offset=len(all_input_ids) - stopping_criteria.current_tokens - 1, read_offset=len(all_input_ids) - stopping_criteria.current_tokens, skip_special_tokens=True, ) generated_text = GeneratedText( output_text, stopping_criteria.current_tokens, reason, seed if do_sample else None, ) else: generated_text = None # Prefill if prefill and request.prefill_logprobs: out_start_index = batch.prefill_cu_outlens[i] out_end_index = batch.prefill_cu_outlens[i + 1] # Remove generated token to only have prefill and add nan for first prompt token request_prefill_logprobs = [float("nan")] + prefill_logprobs[ out_start_index : out_end_index - 1 ] prefill_token_ids = all_input_ids[:-1] prefill_texts = self.tokenizer.batch_decode( prefill_token_ids, clean_up_tokenization_spaces=False, skip_special_tokens=False, ) prefill_tokens = Tokens( prefill_token_ids, request_prefill_logprobs, prefill_texts, is_special=[], ) else: prefill_tokens = None if top_n_tokens > 0: all_top_tokens = [] for (top_token_ids, top_token_logprobs) in zip(top_token_ids, top_token_logprobs): toptoken_texts = self.tokenizer.batch_decode( top_token_ids, clean_up_tokenization_spaces=False, skip_special_tokens=False, ) special_toptokens = [ token_id in self.all_special_ids for token_id in top_token_ids ] top_tokens = Tokens( top_token_ids, top_token_logprobs, toptoken_texts, special_toptokens, ) all_top_tokens.append(top_tokens) top_tokens = all_top_tokens else: top_tokens = None generation = Generation( request.id, prefill_tokens, Tokens( _next_token_ids, _next_token_logprobs, next_token_texts, [nid in self.all_special_ids for nid in _next_token_ids], ), generated_text, top_tokens, ) generations.append(generation) # Update values batch.input_lengths[i] = input_length + n_accepted_ids if batch.input_lengths[i] > batch.max_seqlen: batch.max_seqlen = batch.input_lengths[i] batch.prefix_offsets[i] = prefix_offset batch.read_offsets[i] = read_offset batch.all_input_ids[i] = all_input_ids if stopped: del batch # No need to return a batch if we know that all requests stopped forward_ns = start_decode - start decode_ns = time.time_ns() - start_decode return generations, None, (forward_ns, decode_ns) batch.prefill_cu_outlens = None batch.prefill_head_indices = None batch.prefill_next_token_indices = None forward_ns = start_decode - start decode_ns = time.time_ns() - start_decode return generations, batch, (forward_ns, decode_ns)

text-generation-inference/server/text_generation_server/models/flash_causal_lm.py/0

import torch from transformers import AutoTokenizer, AutoModelForCausalLM from typing import List, Optional, Tuple from text_generation_server.models import CausalLM class RW(CausalLM): def __init__( self, model_id: str, revision: Optional[str] = None, quantize: Optional[str] = None, dtype: Optional[torch.dtype] = None, trust_remote_code: bool = False, ): if torch.cuda.is_available(): device = torch.device("cuda") dtype = torch.float16 if dtype is None else dtype else: if quantize: raise ValueError("quantization is not available on CPU") device = torch.device("cpu") dtype = torch.float32 if dtype is None else dtype tokenizer = AutoTokenizer.from_pretrained( model_id, revision=revision, padding_side="left", truncation_side="left", trust_remote_code=trust_remote_code, ) model = AutoModelForCausalLM.from_pretrained( model_id, revision=revision, torch_dtype=dtype, device_map="auto" if torch.cuda.is_available() and torch.cuda.device_count() > 1 else None, load_in_8bit=quantize == "bitsandbytes", trust_remote_code=trust_remote_code, ) if torch.cuda.is_available() and torch.cuda.device_count() == 1: model = model.cuda() if tokenizer.pad_token_id is None: if model.config.pad_token_id is not None: tokenizer.pad_token_id = model.config.pad_token_id elif model.config.eos_token_id is not None: tokenizer.pad_token_id = model.config.eos_token_id elif tokenizer.eos_token_id is not None: tokenizer.pad_token_id = tokenizer.eos_token_id else: tokenizer.add_special_tokens({"pad_token": "[PAD]"}) super(CausalLM, self).__init__( model=model, tokenizer=tokenizer, requires_padding=True, dtype=dtype, device=device, ) def forward( self, input_ids, attention_mask, position_ids, past_key_values: Optional = None ) -> Tuple[torch.Tensor, List[Tuple[torch.Tensor, torch.Tensor]]]: # Model Forward outputs = self.model.forward( input_ids=input_ids, attention_mask=attention_mask, past_key_values=past_key_values, ) return outputs.logits, outputs.past_key_values

text-generation-inference/server/text_generation_server/models/rw.py/0

import math import numpy as np import torch import torch.nn as nn from torch.cuda.amp import custom_bwd, custom_fwd try: import triton import triton.language as tl from . import custom_autotune # code based https://github.com/fpgaminer/GPTQ-triton @custom_autotune.autotune( configs=[ triton.Config( { "BLOCK_SIZE_M": 64, "BLOCK_SIZE_N": 256, "BLOCK_SIZE_K": 32, "GROUP_SIZE_M": 8, }, num_stages=4, num_warps=4, ), triton.Config( { "BLOCK_SIZE_M": 128, "BLOCK_SIZE_N": 128, "BLOCK_SIZE_K": 32, "GROUP_SIZE_M": 8, }, num_stages=4, num_warps=4, ), triton.Config( { "BLOCK_SIZE_M": 64, "BLOCK_SIZE_N": 128, "BLOCK_SIZE_K": 32, "GROUP_SIZE_M": 8, }, num_stages=4, num_warps=4, ), triton.Config( { "BLOCK_SIZE_M": 128, "BLOCK_SIZE_N": 32, "BLOCK_SIZE_K": 32, "GROUP_SIZE_M": 8, }, num_stages=4, num_warps=4, ), triton.Config( { "BLOCK_SIZE_M": 64, "BLOCK_SIZE_N": 64, "BLOCK_SIZE_K": 32, "GROUP_SIZE_M": 8, }, num_stages=4, num_warps=4, ), triton.Config( { "BLOCK_SIZE_M": 64, "BLOCK_SIZE_N": 128, "BLOCK_SIZE_K": 32, "GROUP_SIZE_M": 8, }, num_stages=2, num_warps=8, ), triton.Config( { "BLOCK_SIZE_M": 64, "BLOCK_SIZE_N": 64, "BLOCK_SIZE_K": 64, "GROUP_SIZE_M": 8, }, num_stages=3, num_warps=8, ), triton.Config( { "BLOCK_SIZE_M": 32, "BLOCK_SIZE_N": 32, "BLOCK_SIZE_K": 128, "GROUP_SIZE_M": 8, }, num_stages=2, num_warps=4, ), ], key=["M", "N", "K"], nearest_power_of_two=True, prune_configs_by={ "early_config_prune": custom_autotune.matmul248_kernel_config_pruner, "perf_model": None, "top_k": None, }, ) @triton.jit def matmul_248_kernel( a_ptr, b_ptr, c_ptr, scales_ptr, zeros_ptr, g_ptr, M, N, K, bits, maxq, stride_am, stride_ak, stride_bk, stride_bn, stride_cm, stride_cn, stride_scales, stride_zeros, BLOCK_SIZE_M: tl.constexpr, BLOCK_SIZE_N: tl.constexpr, BLOCK_SIZE_K: tl.constexpr, GROUP_SIZE_M: tl.constexpr, ): """ Compute the matrix multiplication C = A x B. A is of shape (M, K) float16 B is of shape (K//8, N) int32 C is of shape (M, N) float16 scales is of shape (G, N) float16 zeros is of shape (G, N) float16 g_ptr is of shape (K) int32 """ infearure_per_bits = 32 // bits pid = tl.program_id(axis=0) num_pid_m = tl.cdiv(M, BLOCK_SIZE_M) num_pid_n = tl.cdiv(N, BLOCK_SIZE_N) num_pid_k = tl.cdiv(K, BLOCK_SIZE_K) num_pid_in_group = GROUP_SIZE_M * num_pid_n group_id = pid // num_pid_in_group first_pid_m = group_id * GROUP_SIZE_M group_size_m = min(num_pid_m - first_pid_m, GROUP_SIZE_M) pid_m = first_pid_m + (pid % group_size_m) pid_n = (pid % num_pid_in_group) // group_size_m offs_am = pid_m * BLOCK_SIZE_M + tl.arange(0, BLOCK_SIZE_M) offs_bn = pid_n * BLOCK_SIZE_N + tl.arange(0, BLOCK_SIZE_N) offs_k = tl.arange(0, BLOCK_SIZE_K) a_ptrs = a_ptr + ( offs_am[:, None] * stride_am + offs_k[None, :] * stride_ak ) # (BLOCK_SIZE_M, BLOCK_SIZE_K) a_mask = offs_am[:, None] < M # b_ptrs is set up such that it repeats elements along the K axis 8 times b_ptrs = b_ptr + ( (offs_k[:, None] // infearure_per_bits) * stride_bk + offs_bn[None, :] * stride_bn ) # (BLOCK_SIZE_K, BLOCK_SIZE_N) g_ptrs = g_ptr + offs_k # shifter is used to extract the N bits of each element in the 32-bit word from B scales_ptrs = scales_ptr + offs_bn[None, :] zeros_ptrs = zeros_ptr + (offs_bn[None, :] // infearure_per_bits) shifter = (offs_k % infearure_per_bits) * bits zeros_shifter = (offs_bn % infearure_per_bits) * bits accumulator = tl.zeros((BLOCK_SIZE_M, BLOCK_SIZE_N), dtype=tl.float32) for k in range(0, num_pid_k): g_idx = tl.load(g_ptrs) # Fetch scales and zeros; these are per-outfeature and thus reused in the inner loop scales = tl.load( scales_ptrs + g_idx[:, None] * stride_scales ) # (BLOCK_SIZE_K, BLOCK_SIZE_N,) zeros = tl.load( zeros_ptrs + g_idx[:, None] * stride_zeros ) # (BLOCK_SIZE_K, BLOCK_SIZE_N,) zeros = (zeros >> zeros_shifter[None, :]) & maxq zeros = zeros + 1 a = tl.load(a_ptrs, mask=a_mask, other=0.0) # (BLOCK_SIZE_M, BLOCK_SIZE_K) b = tl.load(b_ptrs) # (BLOCK_SIZE_K, BLOCK_SIZE_N), but repeated # Now we need to unpack b (which is N-bit values) into 32-bit values b = (b >> shifter[:, None]) & maxq # Extract the N-bit values b = (b - zeros) * scales # Scale and shift accumulator += tl.dot(a, b) a_ptrs += BLOCK_SIZE_K b_ptrs += (BLOCK_SIZE_K // infearure_per_bits) * stride_bk g_ptrs += BLOCK_SIZE_K c_ptrs = c_ptr + stride_cm * offs_am[:, None] + stride_cn * offs_bn[None, :] c_mask = (offs_am[:, None] < M) & (offs_bn[None, :] < N) tl.store(c_ptrs, accumulator, mask=c_mask) except: print("triton not installed.") def matmul248(input, qweight, scales, qzeros, g_idx, bits, maxq): with torch.cuda.device(input.device): output = torch.empty( (input.shape[0], qweight.shape[1]), device=input.device, dtype=torch.float16 ) grid = lambda META: ( triton.cdiv(input.shape[0], META["BLOCK_SIZE_M"]) * triton.cdiv(qweight.shape[1], META["BLOCK_SIZE_N"]), ) matmul_248_kernel[grid]( input, qweight, output, scales, qzeros, g_idx, input.shape[0], qweight.shape[1], input.shape[1], bits, maxq, input.stride(0), input.stride(1), qweight.stride(0), qweight.stride(1), output.stride(0), output.stride(1), scales.stride(0), qzeros.stride(0), ) return output class QuantLinearFunction(torch.autograd.Function): @staticmethod @custom_fwd(cast_inputs=torch.float16) def forward(ctx, input, qweight, scales, qzeros, g_idx, bits, maxq): output = matmul248(input, qweight, scales, qzeros, g_idx, bits, maxq) return output class QuantLinear(nn.Module): def __init__(self, qweight, qzeros, scales, g_idx, bias, bits, groupsize): super().__init__() self.register_buffer("qweight", qweight) self.register_buffer("qzeros", qzeros) self.register_buffer("scales", scales) self.register_buffer("g_idx", g_idx) if bias is not None: self.register_buffer("bias", bias) else: self.bias = None if bits not in [2, 4, 8]: raise NotImplementedError("Only 2,4,8 bits are supported.") self.bits = bits self.maxq = 2**self.bits - 1 self.groupsize = groupsize self.outfeatures = qweight.shape[1] self.infeatures = qweight.shape[0] * 32 // bits @classmethod def new(cls, bits, groupsize, infeatures, outfeatures, bias): if bits not in [2, 4, 8]: raise NotImplementedError("Only 2,4,8 bits are supported.") qweight = torch.zeros((infeatures // 32 * bits, outfeatures), dtype=torch.int32) qzeros = torch.zeros( (math.ceil(infeatures / groupsize), outfeatures // 32 * bits), dtype=torch.int32, ) scales = torch.zeros( (math.ceil(infeatures / groupsize), outfeatures), dtype=torch.float16 ) g_idx = torch.tensor( [i // groupsize for i in range(infeatures)], dtype=torch.int32 ) if bias: bias = torch.zeros((outfeatures), dtype=torch.float16) else: bias = None return cls(qweight, qzeros, scales, g_idx, bias, bits, groupsize) def pack(self, linear, scales, zeros, g_idx=None): self.g_idx = g_idx.clone() if g_idx is not None else self.g_idx scales = scales.t().contiguous() zeros = zeros.t().contiguous() scale_zeros = zeros * scales self.scales = scales.clone().half() if linear.bias is not None: self.bias = linear.bias.clone().half() intweight = [] for idx in range(self.infeatures): intweight.append( torch.round( (linear.weight.data[:, idx] + scale_zeros[self.g_idx[idx]]) / self.scales[self.g_idx[idx]] ).to(torch.int)[:, None] ) intweight = torch.cat(intweight, dim=1) intweight = intweight.t().contiguous() intweight = intweight.numpy().astype(np.uint32) qweight = np.zeros( (intweight.shape[0] // 32 * self.bits, intweight.shape[1]), dtype=np.uint32 ) i = 0 row = 0 while row < qweight.shape[0]: if self.bits in [2, 4, 8]: for j in range(i, i + (32 // self.bits)): qweight[row] |= intweight[j] << (self.bits * (j - i)) i += 32 // self.bits row += 1 else: raise NotImplementedError("Only 2,4,8 bits are supported.") qweight = qweight.astype(np.int32) self.qweight = torch.from_numpy(qweight) zeros -= 1 zeros = zeros.numpy().astype(np.uint32) qzeros = np.zeros( (zeros.shape[0], zeros.shape[1] // 32 * self.bits), dtype=np.uint32 ) i = 0 col = 0 while col < qzeros.shape[1]: if self.bits in [2, 4, 8]: for j in range(i, i + (32 // self.bits)): qzeros[:, col] |= zeros[:, j] << (self.bits * (j - i)) i += 32 // self.bits col += 1 else: raise NotImplementedError("Only 2,4,8 bits are supported.") qzeros = qzeros.astype(np.int32) self.qzeros = torch.from_numpy(qzeros) def forward(self, x): out_shape = x.shape[:-1] + (self.outfeatures,) out = QuantLinearFunction.apply( x.reshape(-1, x.shape[-1]), self.qweight, self.scales, self.qzeros, self.g_idx, self.bits, self.maxq, ) out = out + self.bias if self.bias is not None else out return out.reshape(out_shape)

text-generation-inference/server/text_generation_server/utils/gptq/quant_linear.py/0

# EditorConfig helps developers define and maintain consistent # coding styles between different editors or IDEs # http://editorconfig.org root = true [*] indent_style = space indent_size = 2 end_of_line = lf charset = utf-8 trim_trailing_whitespace = true insert_final_newline = true [*.md] trim_trailing_whitespace = false

tokenizers/bindings/node/.editorconfig/0

/* tslint:disable */ /* eslint-disable */ /* prettier-ignore */ /* auto-generated by NAPI-RS */ const { existsSync, readFileSync } = require('fs') const { join } = require('path') const { platform, arch } = process let nativeBinding = null let localFileExisted = false let loadError = null function isMusl() { // For Node 10 if (!process.report || typeof process.report.getReport !== 'function') { try { const lddPath = require('child_process').execSync('which ldd').toString().trim() return readFileSync(lddPath, 'utf8').includes('musl') } catch (e) { return true } } else { const { glibcVersionRuntime } = process.report.getReport().header return !glibcVersionRuntime } } switch (platform) { case 'android': switch (arch) { case 'arm64': localFileExisted = existsSync(join(__dirname, 'tokenizers.android-arm64.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.android-arm64.node') } else { nativeBinding = require('tokenizers-android-arm64') } } catch (e) { loadError = e } break case 'arm': localFileExisted = existsSync(join(__dirname, 'tokenizers.android-arm-eabi.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.android-arm-eabi.node') } else { nativeBinding = require('tokenizers-android-arm-eabi') } } catch (e) { loadError = e } break default: throw new Error(`Unsupported architecture on Android ${arch}`) } break case 'win32': switch (arch) { case 'x64': localFileExisted = existsSync(join(__dirname, 'tokenizers.win32-x64-msvc.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.win32-x64-msvc.node') } else { nativeBinding = require('tokenizers-win32-x64-msvc') } } catch (e) { loadError = e } break case 'ia32': localFileExisted = existsSync(join(__dirname, 'tokenizers.win32-ia32-msvc.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.win32-ia32-msvc.node') } else { nativeBinding = require('tokenizers-win32-ia32-msvc') } } catch (e) { loadError = e } break case 'arm64': localFileExisted = existsSync(join(__dirname, 'tokenizers.win32-arm64-msvc.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.win32-arm64-msvc.node') } else { nativeBinding = require('tokenizers-win32-arm64-msvc') } } catch (e) { loadError = e } break default: throw new Error(`Unsupported architecture on Windows: ${arch}`) } break case 'darwin': localFileExisted = existsSync(join(__dirname, 'tokenizers.darwin-universal.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.darwin-universal.node') } else { nativeBinding = require('tokenizers-darwin-universal') } break } catch {} switch (arch) { case 'x64': localFileExisted = existsSync(join(__dirname, 'tokenizers.darwin-x64.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.darwin-x64.node') } else { nativeBinding = require('tokenizers-darwin-x64') } } catch (e) { loadError = e } break case 'arm64': localFileExisted = existsSync(join(__dirname, 'tokenizers.darwin-arm64.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.darwin-arm64.node') } else { nativeBinding = require('tokenizers-darwin-arm64') } } catch (e) { loadError = e } break default: throw new Error(`Unsupported architecture on macOS: ${arch}`) } break case 'freebsd': if (arch !== 'x64') { throw new Error(`Unsupported architecture on FreeBSD: ${arch}`) } localFileExisted = existsSync(join(__dirname, 'tokenizers.freebsd-x64.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.freebsd-x64.node') } else { nativeBinding = require('tokenizers-freebsd-x64') } } catch (e) { loadError = e } break case 'linux': switch (arch) { case 'x64': if (isMusl()) { localFileExisted = existsSync(join(__dirname, 'tokenizers.linux-x64-musl.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.linux-x64-musl.node') } else { nativeBinding = require('tokenizers-linux-x64-musl') } } catch (e) { loadError = e } } else { localFileExisted = existsSync(join(__dirname, 'tokenizers.linux-x64-gnu.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.linux-x64-gnu.node') } else { nativeBinding = require('tokenizers-linux-x64-gnu') } } catch (e) { loadError = e } } break case 'arm64': if (isMusl()) { localFileExisted = existsSync(join(__dirname, 'tokenizers.linux-arm64-musl.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.linux-arm64-musl.node') } else { nativeBinding = require('tokenizers-linux-arm64-musl') } } catch (e) { loadError = e } } else { localFileExisted = existsSync(join(__dirname, 'tokenizers.linux-arm64-gnu.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.linux-arm64-gnu.node') } else { nativeBinding = require('tokenizers-linux-arm64-gnu') } } catch (e) { loadError = e } } break case 'arm': localFileExisted = existsSync(join(__dirname, 'tokenizers.linux-arm-gnueabihf.node')) try { if (localFileExisted) { nativeBinding = require('./tokenizers.linux-arm-gnueabihf.node') } else { nativeBinding = require('tokenizers-linux-arm-gnueabihf') } } catch (e) { loadError = e } break default: throw new Error(`Unsupported architecture on Linux: ${arch}`) } break default: throw new Error(`Unsupported OS: ${platform}, architecture: ${arch}`) } if (!nativeBinding) { if (loadError) { throw loadError } throw new Error(`Failed to load native binding`) } const { Decoder, bpeDecoder, byteFallbackDecoder, ctcDecoder, fuseDecoder, metaspaceDecoder, replaceDecoder, sequenceDecoder, stripDecoder, wordPieceDecoder, Encoding, TruncationDirection, TruncationStrategy, Model, BPE, WordPiece, WordLevel, Unigram, Normalizer, prependNormalizer, stripAccentsNormalizer, bertNormalizer, nfdNormalizer, nfkdNormalizer, nfcNormalizer, nfkcNormalizer, stripNormalizer, sequenceNormalizer, lowercase, replace, nmt, precompiled, JsSplitDelimiterBehavior, PreTokenizer, byteLevelPreTokenizer, byteLevelAlphabet, whitespacePreTokenizer, whitespaceSplitPreTokenizer, bertPreTokenizer, metaspacePreTokenizer, splitPreTokenizer, punctuationPreTokenizer, sequencePreTokenizer, charDelimiterSplit, digitsPreTokenizer, Processor, bertProcessing, robertaProcessing, byteLevelProcessing, templateProcessing, sequenceProcessing, PaddingDirection, AddedToken, Tokenizer, Trainer, slice, mergeEncodings, } = nativeBinding module.exports.Decoder = Decoder module.exports.bpeDecoder = bpeDecoder module.exports.byteFallbackDecoder = byteFallbackDecoder module.exports.ctcDecoder = ctcDecoder module.exports.fuseDecoder = fuseDecoder module.exports.metaspaceDecoder = metaspaceDecoder module.exports.replaceDecoder = replaceDecoder module.exports.sequenceDecoder = sequenceDecoder module.exports.stripDecoder = stripDecoder module.exports.wordPieceDecoder = wordPieceDecoder module.exports.Encoding = Encoding module.exports.TruncationDirection = TruncationDirection module.exports.TruncationStrategy = TruncationStrategy module.exports.Model = Model module.exports.BPE = BPE module.exports.WordPiece = WordPiece module.exports.WordLevel = WordLevel module.exports.Unigram = Unigram module.exports.Normalizer = Normalizer module.exports.prependNormalizer = prependNormalizer module.exports.stripAccentsNormalizer = stripAccentsNormalizer module.exports.bertNormalizer = bertNormalizer module.exports.nfdNormalizer = nfdNormalizer module.exports.nfkdNormalizer = nfkdNormalizer module.exports.nfcNormalizer = nfcNormalizer module.exports.nfkcNormalizer = nfkcNormalizer module.exports.stripNormalizer = stripNormalizer module.exports.sequenceNormalizer = sequenceNormalizer module.exports.lowercase = lowercase module.exports.replace = replace module.exports.nmt = nmt module.exports.precompiled = precompiled module.exports.JsSplitDelimiterBehavior = JsSplitDelimiterBehavior module.exports.PreTokenizer = PreTokenizer module.exports.byteLevelPreTokenizer = byteLevelPreTokenizer module.exports.byteLevelAlphabet = byteLevelAlphabet module.exports.whitespacePreTokenizer = whitespacePreTokenizer module.exports.whitespaceSplitPreTokenizer = whitespaceSplitPreTokenizer module.exports.bertPreTokenizer = bertPreTokenizer module.exports.metaspacePreTokenizer = metaspacePreTokenizer module.exports.splitPreTokenizer = splitPreTokenizer module.exports.punctuationPreTokenizer = punctuationPreTokenizer module.exports.sequencePreTokenizer = sequencePreTokenizer module.exports.charDelimiterSplit = charDelimiterSplit module.exports.digitsPreTokenizer = digitsPreTokenizer module.exports.Processor = Processor module.exports.bertProcessing = bertProcessing module.exports.robertaProcessing = robertaProcessing module.exports.byteLevelProcessing = byteLevelProcessing module.exports.templateProcessing = templateProcessing module.exports.sequenceProcessing = sequenceProcessing module.exports.PaddingDirection = PaddingDirection module.exports.AddedToken = AddedToken module.exports.Tokenizer = Tokenizer module.exports.Trainer = Trainer module.exports.slice = slice module.exports.mergeEncodings = mergeEncodings

tokenizers/bindings/node/index.js/0

{ "name": "tokenizers-android-arm64", "version": "0.13.4-rc1", "os": [ "android" ], "cpu": [ "arm64" ], "main": "tokenizers.android-arm64.node", "files": [ "tokenizers.android-arm64.node" ], "description": "Tokenizers platform specific bindings", "keywords": [ "napi-rs", "NAPI", "N-API", "Rust", "node-addon", "node-addon-api" ], "license": "MIT", "engines": { "node": ">= 10" }, "publishConfig": { "registry": "https://registry.npmjs.org/", "access": "public" }, "repository": "tokenizers" }

tokenizers/bindings/node/npm/android-arm64/package.json/0

{ "name": "tokenizers-linux-x64-musl", "version": "0.13.4-rc1", "os": [ "linux" ], "cpu": [ "x64" ], "main": "tokenizers.linux-x64-musl.node", "files": [ "tokenizers.linux-x64-musl.node" ], "description": "Tokenizers platform specific bindings", "keywords": [ "napi-rs", "NAPI", "N-API", "Rust", "node-addon", "node-addon-api" ], "license": "MIT", "engines": { "node": ">= 10" }, "publishConfig": { "registry": "https://registry.npmjs.org/", "access": "public" }, "repository": "tokenizers", "libc": [ "musl" ] }

tokenizers/bindings/node/npm/linux-x64-musl/package.json/0