Spaces:
Runtime error
Runtime error
zhangzhe45
commited on
Commit
·
50d1ff1
1
Parent(s):
b1893ac
add
Browse files- app.py +83 -0
- configs/med_large_config.json +21 -0
- mm_commerce.py +146 -0
- models/__init__.py +0 -0
- models/med.py +955 -0
- models/modeling_clip.py +1054 -0
- models/vit.py +305 -0
- requirements.txt +7 -0
- resources/bert-large-chinese/config.json +28 -0
- resources/bert-large-chinese/tokenizer_config.json +1 -0
- resources/bert-large-chinese/vocab.txt +0 -0
- resources/clip_vit_large_patch14/config.json +17 -0
- resources/examples/charger-hw.jpg +3 -0
- resources/examples/charger-ugreen.jpg +3 -0
- resources/examples/charger.jpg +3 -0
- resources/examples/jiandao.jpg +3 -0
- resources/examples/lego-yellow.jpg +3 -0
- starrynight.jpeg +3 -0
- transform/randaugment.py +340 -0
- utils.py +278 -0
app.py
ADDED
@@ -0,0 +1,83 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
from urllib.parse import urlparse
|
3 |
+
|
4 |
+
from PIL import Image
|
5 |
+
import requests
|
6 |
+
import torch
|
7 |
+
from timm.models.hub import download_cached_file
|
8 |
+
from torchvision import transforms
|
9 |
+
from torchvision.transforms.functional import InterpolationMode
|
10 |
+
import gradio as gr
|
11 |
+
from mm_commerce import BLIP_Decoder
|
12 |
+
|
13 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
14 |
+
|
15 |
+
|
16 |
+
def is_url(url_or_filename):
|
17 |
+
parsed = urlparse(url_or_filename)
|
18 |
+
return parsed.scheme in ("http", "https")
|
19 |
+
|
20 |
+
|
21 |
+
def load_checkpoint(url_or_filename):
|
22 |
+
if is_url(url_or_filename):
|
23 |
+
cached_file = download_cached_file(url_or_filename, check_hash=False, progress=True)
|
24 |
+
checkpoint = torch.load(cached_file, map_location='cpu')
|
25 |
+
elif os.path.isfile(url_or_filename):
|
26 |
+
checkpoint = torch.load(url_or_filename, map_location='cpu')
|
27 |
+
else:
|
28 |
+
raise RuntimeError('checkpoint url or path is invalid')
|
29 |
+
return checkpoint
|
30 |
+
|
31 |
+
|
32 |
+
image_size = 224
|
33 |
+
transform = transforms.Compose([
|
34 |
+
transforms.Resize((image_size,image_size),interpolation=InterpolationMode.BICUBIC),
|
35 |
+
transforms.ToTensor(),
|
36 |
+
transforms.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711))
|
37 |
+
])
|
38 |
+
|
39 |
+
model = BLIP_Decoder(med_config='configs/med_large_config.json', vit='large_v2', prompt='[DEC]')
|
40 |
+
|
41 |
+
ckpt = 'https://huggingface.co/zhezh/mm_commerce_zhcn/resolve/main/model.pth'
|
42 |
+
sd = load_checkpoint(ckpt)
|
43 |
+
model.load_state_dict(sd, strict=True)
|
44 |
+
|
45 |
+
model.eval()
|
46 |
+
model = model.to('cuda')
|
47 |
+
|
48 |
+
|
49 |
+
def inference(raw_image, strategy):
|
50 |
+
image = transform(raw_image).unsqueeze(0).to(device)
|
51 |
+
with torch.no_grad():
|
52 |
+
if strategy == "Beam search":
|
53 |
+
caption = model.generate(image, sample=False, num_beams=10, max_length=100, min_length=10)
|
54 |
+
else:
|
55 |
+
caption = model.generate(image, sample=True, top_p=0.9, max_length=100, min_length=10)
|
56 |
+
return '商品描述: ' + '"' + ''.join(caption[0][6:-5].split()) + '"'
|
57 |
+
|
58 |
+
|
59 |
+
inputs = [
|
60 |
+
gr.inputs.Image(type='pil'),
|
61 |
+
gr.inputs.Radio(choices=['Beam search','Nucleus sampling'], type="value", default="Beam search", label="文本生成策略")
|
62 |
+
]
|
63 |
+
outputs = gr.outputs.Textbox(label="生成的标题(Output)")
|
64 |
+
|
65 |
+
title = "MM Commerce ZhCN (中文商品描述生成)"
|
66 |
+
|
67 |
+
description = "中文商品描述生成 -- By Zhe Zhang"
|
68 |
+
|
69 |
+
demo = gr.Interface(
|
70 |
+
inference, inputs, outputs, title=title, description=description,
|
71 |
+
# article=article,
|
72 |
+
examples=[
|
73 |
+
['starrynight.jpeg', "Nucleus sampling"],
|
74 |
+
['resources/examples/zhuobu.jpg', "Beam search"],
|
75 |
+
['resources/examples/jiandao.jpg', "Beam search"],
|
76 |
+
['resources/examples/lego-yellow.jpg', "Beam search"],
|
77 |
+
['resources/examples/charger.jpg', "Beam search"],
|
78 |
+
['resources/examples/charger-ugreen.jpg', "Beam search"],
|
79 |
+
['resources/examples/charger-hw.jpg', "Beam search"],
|
80 |
+
],
|
81 |
+
)
|
82 |
+
# demo.launch(enable_queue=True, share=True, server_name='0.0.0.0', server_port=8080,)
|
83 |
+
demo.launch(enable_queue=True)
|
configs/med_large_config.json
ADDED
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"BertModel"
|
4 |
+
],
|
5 |
+
"attention_probs_dropout_prob": 0.1,
|
6 |
+
"hidden_act": "gelu",
|
7 |
+
"hidden_dropout_prob": 0.1,
|
8 |
+
"hidden_size": 1024,
|
9 |
+
"initializer_range": 0.02,
|
10 |
+
"intermediate_size": 4096,
|
11 |
+
"layer_norm_eps": 1e-12,
|
12 |
+
"max_position_embeddings": 512,
|
13 |
+
"model_type": "bert",
|
14 |
+
"num_attention_heads": 16,
|
15 |
+
"num_hidden_layers": 24,
|
16 |
+
"pad_token_id": 0,
|
17 |
+
"type_vocab_size": 2,
|
18 |
+
"vocab_size": 21130,
|
19 |
+
"encoder_width": 768,
|
20 |
+
"add_cross_attention": true
|
21 |
+
}
|
mm_commerce.py
ADDED
@@ -0,0 +1,146 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import warnings
|
3 |
+
|
4 |
+
from PIL import Image
|
5 |
+
from torchvision import transforms
|
6 |
+
from torchvision.transforms import InterpolationMode
|
7 |
+
|
8 |
+
warnings.filterwarnings("ignore")
|
9 |
+
|
10 |
+
from models.vit import VisionTransformer, interpolate_pos_embed
|
11 |
+
from models.med import BertConfig, BertModel, BertLMHeadModel
|
12 |
+
from transformers import BertTokenizer, CLIPConfig
|
13 |
+
from models.modeling_clip import CLIPModel, CLIPVisionModel, CLIPVisionConfig
|
14 |
+
|
15 |
+
import torch
|
16 |
+
from torch import nn
|
17 |
+
import torch.nn.functional as F
|
18 |
+
|
19 |
+
|
20 |
+
class BLIP_Decoder(nn.Module):
|
21 |
+
def __init__(self,
|
22 |
+
med_config='configs/med_config.json',
|
23 |
+
image_size=384,
|
24 |
+
vit='base',
|
25 |
+
vit_grad_ckpt=False,
|
26 |
+
vit_ckpt_layer=0,
|
27 |
+
prompt='[DEC]',
|
28 |
+
):
|
29 |
+
super().__init__()
|
30 |
+
self.visual_encoder, vision_width = create_vit(vit, image_size, vit_grad_ckpt, vit_ckpt_layer, 0)
|
31 |
+
self.tokenizer = init_tokenizer()
|
32 |
+
med_config = BertConfig.from_json_file(med_config)
|
33 |
+
med_config.encoder_width = vision_width
|
34 |
+
self.text_decoder = BertLMHeadModel(config=med_config)
|
35 |
+
|
36 |
+
self.prompt = prompt
|
37 |
+
self.prompt_length = len(self.tokenizer(self.prompt).input_ids) - 1
|
38 |
+
|
39 |
+
def forward(self, image, caption):
|
40 |
+
|
41 |
+
image_embeds = self.visual_encoder(image)
|
42 |
+
image_atts = torch.ones(image_embeds.size()[:-1], dtype=torch.long).to(image.device)
|
43 |
+
|
44 |
+
text = self.tokenizer(caption, padding='longest', truncation=True, max_length=40, return_tensors="pt").to(image.device)
|
45 |
+
|
46 |
+
text.input_ids[:, 0] = self.tokenizer.bos_token_id
|
47 |
+
|
48 |
+
decoder_targets = text.input_ids.masked_fill(text.input_ids == self.tokenizer.pad_token_id, -100)
|
49 |
+
decoder_targets[:, :self.prompt_length] = -100
|
50 |
+
|
51 |
+
decoder_output = self.text_decoder(text.input_ids,
|
52 |
+
attention_mask=text.attention_mask,
|
53 |
+
encoder_hidden_states=image_embeds,
|
54 |
+
encoder_attention_mask=image_atts,
|
55 |
+
labels=decoder_targets,
|
56 |
+
return_dict=True,
|
57 |
+
)
|
58 |
+
loss_lm = decoder_output.loss
|
59 |
+
|
60 |
+
return loss_lm
|
61 |
+
|
62 |
+
def generate(self, image, sample=False, num_beams=3, max_length=30, min_length=10, top_p=0.9, repetition_penalty=1.0):
|
63 |
+
image_embeds = self.visual_encoder(image)
|
64 |
+
|
65 |
+
if not sample:
|
66 |
+
image_embeds = image_embeds.repeat_interleave(num_beams, dim=0)
|
67 |
+
|
68 |
+
image_atts = torch.ones(image_embeds.size()[:-1], dtype=torch.long).to(image.device)
|
69 |
+
model_kwargs = {"encoder_hidden_states": image_embeds, "encoder_attention_mask": image_atts}
|
70 |
+
|
71 |
+
prompt = [self.prompt] * image.size(0)
|
72 |
+
input_ids = self.tokenizer(prompt, return_tensors="pt").input_ids.to(image.device)
|
73 |
+
input_ids[:, 0] = self.tokenizer.bos_token_id
|
74 |
+
input_ids = input_ids[:, :-1]
|
75 |
+
|
76 |
+
if sample:
|
77 |
+
# nucleus sampling
|
78 |
+
outputs = self.text_decoder.generate(input_ids=input_ids,
|
79 |
+
max_length=max_length,
|
80 |
+
min_length=min_length,
|
81 |
+
do_sample=True,
|
82 |
+
top_p=top_p,
|
83 |
+
num_return_sequences=1,
|
84 |
+
eos_token_id=self.tokenizer.sep_token_id,
|
85 |
+
pad_token_id=self.tokenizer.pad_token_id,
|
86 |
+
repetition_penalty=1.1,
|
87 |
+
**model_kwargs)
|
88 |
+
else:
|
89 |
+
# beam search
|
90 |
+
outputs = self.text_decoder.generate(input_ids=input_ids,
|
91 |
+
max_length=max_length,
|
92 |
+
min_length=min_length,
|
93 |
+
num_beams=num_beams,
|
94 |
+
eos_token_id=self.tokenizer.sep_token_id,
|
95 |
+
pad_token_id=self.tokenizer.pad_token_id,
|
96 |
+
repetition_penalty=repetition_penalty,
|
97 |
+
**model_kwargs)
|
98 |
+
|
99 |
+
captions = []
|
100 |
+
for output in outputs:
|
101 |
+
caption = self.tokenizer.decode(output, skip_special_tokens=False)
|
102 |
+
captions.append(caption[len(self.prompt):])
|
103 |
+
return captions
|
104 |
+
|
105 |
+
|
106 |
+
def init_tokenizer():
|
107 |
+
tokenizer = BertTokenizer.from_pretrained('resources/bert-large-chinese', do_lower_case=True)
|
108 |
+
tokenizer.add_special_tokens({'bos_token': '[DEC]'})
|
109 |
+
tokenizer.add_special_tokens({'additional_special_tokens': ['[ENC]']})
|
110 |
+
tokenizer.enc_token_id = tokenizer.additional_special_tokens_ids[0]
|
111 |
+
return tokenizer
|
112 |
+
|
113 |
+
|
114 |
+
def create_vit(vit, image_size, use_grad_checkpointing=False, ckpt_layer=0, drop_path_rate=0):
|
115 |
+
assert vit in ['base', 'large', 'large_v2'], "vit parameter must be base or large"
|
116 |
+
if vit == 'base':
|
117 |
+
vision_width = 768
|
118 |
+
visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=12,
|
119 |
+
num_heads=12, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer,
|
120 |
+
drop_path_rate=0 or drop_path_rate
|
121 |
+
)
|
122 |
+
elif vit == 'large':
|
123 |
+
vision_width = 1024
|
124 |
+
visual_encoder = VisionTransformer(img_size=image_size, patch_size=16, embed_dim=vision_width, depth=24,
|
125 |
+
num_heads=16, use_grad_checkpointing=use_grad_checkpointing, ckpt_layer=ckpt_layer,
|
126 |
+
drop_path_rate=0.1 or drop_path_rate
|
127 |
+
)
|
128 |
+
elif vit == 'large_v2':
|
129 |
+
vision_width = 1024
|
130 |
+
clip_config = CLIPConfig.from_pretrained('resources/clip_vit_large_patch14')
|
131 |
+
visual_encoder = CLIPVisionModel(clip_config)
|
132 |
+
return visual_encoder, vision_width
|
133 |
+
|
134 |
+
|
135 |
+
def load_image(image, image_size, device):
|
136 |
+
raw_image = Image.open(str(image)).convert('RGB')
|
137 |
+
|
138 |
+
w, h = raw_image.size
|
139 |
+
|
140 |
+
transform = transforms.Compose([
|
141 |
+
transforms.Resize((image_size, image_size), interpolation=InterpolationMode.BICUBIC),
|
142 |
+
transforms.ToTensor(),
|
143 |
+
transforms.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711))
|
144 |
+
])
|
145 |
+
image = transform(raw_image).unsqueeze(0).to(device)
|
146 |
+
return image
|
models/__init__.py
ADDED
File without changes
|
models/med.py
ADDED
@@ -0,0 +1,955 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
'''
|
2 |
+
* Copyright (c) 2022, salesforce.com, inc.
|
3 |
+
* All rights reserved.
|
4 |
+
* SPDX-License-Identifier: BSD-3-Clause
|
5 |
+
* For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
|
6 |
+
* By Junnan Li
|
7 |
+
* Based on huggingface code base
|
8 |
+
* https://github.com/huggingface/transformers/blob/v4.15.0/src/transformers/models/bert
|
9 |
+
'''
|
10 |
+
|
11 |
+
import math
|
12 |
+
import os
|
13 |
+
import warnings
|
14 |
+
from dataclasses import dataclass
|
15 |
+
from typing import Optional, Tuple
|
16 |
+
|
17 |
+
import torch
|
18 |
+
from torch import Tensor, device, dtype, nn
|
19 |
+
import torch.utils.checkpoint
|
20 |
+
from torch import nn
|
21 |
+
from torch.nn import CrossEntropyLoss
|
22 |
+
import torch.nn.functional as F
|
23 |
+
|
24 |
+
from transformers.activations import ACT2FN
|
25 |
+
from transformers.file_utils import (
|
26 |
+
ModelOutput,
|
27 |
+
)
|
28 |
+
from transformers.modeling_outputs import (
|
29 |
+
BaseModelOutputWithPastAndCrossAttentions,
|
30 |
+
BaseModelOutputWithPoolingAndCrossAttentions,
|
31 |
+
CausalLMOutputWithCrossAttentions,
|
32 |
+
MaskedLMOutput,
|
33 |
+
MultipleChoiceModelOutput,
|
34 |
+
NextSentencePredictorOutput,
|
35 |
+
QuestionAnsweringModelOutput,
|
36 |
+
SequenceClassifierOutput,
|
37 |
+
TokenClassifierOutput,
|
38 |
+
)
|
39 |
+
from transformers.modeling_utils import (
|
40 |
+
PreTrainedModel,
|
41 |
+
apply_chunking_to_forward,
|
42 |
+
find_pruneable_heads_and_indices,
|
43 |
+
prune_linear_layer,
|
44 |
+
)
|
45 |
+
from transformers.utils import logging
|
46 |
+
from transformers.models.bert.configuration_bert import BertConfig
|
47 |
+
|
48 |
+
|
49 |
+
logger = logging.get_logger(__name__)
|
50 |
+
|
51 |
+
|
52 |
+
class BertEmbeddings(nn.Module):
|
53 |
+
"""Construct the embeddings from word and position embeddings."""
|
54 |
+
|
55 |
+
def __init__(self, config):
|
56 |
+
super().__init__()
|
57 |
+
self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
|
58 |
+
self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
|
59 |
+
|
60 |
+
# self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
|
61 |
+
# any TensorFlow checkpoint file
|
62 |
+
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
63 |
+
self.dropout = nn.Dropout(config.hidden_dropout_prob)
|
64 |
+
|
65 |
+
# position_ids (1, len position emb) is contiguous in memory and exported when serialized
|
66 |
+
self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))
|
67 |
+
self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
|
68 |
+
|
69 |
+
self.config = config
|
70 |
+
|
71 |
+
def forward(
|
72 |
+
self, input_ids=None, position_ids=None, inputs_embeds=None, past_key_values_length=0
|
73 |
+
):
|
74 |
+
if input_ids is not None:
|
75 |
+
input_shape = input_ids.size()
|
76 |
+
else:
|
77 |
+
input_shape = inputs_embeds.size()[:-1]
|
78 |
+
|
79 |
+
seq_length = input_shape[1]
|
80 |
+
|
81 |
+
if position_ids is None:
|
82 |
+
position_ids = self.position_ids[:, past_key_values_length : seq_length + past_key_values_length]
|
83 |
+
|
84 |
+
if inputs_embeds is None:
|
85 |
+
inputs_embeds = self.word_embeddings(input_ids)
|
86 |
+
|
87 |
+
embeddings = inputs_embeds
|
88 |
+
|
89 |
+
if self.position_embedding_type == "absolute":
|
90 |
+
position_embeddings = self.position_embeddings(position_ids)
|
91 |
+
embeddings += position_embeddings
|
92 |
+
embeddings = self.LayerNorm(embeddings)
|
93 |
+
embeddings = self.dropout(embeddings)
|
94 |
+
return embeddings
|
95 |
+
|
96 |
+
|
97 |
+
class BertSelfAttention(nn.Module):
|
98 |
+
def __init__(self, config, is_cross_attention):
|
99 |
+
super().__init__()
|
100 |
+
self.config = config
|
101 |
+
if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
|
102 |
+
raise ValueError(
|
103 |
+
"The hidden size (%d) is not a multiple of the number of attention "
|
104 |
+
"heads (%d)" % (config.hidden_size, config.num_attention_heads)
|
105 |
+
)
|
106 |
+
|
107 |
+
self.num_attention_heads = config.num_attention_heads
|
108 |
+
self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
|
109 |
+
self.all_head_size = self.num_attention_heads * self.attention_head_size
|
110 |
+
|
111 |
+
self.query = nn.Linear(config.hidden_size, self.all_head_size)
|
112 |
+
if is_cross_attention:
|
113 |
+
self.key = nn.Linear(config.encoder_width, self.all_head_size)
|
114 |
+
self.value = nn.Linear(config.encoder_width, self.all_head_size)
|
115 |
+
else:
|
116 |
+
self.key = nn.Linear(config.hidden_size, self.all_head_size)
|
117 |
+
self.value = nn.Linear(config.hidden_size, self.all_head_size)
|
118 |
+
|
119 |
+
self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
|
120 |
+
self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
|
121 |
+
if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
|
122 |
+
self.max_position_embeddings = config.max_position_embeddings
|
123 |
+
self.distance_embedding = nn.Embedding(2 * config.max_position_embeddings - 1, self.attention_head_size)
|
124 |
+
self.save_attention = False
|
125 |
+
|
126 |
+
def save_attn_gradients(self, attn_gradients):
|
127 |
+
self.attn_gradients = attn_gradients
|
128 |
+
|
129 |
+
def get_attn_gradients(self):
|
130 |
+
return self.attn_gradients
|
131 |
+
|
132 |
+
def save_attention_map(self, attention_map):
|
133 |
+
self.attention_map = attention_map
|
134 |
+
|
135 |
+
def get_attention_map(self):
|
136 |
+
return self.attention_map
|
137 |
+
|
138 |
+
def transpose_for_scores(self, x):
|
139 |
+
new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
|
140 |
+
x = x.view(*new_x_shape)
|
141 |
+
return x.permute(0, 2, 1, 3)
|
142 |
+
|
143 |
+
def forward(
|
144 |
+
self,
|
145 |
+
hidden_states,
|
146 |
+
attention_mask=None,
|
147 |
+
head_mask=None,
|
148 |
+
encoder_hidden_states=None,
|
149 |
+
encoder_attention_mask=None,
|
150 |
+
past_key_value=None,
|
151 |
+
output_attentions=False,
|
152 |
+
):
|
153 |
+
mixed_query_layer = self.query(hidden_states)
|
154 |
+
|
155 |
+
# If this is instantiated as a cross-attention module, the keys
|
156 |
+
# and values come from an encoder; the attention mask needs to be
|
157 |
+
# such that the encoder's padding tokens are not attended to.
|
158 |
+
is_cross_attention = encoder_hidden_states is not None
|
159 |
+
|
160 |
+
if is_cross_attention:
|
161 |
+
key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
|
162 |
+
value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
|
163 |
+
attention_mask = encoder_attention_mask
|
164 |
+
elif past_key_value is not None:
|
165 |
+
key_layer = self.transpose_for_scores(self.key(hidden_states))
|
166 |
+
value_layer = self.transpose_for_scores(self.value(hidden_states))
|
167 |
+
key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
|
168 |
+
value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
|
169 |
+
else:
|
170 |
+
key_layer = self.transpose_for_scores(self.key(hidden_states))
|
171 |
+
value_layer = self.transpose_for_scores(self.value(hidden_states))
|
172 |
+
|
173 |
+
query_layer = self.transpose_for_scores(mixed_query_layer)
|
174 |
+
|
175 |
+
past_key_value = (key_layer, value_layer)
|
176 |
+
|
177 |
+
# Take the dot product between "query" and "key" to get the raw attention scores.
|
178 |
+
attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
|
179 |
+
|
180 |
+
if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
|
181 |
+
seq_length = hidden_states.size()[1]
|
182 |
+
position_ids_l = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
|
183 |
+
position_ids_r = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(1, -1)
|
184 |
+
distance = position_ids_l - position_ids_r
|
185 |
+
positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
|
186 |
+
positional_embedding = positional_embedding.to(dtype=query_layer.dtype) # fp16 compatibility
|
187 |
+
|
188 |
+
if self.position_embedding_type == "relative_key":
|
189 |
+
relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
|
190 |
+
attention_scores = attention_scores + relative_position_scores
|
191 |
+
elif self.position_embedding_type == "relative_key_query":
|
192 |
+
relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
|
193 |
+
relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
|
194 |
+
attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key
|
195 |
+
|
196 |
+
attention_scores = attention_scores / math.sqrt(self.attention_head_size)
|
197 |
+
if attention_mask is not None:
|
198 |
+
# Apply the attention mask is (precomputed for all layers in BertModel forward() function)
|
199 |
+
attention_scores = attention_scores + attention_mask
|
200 |
+
|
201 |
+
# Normalize the attention scores to probabilities.
|
202 |
+
attention_probs = nn.Softmax(dim=-1)(attention_scores)
|
203 |
+
|
204 |
+
if is_cross_attention and self.save_attention:
|
205 |
+
self.save_attention_map(attention_probs)
|
206 |
+
attention_probs.register_hook(self.save_attn_gradients)
|
207 |
+
|
208 |
+
# This is actually dropping out entire tokens to attend to, which might
|
209 |
+
# seem a bit unusual, but is taken from the original Transformer paper.
|
210 |
+
attention_probs_dropped = self.dropout(attention_probs)
|
211 |
+
|
212 |
+
# Mask heads if we want to
|
213 |
+
if head_mask is not None:
|
214 |
+
attention_probs_dropped = attention_probs_dropped * head_mask
|
215 |
+
|
216 |
+
context_layer = torch.matmul(attention_probs_dropped, value_layer)
|
217 |
+
|
218 |
+
context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
|
219 |
+
new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
|
220 |
+
context_layer = context_layer.view(*new_context_layer_shape)
|
221 |
+
|
222 |
+
outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)
|
223 |
+
|
224 |
+
outputs = outputs + (past_key_value,)
|
225 |
+
return outputs
|
226 |
+
|
227 |
+
|
228 |
+
class BertSelfOutput(nn.Module):
|
229 |
+
def __init__(self, config):
|
230 |
+
super().__init__()
|
231 |
+
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
|
232 |
+
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
233 |
+
self.dropout = nn.Dropout(config.hidden_dropout_prob)
|
234 |
+
|
235 |
+
def forward(self, hidden_states, input_tensor):
|
236 |
+
hidden_states = self.dense(hidden_states)
|
237 |
+
hidden_states = self.dropout(hidden_states)
|
238 |
+
hidden_states = self.LayerNorm(hidden_states + input_tensor)
|
239 |
+
return hidden_states
|
240 |
+
|
241 |
+
|
242 |
+
class BertAttention(nn.Module):
|
243 |
+
def __init__(self, config, is_cross_attention=False):
|
244 |
+
super().__init__()
|
245 |
+
self.self = BertSelfAttention(config, is_cross_attention)
|
246 |
+
self.output = BertSelfOutput(config)
|
247 |
+
self.pruned_heads = set()
|
248 |
+
|
249 |
+
def prune_heads(self, heads):
|
250 |
+
if len(heads) == 0:
|
251 |
+
return
|
252 |
+
heads, index = find_pruneable_heads_and_indices(
|
253 |
+
heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
|
254 |
+
)
|
255 |
+
|
256 |
+
# Prune linear layers
|
257 |
+
self.self.query = prune_linear_layer(self.self.query, index)
|
258 |
+
self.self.key = prune_linear_layer(self.self.key, index)
|
259 |
+
self.self.value = prune_linear_layer(self.self.value, index)
|
260 |
+
self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)
|
261 |
+
|
262 |
+
# Update hyper params and store pruned heads
|
263 |
+
self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
|
264 |
+
self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads
|
265 |
+
self.pruned_heads = self.pruned_heads.union(heads)
|
266 |
+
|
267 |
+
def forward(
|
268 |
+
self,
|
269 |
+
hidden_states,
|
270 |
+
attention_mask=None,
|
271 |
+
head_mask=None,
|
272 |
+
encoder_hidden_states=None,
|
273 |
+
encoder_attention_mask=None,
|
274 |
+
past_key_value=None,
|
275 |
+
output_attentions=False,
|
276 |
+
):
|
277 |
+
self_outputs = self.self(
|
278 |
+
hidden_states,
|
279 |
+
attention_mask,
|
280 |
+
head_mask,
|
281 |
+
encoder_hidden_states,
|
282 |
+
encoder_attention_mask,
|
283 |
+
past_key_value,
|
284 |
+
output_attentions,
|
285 |
+
)
|
286 |
+
attention_output = self.output(self_outputs[0], hidden_states)
|
287 |
+
outputs = (attention_output,) + self_outputs[1:] # add attentions if we output them
|
288 |
+
return outputs
|
289 |
+
|
290 |
+
|
291 |
+
class BertIntermediate(nn.Module):
|
292 |
+
def __init__(self, config):
|
293 |
+
super().__init__()
|
294 |
+
self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
|
295 |
+
if isinstance(config.hidden_act, str):
|
296 |
+
self.intermediate_act_fn = ACT2FN[config.hidden_act]
|
297 |
+
else:
|
298 |
+
self.intermediate_act_fn = config.hidden_act
|
299 |
+
|
300 |
+
def forward(self, hidden_states):
|
301 |
+
hidden_states = self.dense(hidden_states)
|
302 |
+
hidden_states = self.intermediate_act_fn(hidden_states)
|
303 |
+
return hidden_states
|
304 |
+
|
305 |
+
|
306 |
+
class BertOutput(nn.Module):
|
307 |
+
def __init__(self, config):
|
308 |
+
super().__init__()
|
309 |
+
self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
|
310 |
+
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
311 |
+
self.dropout = nn.Dropout(config.hidden_dropout_prob)
|
312 |
+
|
313 |
+
def forward(self, hidden_states, input_tensor):
|
314 |
+
hidden_states = self.dense(hidden_states)
|
315 |
+
hidden_states = self.dropout(hidden_states)
|
316 |
+
hidden_states = self.LayerNorm(hidden_states + input_tensor)
|
317 |
+
return hidden_states
|
318 |
+
|
319 |
+
|
320 |
+
class BertLayer(nn.Module):
|
321 |
+
def __init__(self, config, layer_num):
|
322 |
+
super().__init__()
|
323 |
+
self.config = config
|
324 |
+
self.chunk_size_feed_forward = config.chunk_size_feed_forward
|
325 |
+
self.seq_len_dim = 1
|
326 |
+
self.attention = BertAttention(config)
|
327 |
+
self.layer_num = layer_num
|
328 |
+
if self.config.add_cross_attention:
|
329 |
+
self.crossattention = BertAttention(config, is_cross_attention=self.config.add_cross_attention)
|
330 |
+
self.intermediate = BertIntermediate(config)
|
331 |
+
self.output = BertOutput(config)
|
332 |
+
|
333 |
+
def forward(
|
334 |
+
self,
|
335 |
+
hidden_states,
|
336 |
+
attention_mask=None,
|
337 |
+
head_mask=None,
|
338 |
+
encoder_hidden_states=None,
|
339 |
+
encoder_attention_mask=None,
|
340 |
+
past_key_value=None,
|
341 |
+
output_attentions=False,
|
342 |
+
mode=None,
|
343 |
+
):
|
344 |
+
# decoder uni-directional self-attention cached key/values tuple is at positions 1,2
|
345 |
+
self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
|
346 |
+
self_attention_outputs = self.attention(
|
347 |
+
hidden_states,
|
348 |
+
attention_mask,
|
349 |
+
head_mask,
|
350 |
+
output_attentions=output_attentions,
|
351 |
+
past_key_value=self_attn_past_key_value,
|
352 |
+
)
|
353 |
+
attention_output = self_attention_outputs[0]
|
354 |
+
|
355 |
+
outputs = self_attention_outputs[1:-1]
|
356 |
+
present_key_value = self_attention_outputs[-1]
|
357 |
+
|
358 |
+
if mode=='multimodal':
|
359 |
+
assert encoder_hidden_states is not None, "encoder_hidden_states must be given for cross-attention layers"
|
360 |
+
|
361 |
+
cross_attention_outputs = self.crossattention(
|
362 |
+
attention_output,
|
363 |
+
attention_mask,
|
364 |
+
head_mask,
|
365 |
+
encoder_hidden_states,
|
366 |
+
encoder_attention_mask,
|
367 |
+
output_attentions=output_attentions,
|
368 |
+
)
|
369 |
+
attention_output = cross_attention_outputs[0]
|
370 |
+
outputs = outputs + cross_attention_outputs[1:-1] # add cross attentions if we output attention weights
|
371 |
+
layer_output = apply_chunking_to_forward(
|
372 |
+
self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
|
373 |
+
)
|
374 |
+
outputs = (layer_output,) + outputs
|
375 |
+
|
376 |
+
outputs = outputs + (present_key_value,)
|
377 |
+
|
378 |
+
return outputs
|
379 |
+
|
380 |
+
def feed_forward_chunk(self, attention_output):
|
381 |
+
intermediate_output = self.intermediate(attention_output)
|
382 |
+
layer_output = self.output(intermediate_output, attention_output)
|
383 |
+
return layer_output
|
384 |
+
|
385 |
+
|
386 |
+
class BertEncoder(nn.Module):
|
387 |
+
def __init__(self, config):
|
388 |
+
super().__init__()
|
389 |
+
self.config = config
|
390 |
+
self.layer = nn.ModuleList([BertLayer(config,i) for i in range(config.num_hidden_layers)])
|
391 |
+
self.gradient_checkpointing = False
|
392 |
+
|
393 |
+
def forward(
|
394 |
+
self,
|
395 |
+
hidden_states,
|
396 |
+
attention_mask=None,
|
397 |
+
head_mask=None,
|
398 |
+
encoder_hidden_states=None,
|
399 |
+
encoder_attention_mask=None,
|
400 |
+
past_key_values=None,
|
401 |
+
use_cache=None,
|
402 |
+
output_attentions=False,
|
403 |
+
output_hidden_states=False,
|
404 |
+
return_dict=True,
|
405 |
+
mode='multimodal',
|
406 |
+
):
|
407 |
+
all_hidden_states = () if output_hidden_states else None
|
408 |
+
all_self_attentions = () if output_attentions else None
|
409 |
+
all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
|
410 |
+
|
411 |
+
next_decoder_cache = () if use_cache else None
|
412 |
+
|
413 |
+
for i in range(self.config.num_hidden_layers):
|
414 |
+
layer_module = self.layer[i]
|
415 |
+
if output_hidden_states:
|
416 |
+
all_hidden_states = all_hidden_states + (hidden_states,)
|
417 |
+
|
418 |
+
layer_head_mask = head_mask[i] if head_mask is not None else None
|
419 |
+
past_key_value = past_key_values[i] if past_key_values is not None else None
|
420 |
+
|
421 |
+
if self.gradient_checkpointing and self.training:
|
422 |
+
|
423 |
+
if use_cache:
|
424 |
+
logger.warn(
|
425 |
+
"`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
|
426 |
+
)
|
427 |
+
use_cache = False
|
428 |
+
|
429 |
+
def create_custom_forward(module):
|
430 |
+
def custom_forward(*inputs):
|
431 |
+
return module(*inputs, past_key_value, output_attentions)
|
432 |
+
|
433 |
+
return custom_forward
|
434 |
+
|
435 |
+
layer_outputs = torch.utils.checkpoint.checkpoint(
|
436 |
+
create_custom_forward(layer_module),
|
437 |
+
hidden_states,
|
438 |
+
attention_mask,
|
439 |
+
layer_head_mask,
|
440 |
+
encoder_hidden_states,
|
441 |
+
encoder_attention_mask,
|
442 |
+
mode=mode,
|
443 |
+
)
|
444 |
+
else:
|
445 |
+
layer_outputs = layer_module(
|
446 |
+
hidden_states,
|
447 |
+
attention_mask,
|
448 |
+
layer_head_mask,
|
449 |
+
encoder_hidden_states,
|
450 |
+
encoder_attention_mask,
|
451 |
+
past_key_value,
|
452 |
+
output_attentions,
|
453 |
+
mode=mode,
|
454 |
+
)
|
455 |
+
|
456 |
+
hidden_states = layer_outputs[0]
|
457 |
+
if use_cache:
|
458 |
+
next_decoder_cache += (layer_outputs[-1],)
|
459 |
+
if output_attentions:
|
460 |
+
all_self_attentions = all_self_attentions + (layer_outputs[1],)
|
461 |
+
|
462 |
+
if output_hidden_states:
|
463 |
+
all_hidden_states = all_hidden_states + (hidden_states,)
|
464 |
+
|
465 |
+
if not return_dict:
|
466 |
+
return tuple(
|
467 |
+
v
|
468 |
+
for v in [
|
469 |
+
hidden_states,
|
470 |
+
next_decoder_cache,
|
471 |
+
all_hidden_states,
|
472 |
+
all_self_attentions,
|
473 |
+
all_cross_attentions,
|
474 |
+
]
|
475 |
+
if v is not None
|
476 |
+
)
|
477 |
+
return BaseModelOutputWithPastAndCrossAttentions(
|
478 |
+
last_hidden_state=hidden_states,
|
479 |
+
past_key_values=next_decoder_cache,
|
480 |
+
hidden_states=all_hidden_states,
|
481 |
+
attentions=all_self_attentions,
|
482 |
+
cross_attentions=all_cross_attentions,
|
483 |
+
)
|
484 |
+
|
485 |
+
|
486 |
+
class BertPooler(nn.Module):
|
487 |
+
def __init__(self, config):
|
488 |
+
super().__init__()
|
489 |
+
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
|
490 |
+
self.activation = nn.Tanh()
|
491 |
+
|
492 |
+
def forward(self, hidden_states):
|
493 |
+
# We "pool" the model by simply taking the hidden state corresponding
|
494 |
+
# to the first token.
|
495 |
+
first_token_tensor = hidden_states[:, 0]
|
496 |
+
pooled_output = self.dense(first_token_tensor)
|
497 |
+
pooled_output = self.activation(pooled_output)
|
498 |
+
return pooled_output
|
499 |
+
|
500 |
+
|
501 |
+
class BertPredictionHeadTransform(nn.Module):
|
502 |
+
def __init__(self, config):
|
503 |
+
super().__init__()
|
504 |
+
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
|
505 |
+
if isinstance(config.hidden_act, str):
|
506 |
+
self.transform_act_fn = ACT2FN[config.hidden_act]
|
507 |
+
else:
|
508 |
+
self.transform_act_fn = config.hidden_act
|
509 |
+
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
510 |
+
|
511 |
+
def forward(self, hidden_states):
|
512 |
+
hidden_states = self.dense(hidden_states)
|
513 |
+
hidden_states = self.transform_act_fn(hidden_states)
|
514 |
+
hidden_states = self.LayerNorm(hidden_states)
|
515 |
+
return hidden_states
|
516 |
+
|
517 |
+
|
518 |
+
class BertLMPredictionHead(nn.Module):
|
519 |
+
def __init__(self, config):
|
520 |
+
super().__init__()
|
521 |
+
self.transform = BertPredictionHeadTransform(config)
|
522 |
+
|
523 |
+
# The output weights are the same as the input embeddings, but there is
|
524 |
+
# an output-only bias for each token.
|
525 |
+
self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
|
526 |
+
|
527 |
+
self.bias = nn.Parameter(torch.zeros(config.vocab_size))
|
528 |
+
|
529 |
+
# Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
|
530 |
+
self.decoder.bias = self.bias
|
531 |
+
|
532 |
+
def forward(self, hidden_states):
|
533 |
+
hidden_states = self.transform(hidden_states)
|
534 |
+
hidden_states = self.decoder(hidden_states)
|
535 |
+
return hidden_states
|
536 |
+
|
537 |
+
|
538 |
+
class BertOnlyMLMHead(nn.Module):
|
539 |
+
def __init__(self, config):
|
540 |
+
super().__init__()
|
541 |
+
self.predictions = BertLMPredictionHead(config)
|
542 |
+
|
543 |
+
def forward(self, sequence_output):
|
544 |
+
prediction_scores = self.predictions(sequence_output)
|
545 |
+
return prediction_scores
|
546 |
+
|
547 |
+
|
548 |
+
class BertPreTrainedModel(PreTrainedModel):
|
549 |
+
"""
|
550 |
+
An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
|
551 |
+
models.
|
552 |
+
"""
|
553 |
+
|
554 |
+
config_class = BertConfig
|
555 |
+
base_model_prefix = "bert"
|
556 |
+
_keys_to_ignore_on_load_missing = [r"position_ids"]
|
557 |
+
|
558 |
+
def _init_weights(self, module):
|
559 |
+
""" Initialize the weights """
|
560 |
+
if isinstance(module, (nn.Linear, nn.Embedding)):
|
561 |
+
# Slightly different from the TF version which uses truncated_normal for initialization
|
562 |
+
# cf https://github.com/pytorch/pytorch/pull/5617
|
563 |
+
module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
|
564 |
+
elif isinstance(module, nn.LayerNorm):
|
565 |
+
module.bias.data.zero_()
|
566 |
+
module.weight.data.fill_(1.0)
|
567 |
+
if isinstance(module, nn.Linear) and module.bias is not None:
|
568 |
+
module.bias.data.zero_()
|
569 |
+
|
570 |
+
|
571 |
+
class BertModel(BertPreTrainedModel):
|
572 |
+
"""
|
573 |
+
The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
|
574 |
+
cross-attention is added between the self-attention layers, following the architecture described in `Attention is
|
575 |
+
all you need <https://arxiv.org/abs/1706.03762>`__ by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
|
576 |
+
Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
|
577 |
+
argument and :obj:`add_cross_attention` set to :obj:`True`; an :obj:`encoder_hidden_states` is then expected as an
|
578 |
+
input to the forward pass.
|
579 |
+
"""
|
580 |
+
|
581 |
+
def __init__(self, config, add_pooling_layer=True):
|
582 |
+
super().__init__(config)
|
583 |
+
self.config = config
|
584 |
+
|
585 |
+
self.embeddings = BertEmbeddings(config)
|
586 |
+
|
587 |
+
self.encoder = BertEncoder(config)
|
588 |
+
|
589 |
+
self.pooler = BertPooler(config) if add_pooling_layer else None
|
590 |
+
|
591 |
+
self.init_weights()
|
592 |
+
|
593 |
+
|
594 |
+
def get_input_embeddings(self):
|
595 |
+
return self.embeddings.word_embeddings
|
596 |
+
|
597 |
+
def set_input_embeddings(self, value):
|
598 |
+
self.embeddings.word_embeddings = value
|
599 |
+
|
600 |
+
def _prune_heads(self, heads_to_prune):
|
601 |
+
"""
|
602 |
+
Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
|
603 |
+
class PreTrainedModel
|
604 |
+
"""
|
605 |
+
for layer, heads in heads_to_prune.items():
|
606 |
+
self.encoder.layer[layer].attention.prune_heads(heads)
|
607 |
+
|
608 |
+
|
609 |
+
def get_extended_attention_mask(self, attention_mask: Tensor, input_shape: Tuple[int], device: device, is_decoder: bool) -> Tensor:
|
610 |
+
"""
|
611 |
+
Makes broadcastable attention and causal masks so that future and masked tokens are ignored.
|
612 |
+
|
613 |
+
Arguments:
|
614 |
+
attention_mask (:obj:`torch.Tensor`):
|
615 |
+
Mask with ones indicating tokens to attend to, zeros for tokens to ignore.
|
616 |
+
input_shape (:obj:`Tuple[int]`):
|
617 |
+
The shape of the input to the model.
|
618 |
+
device: (:obj:`torch.device`):
|
619 |
+
The device of the input to the model.
|
620 |
+
|
621 |
+
Returns:
|
622 |
+
:obj:`torch.Tensor` The extended attention mask, with a the same dtype as :obj:`attention_mask.dtype`.
|
623 |
+
"""
|
624 |
+
# We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
|
625 |
+
# ourselves in which case we just need to make it broadcastable to all heads.
|
626 |
+
if attention_mask.dim() == 3:
|
627 |
+
extended_attention_mask = attention_mask[:, None, :, :]
|
628 |
+
elif attention_mask.dim() == 2:
|
629 |
+
# Provided a padding mask of dimensions [batch_size, seq_length]
|
630 |
+
# - if the model is a decoder, apply a causal mask in addition to the padding mask
|
631 |
+
# - if the model is an encoder, make the mask broadcastable to [batch_size, num_heads, seq_length, seq_length]
|
632 |
+
if is_decoder:
|
633 |
+
batch_size, seq_length = input_shape
|
634 |
+
|
635 |
+
seq_ids = torch.arange(seq_length, device=device)
|
636 |
+
causal_mask = seq_ids[None, None, :].repeat(batch_size, seq_length, 1) <= seq_ids[None, :, None]
|
637 |
+
# in case past_key_values are used we need to add a prefix ones mask to the causal mask
|
638 |
+
# causal and attention masks must have same type with pytorch version < 1.3
|
639 |
+
causal_mask = causal_mask.to(attention_mask.dtype)
|
640 |
+
|
641 |
+
if causal_mask.shape[1] < attention_mask.shape[1]:
|
642 |
+
prefix_seq_len = attention_mask.shape[1] - causal_mask.shape[1]
|
643 |
+
causal_mask = torch.cat(
|
644 |
+
[
|
645 |
+
torch.ones((batch_size, seq_length, prefix_seq_len), device=device, dtype=causal_mask.dtype),
|
646 |
+
causal_mask,
|
647 |
+
],
|
648 |
+
axis=-1,
|
649 |
+
)
|
650 |
+
|
651 |
+
extended_attention_mask = causal_mask[:, None, :, :] * attention_mask[:, None, None, :]
|
652 |
+
else:
|
653 |
+
extended_attention_mask = attention_mask[:, None, None, :]
|
654 |
+
else:
|
655 |
+
raise ValueError(
|
656 |
+
"Wrong shape for input_ids (shape {}) or attention_mask (shape {})".format(
|
657 |
+
input_shape, attention_mask.shape
|
658 |
+
)
|
659 |
+
)
|
660 |
+
|
661 |
+
# Since attention_mask is 1.0 for positions we want to attend and 0.0 for
|
662 |
+
# masked positions, this operation will create a tensor which is 0.0 for
|
663 |
+
# positions we want to attend and -10000.0 for masked positions.
|
664 |
+
# Since we are adding it to the raw scores before the softmax, this is
|
665 |
+
# effectively the same as removing these entirely.
|
666 |
+
extended_attention_mask = extended_attention_mask.to(dtype=self.dtype) # fp16 compatibility
|
667 |
+
extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
|
668 |
+
return extended_attention_mask
|
669 |
+
|
670 |
+
def forward(
|
671 |
+
self,
|
672 |
+
input_ids=None,
|
673 |
+
attention_mask=None,
|
674 |
+
position_ids=None,
|
675 |
+
head_mask=None,
|
676 |
+
inputs_embeds=None,
|
677 |
+
encoder_embeds=None,
|
678 |
+
encoder_hidden_states=None,
|
679 |
+
encoder_attention_mask=None,
|
680 |
+
past_key_values=None,
|
681 |
+
use_cache=None,
|
682 |
+
output_attentions=None,
|
683 |
+
output_hidden_states=None,
|
684 |
+
return_dict=None,
|
685 |
+
is_decoder=False,
|
686 |
+
mode='multimodal',
|
687 |
+
):
|
688 |
+
r"""
|
689 |
+
encoder_hidden_states (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
|
690 |
+
Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
|
691 |
+
the model is configured as a decoder.
|
692 |
+
encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
|
693 |
+
Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
|
694 |
+
the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
|
695 |
+
- 1 for tokens that are **not masked**,
|
696 |
+
- 0 for tokens that are **masked**.
|
697 |
+
past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
|
698 |
+
Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
|
699 |
+
If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
|
700 |
+
(those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
|
701 |
+
instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
|
702 |
+
use_cache (:obj:`bool`, `optional`):
|
703 |
+
If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
|
704 |
+
decoding (see :obj:`past_key_values`).
|
705 |
+
"""
|
706 |
+
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
|
707 |
+
output_hidden_states = (
|
708 |
+
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
|
709 |
+
)
|
710 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
711 |
+
|
712 |
+
if is_decoder:
|
713 |
+
use_cache = use_cache if use_cache is not None else self.config.use_cache
|
714 |
+
else:
|
715 |
+
use_cache = False
|
716 |
+
|
717 |
+
if input_ids is not None and inputs_embeds is not None:
|
718 |
+
raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
|
719 |
+
elif input_ids is not None:
|
720 |
+
input_shape = input_ids.size()
|
721 |
+
batch_size, seq_length = input_shape
|
722 |
+
device = input_ids.device
|
723 |
+
elif inputs_embeds is not None:
|
724 |
+
input_shape = inputs_embeds.size()[:-1]
|
725 |
+
batch_size, seq_length = input_shape
|
726 |
+
device = inputs_embeds.device
|
727 |
+
elif encoder_embeds is not None:
|
728 |
+
input_shape = encoder_embeds.size()[:-1]
|
729 |
+
batch_size, seq_length = input_shape
|
730 |
+
device = encoder_embeds.device
|
731 |
+
else:
|
732 |
+
raise ValueError("You have to specify either input_ids or inputs_embeds or encoder_embeds")
|
733 |
+
|
734 |
+
# past_key_values_length
|
735 |
+
past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
|
736 |
+
|
737 |
+
if attention_mask is None:
|
738 |
+
attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
|
739 |
+
|
740 |
+
# We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
|
741 |
+
# ourselves in which case we just need to make it broadcastable to all heads.
|
742 |
+
extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape,
|
743 |
+
device, is_decoder)
|
744 |
+
|
745 |
+
# If a 2D or 3D attention mask is provided for the cross-attention
|
746 |
+
# we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
|
747 |
+
if encoder_hidden_states is not None:
|
748 |
+
if type(encoder_hidden_states) == list:
|
749 |
+
encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states[0].size()
|
750 |
+
else:
|
751 |
+
encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
|
752 |
+
encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
|
753 |
+
|
754 |
+
if type(encoder_attention_mask) == list:
|
755 |
+
encoder_extended_attention_mask = [self.invert_attention_mask(mask) for mask in encoder_attention_mask]
|
756 |
+
elif encoder_attention_mask is None:
|
757 |
+
encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
|
758 |
+
encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
|
759 |
+
else:
|
760 |
+
encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
|
761 |
+
else:
|
762 |
+
encoder_extended_attention_mask = None
|
763 |
+
|
764 |
+
# Prepare head mask if needed
|
765 |
+
# 1.0 in head_mask indicate we keep the head
|
766 |
+
# attention_probs has shape bsz x n_heads x N x N
|
767 |
+
# input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
|
768 |
+
# and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
|
769 |
+
head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
|
770 |
+
|
771 |
+
if encoder_embeds is None:
|
772 |
+
embedding_output = self.embeddings(
|
773 |
+
input_ids=input_ids,
|
774 |
+
position_ids=position_ids,
|
775 |
+
inputs_embeds=inputs_embeds,
|
776 |
+
past_key_values_length=past_key_values_length,
|
777 |
+
)
|
778 |
+
else:
|
779 |
+
embedding_output = encoder_embeds
|
780 |
+
|
781 |
+
encoder_outputs = self.encoder(
|
782 |
+
embedding_output,
|
783 |
+
attention_mask=extended_attention_mask,
|
784 |
+
head_mask=head_mask,
|
785 |
+
encoder_hidden_states=encoder_hidden_states,
|
786 |
+
encoder_attention_mask=encoder_extended_attention_mask,
|
787 |
+
past_key_values=past_key_values,
|
788 |
+
use_cache=use_cache,
|
789 |
+
output_attentions=output_attentions,
|
790 |
+
output_hidden_states=output_hidden_states,
|
791 |
+
return_dict=return_dict,
|
792 |
+
mode=mode,
|
793 |
+
)
|
794 |
+
sequence_output = encoder_outputs[0]
|
795 |
+
pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
|
796 |
+
|
797 |
+
if not return_dict:
|
798 |
+
return (sequence_output, pooled_output) + encoder_outputs[1:]
|
799 |
+
|
800 |
+
return BaseModelOutputWithPoolingAndCrossAttentions(
|
801 |
+
last_hidden_state=sequence_output,
|
802 |
+
pooler_output=pooled_output,
|
803 |
+
past_key_values=encoder_outputs.past_key_values,
|
804 |
+
hidden_states=encoder_outputs.hidden_states,
|
805 |
+
attentions=encoder_outputs.attentions,
|
806 |
+
cross_attentions=encoder_outputs.cross_attentions,
|
807 |
+
)
|
808 |
+
|
809 |
+
|
810 |
+
|
811 |
+
class BertLMHeadModel(BertPreTrainedModel):
|
812 |
+
|
813 |
+
_keys_to_ignore_on_load_unexpected = [r"pooler"]
|
814 |
+
_keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias"]
|
815 |
+
|
816 |
+
def __init__(self, config):
|
817 |
+
super().__init__(config)
|
818 |
+
|
819 |
+
self.bert = BertModel(config, add_pooling_layer=False)
|
820 |
+
self.cls = BertOnlyMLMHead(config)
|
821 |
+
|
822 |
+
self.init_weights()
|
823 |
+
|
824 |
+
def get_output_embeddings(self):
|
825 |
+
return self.cls.predictions.decoder
|
826 |
+
|
827 |
+
def set_output_embeddings(self, new_embeddings):
|
828 |
+
self.cls.predictions.decoder = new_embeddings
|
829 |
+
|
830 |
+
def forward(
|
831 |
+
self,
|
832 |
+
input_ids=None,
|
833 |
+
attention_mask=None,
|
834 |
+
position_ids=None,
|
835 |
+
head_mask=None,
|
836 |
+
inputs_embeds=None,
|
837 |
+
encoder_hidden_states=None,
|
838 |
+
encoder_attention_mask=None,
|
839 |
+
labels=None,
|
840 |
+
past_key_values=None,
|
841 |
+
use_cache=None,
|
842 |
+
output_attentions=None,
|
843 |
+
output_hidden_states=None,
|
844 |
+
return_dict=None,
|
845 |
+
return_logits=False,
|
846 |
+
is_decoder=True,
|
847 |
+
reduction='mean',
|
848 |
+
mode='multimodal',
|
849 |
+
):
|
850 |
+
r"""
|
851 |
+
encoder_hidden_states (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
|
852 |
+
Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
|
853 |
+
the model is configured as a decoder.
|
854 |
+
encoder_attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
|
855 |
+
Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
|
856 |
+
the cross-attention if the model is configured as a decoder. Mask values selected in ``[0, 1]``:
|
857 |
+
- 1 for tokens that are **not masked**,
|
858 |
+
- 0 for tokens that are **masked**.
|
859 |
+
labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
|
860 |
+
Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in
|
861 |
+
``[-100, 0, ..., config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are
|
862 |
+
ignored (masked), the loss is only computed for the tokens with labels n ``[0, ..., config.vocab_size]``
|
863 |
+
past_key_values (:obj:`tuple(tuple(torch.FloatTensor))` of length :obj:`config.n_layers` with each tuple having 4 tensors of shape :obj:`(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
|
864 |
+
Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
|
865 |
+
If :obj:`past_key_values` are used, the user can optionally input only the last :obj:`decoder_input_ids`
|
866 |
+
(those that don't have their past key value states given to this model) of shape :obj:`(batch_size, 1)`
|
867 |
+
instead of all :obj:`decoder_input_ids` of shape :obj:`(batch_size, sequence_length)`.
|
868 |
+
use_cache (:obj:`bool`, `optional`):
|
869 |
+
If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
|
870 |
+
decoding (see :obj:`past_key_values`).
|
871 |
+
Returns:
|
872 |
+
Example::
|
873 |
+
>>> from transformers import BertTokenizer, BertLMHeadModel, BertConfig
|
874 |
+
>>> import torch
|
875 |
+
>>> tokenizer = BertTokenizer.from_pretrained('bert-base-cased')
|
876 |
+
>>> config = BertConfig.from_pretrained("bert-base-cased")
|
877 |
+
>>> model = BertLMHeadModel.from_pretrained('bert-base-cased', config=config)
|
878 |
+
>>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
|
879 |
+
>>> outputs = model(**inputs)
|
880 |
+
>>> prediction_logits = outputs.logits
|
881 |
+
"""
|
882 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
883 |
+
if labels is not None:
|
884 |
+
use_cache = False
|
885 |
+
|
886 |
+
outputs = self.bert(
|
887 |
+
input_ids,
|
888 |
+
attention_mask=attention_mask,
|
889 |
+
position_ids=position_ids,
|
890 |
+
head_mask=head_mask,
|
891 |
+
inputs_embeds=inputs_embeds,
|
892 |
+
encoder_hidden_states=encoder_hidden_states,
|
893 |
+
encoder_attention_mask=encoder_attention_mask,
|
894 |
+
past_key_values=past_key_values,
|
895 |
+
use_cache=use_cache,
|
896 |
+
output_attentions=output_attentions,
|
897 |
+
output_hidden_states=output_hidden_states,
|
898 |
+
return_dict=return_dict,
|
899 |
+
is_decoder=is_decoder,
|
900 |
+
mode=mode,
|
901 |
+
)
|
902 |
+
|
903 |
+
sequence_output = outputs[0]
|
904 |
+
prediction_scores = self.cls(sequence_output)
|
905 |
+
|
906 |
+
if return_logits:
|
907 |
+
return prediction_scores[:, :-1, :].contiguous()
|
908 |
+
|
909 |
+
lm_loss = None
|
910 |
+
if labels is not None:
|
911 |
+
# we are doing next-token prediction; shift prediction scores and input ids by one
|
912 |
+
shifted_prediction_scores = prediction_scores[:, :-1, :].contiguous()
|
913 |
+
labels = labels[:, 1:].contiguous()
|
914 |
+
loss_fct = CrossEntropyLoss(reduction=reduction, label_smoothing=0.1)
|
915 |
+
lm_loss = loss_fct(shifted_prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))
|
916 |
+
if reduction=='none':
|
917 |
+
lm_loss = lm_loss.view(prediction_scores.size(0),-1).sum(1)
|
918 |
+
|
919 |
+
if not return_dict:
|
920 |
+
output = (prediction_scores,) + outputs[2:]
|
921 |
+
return ((lm_loss,) + output) if lm_loss is not None else output
|
922 |
+
|
923 |
+
return CausalLMOutputWithCrossAttentions(
|
924 |
+
loss=lm_loss,
|
925 |
+
logits=prediction_scores,
|
926 |
+
past_key_values=outputs.past_key_values,
|
927 |
+
hidden_states=outputs.hidden_states,
|
928 |
+
attentions=outputs.attentions,
|
929 |
+
cross_attentions=outputs.cross_attentions,
|
930 |
+
)
|
931 |
+
|
932 |
+
def prepare_inputs_for_generation(self, input_ids, past=None, attention_mask=None, **model_kwargs):
|
933 |
+
input_shape = input_ids.shape
|
934 |
+
# if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
|
935 |
+
if attention_mask is None:
|
936 |
+
attention_mask = input_ids.new_ones(input_shape)
|
937 |
+
|
938 |
+
# cut decoder_input_ids if past is used
|
939 |
+
if past is not None:
|
940 |
+
input_ids = input_ids[:, -1:]
|
941 |
+
|
942 |
+
return {
|
943 |
+
"input_ids": input_ids,
|
944 |
+
"attention_mask": attention_mask,
|
945 |
+
"past_key_values": past,
|
946 |
+
"encoder_hidden_states": model_kwargs.get("encoder_hidden_states", None),
|
947 |
+
"encoder_attention_mask": model_kwargs.get("encoder_attention_mask", None),
|
948 |
+
"is_decoder": True,
|
949 |
+
}
|
950 |
+
|
951 |
+
def _reorder_cache(self, past, beam_idx):
|
952 |
+
reordered_past = ()
|
953 |
+
for layer_past in past:
|
954 |
+
reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
|
955 |
+
return reordered_past
|
models/modeling_clip.py
ADDED
@@ -0,0 +1,1054 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# coding=utf-8
|
2 |
+
# Copyright 2021 The OpenAI Team Authors and The HuggingFace Team. All rights reserved.
|
3 |
+
#
|
4 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
+
# you may not use this file except in compliance with the License.
|
6 |
+
# You may obtain a copy of the License at
|
7 |
+
#
|
8 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
9 |
+
#
|
10 |
+
# Unless required by applicable law or agreed to in writing, software
|
11 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
12 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
13 |
+
# See the License for the specific language governing permissions and
|
14 |
+
# limitations under the License.
|
15 |
+
""" PyTorch CLIP model."""
|
16 |
+
|
17 |
+
|
18 |
+
from dataclasses import dataclass
|
19 |
+
from typing import Any, Optional, Tuple, Union
|
20 |
+
|
21 |
+
import torch
|
22 |
+
import torch.utils.checkpoint
|
23 |
+
from torch import nn
|
24 |
+
|
25 |
+
from transformers.activations import ACT2FN
|
26 |
+
from transformers.file_utils import (
|
27 |
+
ModelOutput,
|
28 |
+
add_start_docstrings,
|
29 |
+
add_start_docstrings_to_model_forward,
|
30 |
+
replace_return_docstrings,
|
31 |
+
)
|
32 |
+
from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
|
33 |
+
from transformers.modeling_utils import PreTrainedModel
|
34 |
+
from transformers.utils import logging
|
35 |
+
from transformers.models.clip.configuration_clip import CLIPConfig, CLIPTextConfig, CLIPVisionConfig
|
36 |
+
|
37 |
+
|
38 |
+
logger = logging.get_logger(__name__)
|
39 |
+
|
40 |
+
_CHECKPOINT_FOR_DOC = "openai/clip-vit-base-patch32"
|
41 |
+
|
42 |
+
CLIP_PRETRAINED_MODEL_ARCHIVE_LIST = [
|
43 |
+
"openai/clip-vit-base-patch32",
|
44 |
+
# See all CLIP models at https://huggingface.co/models?filter=clip
|
45 |
+
]
|
46 |
+
|
47 |
+
|
48 |
+
# Copied from transformers.models.bart.modeling_bart._expand_mask
|
49 |
+
def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
|
50 |
+
"""
|
51 |
+
Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
|
52 |
+
"""
|
53 |
+
bsz, src_len = mask.size()
|
54 |
+
tgt_len = tgt_len if tgt_len is not None else src_len
|
55 |
+
|
56 |
+
expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
|
57 |
+
|
58 |
+
inverted_mask = 1.0 - expanded_mask
|
59 |
+
|
60 |
+
return inverted_mask.masked_fill(inverted_mask.bool(), torch.finfo(dtype).min)
|
61 |
+
|
62 |
+
|
63 |
+
# contrastive loss function, adapted from
|
64 |
+
# https://sachinruk.github.io/blog/pytorch/pytorch%20lightning/loss%20function/gpu/2021/03/07/CLIP.html
|
65 |
+
def contrastive_loss(logits: torch.Tensor) -> torch.Tensor:
|
66 |
+
return nn.functional.cross_entropy(logits, torch.arange(len(logits), device=logits.device))
|
67 |
+
|
68 |
+
|
69 |
+
def clip_loss(similarity: torch.Tensor) -> torch.Tensor:
|
70 |
+
caption_loss = contrastive_loss(similarity)
|
71 |
+
image_loss = contrastive_loss(similarity.T)
|
72 |
+
return (caption_loss + image_loss) / 2.0
|
73 |
+
|
74 |
+
|
75 |
+
@dataclass
|
76 |
+
class CLIPOutput(ModelOutput):
|
77 |
+
"""
|
78 |
+
Args:
|
79 |
+
loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `return_loss` is `True`):
|
80 |
+
Contrastive loss for image-text similarity.
|
81 |
+
logits_per_image:(`torch.FloatTensor` of shape `(image_batch_size, text_batch_size)`):
|
82 |
+
The scaled dot product scores between `image_embeds` and `text_embeds`. This represents the image-text
|
83 |
+
similarity scores.
|
84 |
+
logits_per_text:(`torch.FloatTensor` of shape `(text_batch_size, image_batch_size)`):
|
85 |
+
The scaled dot product scores between `text_embeds` and `image_embeds`. This represents the text-image
|
86 |
+
similarity scores.
|
87 |
+
text_embeds(`torch.FloatTensor` of shape `(batch_size, output_dim`):
|
88 |
+
The text embeddings obtained by applying the projection layer to the pooled output of [`CLIPTextModel`].
|
89 |
+
image_embeds(`torch.FloatTensor` of shape `(batch_size, output_dim`):
|
90 |
+
The image embeddings obtained by applying the projection layer to the pooled output of [`CLIPVisionModel`].
|
91 |
+
text_model_output(`BaseModelOutputWithPooling`):
|
92 |
+
The output of the [`CLIPTextModel`].
|
93 |
+
vision_model_output(`BaseModelOutputWithPooling`):
|
94 |
+
The output of the [`CLIPVisionModel`].
|
95 |
+
"""
|
96 |
+
|
97 |
+
loss: Optional[torch.FloatTensor] = None
|
98 |
+
logits_per_image: torch.FloatTensor = None
|
99 |
+
logits_per_text: torch.FloatTensor = None
|
100 |
+
text_embeds: torch.FloatTensor = None
|
101 |
+
image_embeds: torch.FloatTensor = None
|
102 |
+
text_model_output: BaseModelOutputWithPooling = None
|
103 |
+
vision_model_output: BaseModelOutputWithPooling = None
|
104 |
+
|
105 |
+
def to_tuple(self) -> Tuple[Any]:
|
106 |
+
return tuple(
|
107 |
+
self[k] if k not in ["text_model_output", "vision_model_output"] else getattr(self, k).to_tuple()
|
108 |
+
for k in self.keys()
|
109 |
+
)
|
110 |
+
|
111 |
+
|
112 |
+
class CLIPVisionEmbeddings(nn.Module):
|
113 |
+
def __init__(self, config: CLIPVisionConfig):
|
114 |
+
super().__init__()
|
115 |
+
self.config = config
|
116 |
+
self.embed_dim = config.hidden_size
|
117 |
+
self.image_size = config.image_size
|
118 |
+
self.patch_size = config.patch_size
|
119 |
+
|
120 |
+
self.class_embedding = nn.Parameter(torch.randn(self.embed_dim))
|
121 |
+
|
122 |
+
self.patch_embedding = nn.Conv2d(
|
123 |
+
in_channels=3, out_channels=self.embed_dim, kernel_size=self.patch_size, stride=self.patch_size, bias=False
|
124 |
+
)
|
125 |
+
|
126 |
+
self.num_patches = (self.image_size // self.patch_size) ** 2
|
127 |
+
self.num_positions = self.num_patches + 1
|
128 |
+
self.position_embedding = nn.Embedding(self.num_positions, self.embed_dim)
|
129 |
+
self.register_buffer("position_ids", torch.arange(self.num_positions).expand((1, -1)))
|
130 |
+
|
131 |
+
def forward(self, pixel_values: torch.FloatTensor) -> torch.Tensor:
|
132 |
+
batch_size = pixel_values.shape[0]
|
133 |
+
patch_embeds = self.patch_embedding(pixel_values) # shape = [*, width, grid, grid]
|
134 |
+
patch_embeds = patch_embeds.flatten(2).transpose(1, 2)
|
135 |
+
|
136 |
+
class_embeds = self.class_embedding.expand(batch_size, 1, -1)
|
137 |
+
embeddings = torch.cat([class_embeds, patch_embeds], dim=1)
|
138 |
+
embeddings = embeddings + self.position_embedding(self.position_ids)
|
139 |
+
return embeddings
|
140 |
+
|
141 |
+
|
142 |
+
class CLIPTextEmbeddings(nn.Module):
|
143 |
+
def __init__(self, config: CLIPTextConfig):
|
144 |
+
super().__init__()
|
145 |
+
embed_dim = config.hidden_size
|
146 |
+
|
147 |
+
self.token_embedding = nn.Embedding(config.vocab_size, embed_dim)
|
148 |
+
self.position_embedding = nn.Embedding(config.max_position_embeddings, embed_dim)
|
149 |
+
|
150 |
+
# position_ids (1, len position emb) is contiguous in memory and exported when serialized
|
151 |
+
self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))
|
152 |
+
|
153 |
+
def forward(
|
154 |
+
self,
|
155 |
+
input_ids: Optional[torch.LongTensor] = None,
|
156 |
+
position_ids: Optional[torch.LongTensor] = None,
|
157 |
+
inputs_embeds: Optional[torch.FloatTensor] = None,
|
158 |
+
) -> torch.Tensor:
|
159 |
+
seq_length = input_ids.shape[-1] if input_ids is not None else inputs_embeds.shape[-2]
|
160 |
+
|
161 |
+
if position_ids is None:
|
162 |
+
position_ids = self.position_ids[:, :seq_length]
|
163 |
+
|
164 |
+
if inputs_embeds is None:
|
165 |
+
inputs_embeds = self.token_embedding(input_ids)
|
166 |
+
|
167 |
+
position_embeddings = self.position_embedding(position_ids)
|
168 |
+
embeddings = inputs_embeds + position_embeddings
|
169 |
+
|
170 |
+
return embeddings
|
171 |
+
|
172 |
+
|
173 |
+
class CLIPAttention(nn.Module):
|
174 |
+
"""Multi-headed attention from 'Attention Is All You Need' paper"""
|
175 |
+
|
176 |
+
def __init__(self, config):
|
177 |
+
super().__init__()
|
178 |
+
self.config = config
|
179 |
+
self.embed_dim = config.hidden_size
|
180 |
+
self.num_heads = config.num_attention_heads
|
181 |
+
self.head_dim = self.embed_dim // self.num_heads
|
182 |
+
assert (
|
183 |
+
self.head_dim * self.num_heads == self.embed_dim
|
184 |
+
), f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`: {self.num_heads})."
|
185 |
+
self.scale = self.head_dim**-0.5
|
186 |
+
self.dropout = config.attention_dropout
|
187 |
+
|
188 |
+
self.k_proj = nn.Linear(self.embed_dim, self.embed_dim)
|
189 |
+
self.v_proj = nn.Linear(self.embed_dim, self.embed_dim)
|
190 |
+
self.q_proj = nn.Linear(self.embed_dim, self.embed_dim)
|
191 |
+
self.out_proj = nn.Linear(self.embed_dim, self.embed_dim)
|
192 |
+
|
193 |
+
def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
|
194 |
+
return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
|
195 |
+
|
196 |
+
def forward(
|
197 |
+
self,
|
198 |
+
hidden_states: torch.Tensor,
|
199 |
+
attention_mask: Optional[torch.Tensor] = None,
|
200 |
+
causal_attention_mask: Optional[torch.Tensor] = None,
|
201 |
+
output_attentions: Optional[bool] = False,
|
202 |
+
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
|
203 |
+
"""Input shape: Batch x Time x Channel"""
|
204 |
+
|
205 |
+
bsz, tgt_len, embed_dim = hidden_states.size()
|
206 |
+
|
207 |
+
# get query proj
|
208 |
+
query_states = self.q_proj(hidden_states) * self.scale
|
209 |
+
key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
|
210 |
+
value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
|
211 |
+
|
212 |
+
proj_shape = (bsz * self.num_heads, -1, self.head_dim)
|
213 |
+
query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
|
214 |
+
key_states = key_states.view(*proj_shape)
|
215 |
+
value_states = value_states.view(*proj_shape)
|
216 |
+
|
217 |
+
src_len = key_states.size(1)
|
218 |
+
attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))
|
219 |
+
|
220 |
+
if attn_weights.size() != (bsz * self.num_heads, tgt_len, src_len):
|
221 |
+
raise ValueError(
|
222 |
+
f"Attention weights should be of size {(bsz * self.num_heads, tgt_len, src_len)}, but is {attn_weights.size()}"
|
223 |
+
)
|
224 |
+
|
225 |
+
# apply the causal_attention_mask first
|
226 |
+
if causal_attention_mask is not None:
|
227 |
+
if causal_attention_mask.size() != (bsz, 1, tgt_len, src_len):
|
228 |
+
raise ValueError(
|
229 |
+
f"Attention mask should be of size {(bsz, 1, tgt_len, src_len)}, but is {causal_attention_mask.size()}"
|
230 |
+
)
|
231 |
+
attn_weights = attn_weights.view(bsz, self.num_heads, tgt_len, src_len) + causal_attention_mask
|
232 |
+
attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)
|
233 |
+
|
234 |
+
if attention_mask is not None:
|
235 |
+
if attention_mask.size() != (bsz, 1, tgt_len, src_len):
|
236 |
+
raise ValueError(
|
237 |
+
f"Attention mask should be of size {(bsz, 1, tgt_len, src_len)}, but is {attention_mask.size()}"
|
238 |
+
)
|
239 |
+
attn_weights = attn_weights.view(bsz, self.num_heads, tgt_len, src_len) + attention_mask
|
240 |
+
attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)
|
241 |
+
|
242 |
+
attn_weights = nn.functional.softmax(attn_weights, dim=-1)
|
243 |
+
|
244 |
+
if output_attentions:
|
245 |
+
# this operation is a bit akward, but it's required to
|
246 |
+
# make sure that attn_weights keeps its gradient.
|
247 |
+
# In order to do so, attn_weights have to reshaped
|
248 |
+
# twice and have to be reused in the following
|
249 |
+
attn_weights_reshaped = attn_weights.view(bsz, self.num_heads, tgt_len, src_len)
|
250 |
+
attn_weights = attn_weights_reshaped.view(bsz * self.num_heads, tgt_len, src_len)
|
251 |
+
else:
|
252 |
+
attn_weights_reshaped = None
|
253 |
+
|
254 |
+
attn_probs = nn.functional.dropout(attn_weights, p=self.dropout, training=self.training)
|
255 |
+
|
256 |
+
attn_output = torch.bmm(attn_probs, value_states)
|
257 |
+
|
258 |
+
if attn_output.size() != (bsz * self.num_heads, tgt_len, self.head_dim):
|
259 |
+
raise ValueError(
|
260 |
+
f"`attn_output` should be of size {(bsz, self.num_heads, tgt_len, self.head_dim)}, but is {attn_output.size()}"
|
261 |
+
)
|
262 |
+
|
263 |
+
attn_output = attn_output.view(bsz, self.num_heads, tgt_len, self.head_dim)
|
264 |
+
attn_output = attn_output.transpose(1, 2)
|
265 |
+
attn_output = attn_output.reshape(bsz, tgt_len, embed_dim)
|
266 |
+
|
267 |
+
attn_output = self.out_proj(attn_output)
|
268 |
+
|
269 |
+
return attn_output, attn_weights_reshaped
|
270 |
+
|
271 |
+
|
272 |
+
class CLIPMLP(nn.Module):
|
273 |
+
def __init__(self, config):
|
274 |
+
super().__init__()
|
275 |
+
self.config = config
|
276 |
+
self.activation_fn = ACT2FN[config.hidden_act]
|
277 |
+
self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
|
278 |
+
self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)
|
279 |
+
|
280 |
+
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
|
281 |
+
hidden_states = self.fc1(hidden_states)
|
282 |
+
hidden_states = self.activation_fn(hidden_states)
|
283 |
+
hidden_states = self.fc2(hidden_states)
|
284 |
+
return hidden_states
|
285 |
+
|
286 |
+
|
287 |
+
class CLIPEncoderLayer(nn.Module):
|
288 |
+
def __init__(self, config: CLIPConfig):
|
289 |
+
super().__init__()
|
290 |
+
self.embed_dim = config.hidden_size
|
291 |
+
self.self_attn = CLIPAttention(config)
|
292 |
+
self.layer_norm1 = nn.LayerNorm(self.embed_dim)
|
293 |
+
self.mlp = CLIPMLP(config)
|
294 |
+
self.layer_norm2 = nn.LayerNorm(self.embed_dim)
|
295 |
+
|
296 |
+
def forward(
|
297 |
+
self,
|
298 |
+
hidden_states: torch.Tensor,
|
299 |
+
attention_mask: torch.Tensor,
|
300 |
+
causal_attention_mask: torch.Tensor,
|
301 |
+
output_attentions: Optional[bool] = False,
|
302 |
+
) -> Tuple[torch.FloatTensor]:
|
303 |
+
"""
|
304 |
+
Args:
|
305 |
+
hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
|
306 |
+
attention_mask (`torch.FloatTensor`): attention mask of size
|
307 |
+
`(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
|
308 |
+
`(config.encoder_attention_heads,)`.
|
309 |
+
output_attentions (`bool`, *optional*):
|
310 |
+
Whether or not to return the attentions tensors of all attention layers. See `attentions` under
|
311 |
+
returned tensors for more detail.
|
312 |
+
"""
|
313 |
+
residual = hidden_states
|
314 |
+
|
315 |
+
hidden_states = self.layer_norm1(hidden_states)
|
316 |
+
hidden_states, attn_weights = self.self_attn(
|
317 |
+
hidden_states=hidden_states,
|
318 |
+
attention_mask=attention_mask,
|
319 |
+
causal_attention_mask=causal_attention_mask,
|
320 |
+
output_attentions=output_attentions,
|
321 |
+
)
|
322 |
+
hidden_states = residual + hidden_states
|
323 |
+
|
324 |
+
residual = hidden_states
|
325 |
+
hidden_states = self.layer_norm2(hidden_states)
|
326 |
+
hidden_states = self.mlp(hidden_states)
|
327 |
+
hidden_states = residual + hidden_states
|
328 |
+
|
329 |
+
outputs = (hidden_states,)
|
330 |
+
|
331 |
+
if output_attentions:
|
332 |
+
outputs += (attn_weights,)
|
333 |
+
|
334 |
+
return outputs
|
335 |
+
|
336 |
+
|
337 |
+
class CLIPPreTrainedModel(PreTrainedModel):
|
338 |
+
"""
|
339 |
+
An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
|
340 |
+
models.
|
341 |
+
"""
|
342 |
+
|
343 |
+
config_class = CLIPConfig
|
344 |
+
base_model_prefix = "clip"
|
345 |
+
supports_gradient_checkpointing = True
|
346 |
+
_keys_to_ignore_on_load_missing = [r"position_ids"]
|
347 |
+
|
348 |
+
def _init_weights(self, module):
|
349 |
+
"""Initialize the weights"""
|
350 |
+
factor = self.config.initializer_factor
|
351 |
+
if isinstance(module, CLIPTextEmbeddings):
|
352 |
+
module.token_embedding.weight.data.normal_(mean=0.0, std=factor * 0.02)
|
353 |
+
module.position_embedding.weight.data.normal_(mean=0.0, std=factor * 0.02)
|
354 |
+
elif isinstance(module, CLIPVisionEmbeddings):
|
355 |
+
factor = self.config.initializer_factor
|
356 |
+
nn.init.normal_(module.class_embedding, mean=0.0, std=module.embed_dim**-0.5 * factor)
|
357 |
+
nn.init.normal_(module.patch_embedding.weight, std=module.config.initializer_range * factor)
|
358 |
+
nn.init.normal_(module.position_embedding.weight, std=module.config.initializer_range * factor)
|
359 |
+
elif isinstance(module, CLIPAttention):
|
360 |
+
factor = self.config.initializer_factor
|
361 |
+
in_proj_std = (module.embed_dim**-0.5) * ((2 * module.config.num_hidden_layers) ** -0.5) * factor
|
362 |
+
out_proj_std = (module.embed_dim**-0.5) * factor
|
363 |
+
nn.init.normal_(module.q_proj.weight, std=in_proj_std)
|
364 |
+
nn.init.normal_(module.k_proj.weight, std=in_proj_std)
|
365 |
+
nn.init.normal_(module.v_proj.weight, std=in_proj_std)
|
366 |
+
nn.init.normal_(module.out_proj.weight, std=out_proj_std)
|
367 |
+
elif isinstance(module, CLIPMLP):
|
368 |
+
factor = self.config.initializer_factor
|
369 |
+
in_proj_std = (
|
370 |
+
(module.config.hidden_size**-0.5) * ((2 * module.config.num_hidden_layers) ** -0.5) * factor
|
371 |
+
)
|
372 |
+
fc_std = (2 * module.config.hidden_size) ** -0.5 * factor
|
373 |
+
nn.init.normal_(module.fc1.weight, std=fc_std)
|
374 |
+
nn.init.normal_(module.fc2.weight, std=in_proj_std)
|
375 |
+
elif isinstance(module, CLIPModel):
|
376 |
+
nn.init.normal_(
|
377 |
+
module.text_projection.weight,
|
378 |
+
std=module.text_embed_dim**-0.5 * self.config.initializer_factor,
|
379 |
+
)
|
380 |
+
nn.init.normal_(
|
381 |
+
module.visual_projection.weight,
|
382 |
+
std=module.vision_embed_dim**-0.5 * self.config.initializer_factor,
|
383 |
+
)
|
384 |
+
|
385 |
+
if isinstance(module, nn.LayerNorm):
|
386 |
+
module.bias.data.zero_()
|
387 |
+
module.weight.data.fill_(1.0)
|
388 |
+
if isinstance(module, nn.Linear) and module.bias is not None:
|
389 |
+
module.bias.data.zero_()
|
390 |
+
|
391 |
+
def _set_gradient_checkpointing(self, module, value=False):
|
392 |
+
if isinstance(module, CLIPEncoder):
|
393 |
+
module.gradient_checkpointing = value
|
394 |
+
|
395 |
+
|
396 |
+
CLIP_START_DOCSTRING = r"""
|
397 |
+
This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass. Use it
|
398 |
+
as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and
|
399 |
+
behavior.
|
400 |
+
|
401 |
+
Parameters:
|
402 |
+
config ([`CLIPConfig`]): Model configuration class with all the parameters of the model.
|
403 |
+
Initializing with a config file does not load the weights associated with the model, only the
|
404 |
+
configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
|
405 |
+
"""
|
406 |
+
|
407 |
+
CLIP_TEXT_INPUTS_DOCSTRING = r"""
|
408 |
+
Args:
|
409 |
+
input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
|
410 |
+
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
|
411 |
+
it.
|
412 |
+
|
413 |
+
Indices can be obtained using [`CLIPTokenizer`]. See [`PreTrainedTokenizer.encode`] and
|
414 |
+
[`PreTrainedTokenizer.__call__`] for details.
|
415 |
+
|
416 |
+
[What are input IDs?](../glossary#input-ids)
|
417 |
+
attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
|
418 |
+
Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
|
419 |
+
|
420 |
+
- 1 for tokens that are **not masked**,
|
421 |
+
- 0 for tokens that are **masked**.
|
422 |
+
|
423 |
+
[What are attention masks?](../glossary#attention-mask)
|
424 |
+
position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
|
425 |
+
Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
|
426 |
+
config.max_position_embeddings - 1]`.
|
427 |
+
|
428 |
+
[What are position IDs?](../glossary#position-ids)
|
429 |
+
output_attentions (`bool`, *optional*):
|
430 |
+
Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
|
431 |
+
tensors for more detail.
|
432 |
+
output_hidden_states (`bool`, *optional*):
|
433 |
+
Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
|
434 |
+
more detail.
|
435 |
+
return_dict (`bool`, *optional*):
|
436 |
+
Whether or not to return a [`~file_utils.ModelOutput`] instead of a plain tuple.
|
437 |
+
"""
|
438 |
+
|
439 |
+
CLIP_VISION_INPUTS_DOCSTRING = r"""
|
440 |
+
Args:
|
441 |
+
pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
|
442 |
+
Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained using
|
443 |
+
[`CLIPFeatureExtractor`]. See [`CLIPFeatureExtractor.__call__`] for details.
|
444 |
+
output_attentions (`bool`, *optional*):
|
445 |
+
Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
|
446 |
+
tensors for more detail.
|
447 |
+
output_hidden_states (`bool`, *optional*):
|
448 |
+
Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
|
449 |
+
more detail.
|
450 |
+
return_dict (`bool`, *optional*):
|
451 |
+
Whether or not to return a [`~file_utils.ModelOutput`] instead of a plain tuple.
|
452 |
+
"""
|
453 |
+
|
454 |
+
CLIP_INPUTS_DOCSTRING = r"""
|
455 |
+
Args:
|
456 |
+
input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
|
457 |
+
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
|
458 |
+
it.
|
459 |
+
|
460 |
+
Indices can be obtained using [`CLIPTokenizer`]. See [`PreTrainedTokenizer.encode`] and
|
461 |
+
[`PreTrainedTokenizer.__call__`] for details.
|
462 |
+
|
463 |
+
[What are input IDs?](../glossary#input-ids)
|
464 |
+
attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
|
465 |
+
Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
|
466 |
+
|
467 |
+
- 1 for tokens that are **not masked**,
|
468 |
+
- 0 for tokens that are **masked**.
|
469 |
+
|
470 |
+
[What are attention masks?](../glossary#attention-mask)
|
471 |
+
position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
|
472 |
+
Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
|
473 |
+
config.max_position_embeddings - 1]`.
|
474 |
+
|
475 |
+
[What are position IDs?](../glossary#position-ids)
|
476 |
+
pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
|
477 |
+
Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained using
|
478 |
+
[`CLIPFeatureExtractor`]. See [`CLIPFeatureExtractor.__call__`] for details.
|
479 |
+
return_loss (`bool`, *optional*):
|
480 |
+
Whether or not to return the contrastive loss.
|
481 |
+
output_attentions (`bool`, *optional*):
|
482 |
+
Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
|
483 |
+
tensors for more detail.
|
484 |
+
output_hidden_states (`bool`, *optional*):
|
485 |
+
Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
|
486 |
+
more detail.
|
487 |
+
return_dict (`bool`, *optional*):
|
488 |
+
Whether or not to return a [`~file_utils.ModelOutput`] instead of a plain tuple.
|
489 |
+
"""
|
490 |
+
|
491 |
+
|
492 |
+
class CLIPEncoder(nn.Module):
|
493 |
+
"""
|
494 |
+
Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a
|
495 |
+
[`CLIPEncoderLayer`].
|
496 |
+
|
497 |
+
Args:
|
498 |
+
config: CLIPConfig
|
499 |
+
"""
|
500 |
+
|
501 |
+
def __init__(self, config: CLIPConfig):
|
502 |
+
super().__init__()
|
503 |
+
self.config = config
|
504 |
+
self.layers = nn.ModuleList([CLIPEncoderLayer(config) for _ in range(config.num_hidden_layers)])
|
505 |
+
self.gradient_checkpointing = False
|
506 |
+
|
507 |
+
def forward(
|
508 |
+
self,
|
509 |
+
inputs_embeds,
|
510 |
+
attention_mask: Optional[torch.Tensor] = None,
|
511 |
+
causal_attention_mask: Optional[torch.Tensor] = None,
|
512 |
+
output_attentions: Optional[bool] = None,
|
513 |
+
output_hidden_states: Optional[bool] = None,
|
514 |
+
return_dict: Optional[bool] = None,
|
515 |
+
) -> Union[Tuple, BaseModelOutput]:
|
516 |
+
r"""
|
517 |
+
Args:
|
518 |
+
inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
|
519 |
+
Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
|
520 |
+
This is useful if you want more control over how to convert `input_ids` indices into associated vectors
|
521 |
+
than the model's internal embedding lookup matrix.
|
522 |
+
attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
|
523 |
+
Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
|
524 |
+
|
525 |
+
- 1 for tokens that are **not masked**,
|
526 |
+
- 0 for tokens that are **masked**.
|
527 |
+
|
528 |
+
[What are attention masks?](../glossary#attention-mask)
|
529 |
+
causal_attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
|
530 |
+
Causal mask for the text model. Mask values selected in `[0, 1]`:
|
531 |
+
|
532 |
+
- 1 for tokens that are **not masked**,
|
533 |
+
- 0 for tokens that are **masked**.
|
534 |
+
|
535 |
+
[What are attention masks?](../glossary#attention-mask)
|
536 |
+
output_attentions (`bool`, *optional*):
|
537 |
+
Whether or not to return the attentions tensors of all attention layers. See `attentions` under
|
538 |
+
returned tensors for more detail.
|
539 |
+
output_hidden_states (`bool`, *optional*):
|
540 |
+
Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
|
541 |
+
for more detail.
|
542 |
+
return_dict (`bool`, *optional*):
|
543 |
+
Whether or not to return a [`~file_utils.ModelOutput`] instead of a plain tuple.
|
544 |
+
"""
|
545 |
+
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
|
546 |
+
output_hidden_states = (
|
547 |
+
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
|
548 |
+
)
|
549 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
550 |
+
|
551 |
+
encoder_states = () if output_hidden_states else None
|
552 |
+
all_attentions = () if output_attentions else None
|
553 |
+
|
554 |
+
hidden_states = inputs_embeds
|
555 |
+
for idx, encoder_layer in enumerate(self.layers):
|
556 |
+
if output_hidden_states:
|
557 |
+
encoder_states = encoder_states + (hidden_states,)
|
558 |
+
if self.gradient_checkpointing and self.training:
|
559 |
+
|
560 |
+
def create_custom_forward(module):
|
561 |
+
def custom_forward(*inputs):
|
562 |
+
return module(*inputs, output_attentions)
|
563 |
+
|
564 |
+
return custom_forward
|
565 |
+
|
566 |
+
layer_outputs = torch.utils.checkpoint.checkpoint(
|
567 |
+
create_custom_forward(encoder_layer),
|
568 |
+
hidden_states,
|
569 |
+
attention_mask,
|
570 |
+
causal_attention_mask,
|
571 |
+
)
|
572 |
+
else:
|
573 |
+
layer_outputs = encoder_layer(
|
574 |
+
hidden_states,
|
575 |
+
attention_mask,
|
576 |
+
causal_attention_mask,
|
577 |
+
output_attentions=output_attentions,
|
578 |
+
)
|
579 |
+
|
580 |
+
hidden_states = layer_outputs[0]
|
581 |
+
|
582 |
+
if output_attentions:
|
583 |
+
all_attentions = all_attentions + (layer_outputs[1],)
|
584 |
+
|
585 |
+
if output_hidden_states:
|
586 |
+
encoder_states = encoder_states + (hidden_states,)
|
587 |
+
|
588 |
+
if not return_dict:
|
589 |
+
return tuple(v for v in [hidden_states, encoder_states, all_attentions] if v is not None)
|
590 |
+
return BaseModelOutput(
|
591 |
+
last_hidden_state=hidden_states, hidden_states=encoder_states, attentions=all_attentions
|
592 |
+
)
|
593 |
+
|
594 |
+
|
595 |
+
class CLIPTextTransformer(nn.Module):
|
596 |
+
def __init__(self, config: CLIPTextConfig):
|
597 |
+
super().__init__()
|
598 |
+
self.config = config
|
599 |
+
embed_dim = config.hidden_size
|
600 |
+
self.embeddings = CLIPTextEmbeddings(config)
|
601 |
+
self.encoder = CLIPEncoder(config)
|
602 |
+
self.final_layer_norm = nn.LayerNorm(embed_dim)
|
603 |
+
|
604 |
+
@add_start_docstrings_to_model_forward(CLIP_TEXT_INPUTS_DOCSTRING)
|
605 |
+
@replace_return_docstrings(output_type=BaseModelOutputWithPooling, config_class=CLIPTextConfig)
|
606 |
+
def forward(
|
607 |
+
self,
|
608 |
+
input_ids: Optional[torch.Tensor] = None,
|
609 |
+
attention_mask: Optional[torch.Tensor] = None,
|
610 |
+
position_ids: Optional[torch.Tensor] = None,
|
611 |
+
output_attentions: Optional[bool] = None,
|
612 |
+
output_hidden_states: Optional[bool] = None,
|
613 |
+
return_dict: Optional[bool] = None,
|
614 |
+
) -> Union[Tuple, BaseModelOutputWithPooling]:
|
615 |
+
r"""
|
616 |
+
Returns:
|
617 |
+
|
618 |
+
"""
|
619 |
+
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
|
620 |
+
output_hidden_states = (
|
621 |
+
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
|
622 |
+
)
|
623 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
624 |
+
|
625 |
+
if input_ids is None:
|
626 |
+
raise ValueError("You have to specify either input_ids")
|
627 |
+
|
628 |
+
input_shape = input_ids.size()
|
629 |
+
input_ids = input_ids.view(-1, input_shape[-1])
|
630 |
+
|
631 |
+
hidden_states = self.embeddings(input_ids=input_ids, position_ids=position_ids)
|
632 |
+
|
633 |
+
bsz, seq_len = input_shape
|
634 |
+
# CLIP's text model uses causal mask, prepare it here.
|
635 |
+
# https://github.com/openai/CLIP/blob/cfcffb90e69f37bf2ff1e988237a0fbe41f33c04/clip/model.py#L324
|
636 |
+
causal_attention_mask = self._build_causal_attention_mask(bsz, seq_len).to(hidden_states.device)
|
637 |
+
# expand attention_mask
|
638 |
+
if attention_mask is not None:
|
639 |
+
# [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
|
640 |
+
attention_mask = _expand_mask(attention_mask, hidden_states.dtype)
|
641 |
+
|
642 |
+
encoder_outputs = self.encoder(
|
643 |
+
inputs_embeds=hidden_states,
|
644 |
+
attention_mask=attention_mask,
|
645 |
+
causal_attention_mask=causal_attention_mask,
|
646 |
+
output_attentions=output_attentions,
|
647 |
+
output_hidden_states=output_hidden_states,
|
648 |
+
return_dict=return_dict,
|
649 |
+
)
|
650 |
+
|
651 |
+
last_hidden_state = encoder_outputs[0]
|
652 |
+
last_hidden_state = self.final_layer_norm(last_hidden_state)
|
653 |
+
|
654 |
+
# text_embeds.shape = [batch_size, sequence_length, transformer.width]
|
655 |
+
# take features from the eot embedding (eot_token is the highest number in each sequence)
|
656 |
+
pooled_output = last_hidden_state[torch.arange(last_hidden_state.shape[0]), input_ids.argmax(dim=-1)]
|
657 |
+
|
658 |
+
if not return_dict:
|
659 |
+
return (last_hidden_state, pooled_output) + encoder_outputs[1:]
|
660 |
+
|
661 |
+
return BaseModelOutputWithPooling(
|
662 |
+
last_hidden_state=last_hidden_state,
|
663 |
+
pooler_output=pooled_output,
|
664 |
+
hidden_states=encoder_outputs.hidden_states,
|
665 |
+
attentions=encoder_outputs.attentions,
|
666 |
+
)
|
667 |
+
|
668 |
+
def _build_causal_attention_mask(self, bsz, seq_len):
|
669 |
+
# lazily create causal attention mask, with full attention between the vision tokens
|
670 |
+
# pytorch uses additive attention mask; fill with -inf
|
671 |
+
mask = torch.empty(bsz, seq_len, seq_len)
|
672 |
+
mask.fill_(float("-inf"))
|
673 |
+
mask.triu_(1) # zero out the lower diagonal
|
674 |
+
mask = mask.unsqueeze(1) # expand mask
|
675 |
+
return mask
|
676 |
+
|
677 |
+
|
678 |
+
class CLIPTextModel(CLIPPreTrainedModel):
|
679 |
+
config_class = CLIPTextConfig
|
680 |
+
|
681 |
+
def __init__(self, config: CLIPTextConfig):
|
682 |
+
super().__init__(config)
|
683 |
+
self.text_model = CLIPTextTransformer(config)
|
684 |
+
# Initialize weights and apply final processing
|
685 |
+
self.post_init()
|
686 |
+
|
687 |
+
def get_input_embeddings(self) -> nn.Module:
|
688 |
+
return self.text_model.embeddings.token_embedding
|
689 |
+
|
690 |
+
def set_input_embeddings(self, value):
|
691 |
+
self.text_model.embeddings.token_embedding = value
|
692 |
+
|
693 |
+
@add_start_docstrings_to_model_forward(CLIP_TEXT_INPUTS_DOCSTRING)
|
694 |
+
@replace_return_docstrings(output_type=BaseModelOutputWithPooling, config_class=CLIPTextConfig)
|
695 |
+
def forward(
|
696 |
+
self,
|
697 |
+
input_ids: Optional[torch.Tensor] = None,
|
698 |
+
attention_mask: Optional[torch.Tensor] = None,
|
699 |
+
position_ids: Optional[torch.Tensor] = None,
|
700 |
+
output_attentions: Optional[bool] = None,
|
701 |
+
output_hidden_states: Optional[bool] = None,
|
702 |
+
return_dict: Optional[bool] = None,
|
703 |
+
) -> Union[Tuple, BaseModelOutputWithPooling]:
|
704 |
+
r"""
|
705 |
+
Returns:
|
706 |
+
|
707 |
+
Examples:
|
708 |
+
|
709 |
+
```python
|
710 |
+
>>> from transformers import CLIPTokenizer, CLIPTextModel
|
711 |
+
|
712 |
+
>>> model = CLIPTextModel.from_pretrained("openai/clip-vit-base-patch32")
|
713 |
+
>>> tokenizer = CLIPTokenizer.from_pretrained("openai/clip-vit-base-patch32")
|
714 |
+
|
715 |
+
>>> inputs = tokenizer(["a photo of a cat", "a photo of a dog"], padding=True, return_tensors="pt")
|
716 |
+
|
717 |
+
>>> outputs = model(**inputs)
|
718 |
+
>>> last_hidden_state = outputs.last_hidden_state
|
719 |
+
>>> pooled_output = outputs.pooler_output # pooled (EOS token) states
|
720 |
+
```"""
|
721 |
+
return self.text_model(
|
722 |
+
input_ids=input_ids,
|
723 |
+
attention_mask=attention_mask,
|
724 |
+
position_ids=position_ids,
|
725 |
+
output_attentions=output_attentions,
|
726 |
+
output_hidden_states=output_hidden_states,
|
727 |
+
return_dict=return_dict,
|
728 |
+
)
|
729 |
+
|
730 |
+
|
731 |
+
class CLIPVisionTransformer(nn.Module):
|
732 |
+
def __init__(self, config: CLIPVisionConfig):
|
733 |
+
super().__init__()
|
734 |
+
self.config = config
|
735 |
+
embed_dim = config.hidden_size
|
736 |
+
|
737 |
+
self.embeddings = CLIPVisionEmbeddings(config)
|
738 |
+
self.pre_layrnorm = nn.LayerNorm(embed_dim)
|
739 |
+
self.encoder = CLIPEncoder(config)
|
740 |
+
self.post_layernorm = nn.LayerNorm(embed_dim)
|
741 |
+
|
742 |
+
@add_start_docstrings_to_model_forward(CLIP_VISION_INPUTS_DOCSTRING)
|
743 |
+
@replace_return_docstrings(output_type=BaseModelOutputWithPooling, config_class=CLIPVisionConfig)
|
744 |
+
def forward(
|
745 |
+
self,
|
746 |
+
pixel_values: Optional[torch.FloatTensor] = None,
|
747 |
+
output_attentions: Optional[bool] = None,
|
748 |
+
output_hidden_states: Optional[bool] = None,
|
749 |
+
return_dict: Optional[bool] = None,
|
750 |
+
) -> Union[Tuple, BaseModelOutputWithPooling]:
|
751 |
+
r"""
|
752 |
+
Returns:
|
753 |
+
|
754 |
+
"""
|
755 |
+
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
|
756 |
+
output_hidden_states = (
|
757 |
+
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
|
758 |
+
)
|
759 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
760 |
+
|
761 |
+
if pixel_values is None:
|
762 |
+
raise ValueError("You have to specify pixel_values")
|
763 |
+
|
764 |
+
hidden_states = self.embeddings(pixel_values)
|
765 |
+
hidden_states = self.pre_layrnorm(hidden_states)
|
766 |
+
|
767 |
+
encoder_outputs = self.encoder(
|
768 |
+
inputs_embeds=hidden_states,
|
769 |
+
output_attentions=output_attentions,
|
770 |
+
output_hidden_states=output_hidden_states,
|
771 |
+
return_dict=return_dict,
|
772 |
+
)
|
773 |
+
|
774 |
+
last_hidden_state = encoder_outputs[0]
|
775 |
+
pooled_output = last_hidden_state[:, 0, :]
|
776 |
+
pooled_output = self.post_layernorm(pooled_output)
|
777 |
+
|
778 |
+
if not return_dict:
|
779 |
+
return (last_hidden_state, pooled_output) + encoder_outputs[1:]
|
780 |
+
|
781 |
+
return BaseModelOutputWithPooling(
|
782 |
+
last_hidden_state=last_hidden_state,
|
783 |
+
pooler_output=pooled_output,
|
784 |
+
hidden_states=encoder_outputs.hidden_states,
|
785 |
+
attentions=encoder_outputs.attentions,
|
786 |
+
)
|
787 |
+
|
788 |
+
|
789 |
+
class CLIPVisionModel(CLIPPreTrainedModel):
|
790 |
+
config_class = CLIPVisionConfig
|
791 |
+
main_input_name = "pixel_values"
|
792 |
+
|
793 |
+
def __init__(self, config: CLIPVisionConfig):
|
794 |
+
super().__init__(config)
|
795 |
+
self.vision_model = CLIPVisionTransformer(config)
|
796 |
+
# Initialize weights and apply final processing
|
797 |
+
self.post_norm = nn.LayerNorm(config.hidden_size, eps=1e-6)
|
798 |
+
self.post_init()
|
799 |
+
|
800 |
+
def get_input_embeddings(self) -> nn.Module:
|
801 |
+
return self.vision_model.embeddings.patch_embedding
|
802 |
+
|
803 |
+
@add_start_docstrings_to_model_forward(CLIP_VISION_INPUTS_DOCSTRING)
|
804 |
+
@replace_return_docstrings(output_type=BaseModelOutputWithPooling, config_class=CLIPVisionConfig)
|
805 |
+
def forward(
|
806 |
+
self,
|
807 |
+
pixel_values: Optional[torch.FloatTensor] = None,
|
808 |
+
output_attentions: Optional[bool] = None,
|
809 |
+
output_hidden_states: Optional[bool] = None,
|
810 |
+
return_dict: Optional[bool] = None,
|
811 |
+
) -> Union[Tuple, BaseModelOutputWithPooling]:
|
812 |
+
r"""
|
813 |
+
Returns:
|
814 |
+
|
815 |
+
Examples:
|
816 |
+
|
817 |
+
```python
|
818 |
+
>>> from PIL import Image
|
819 |
+
>>> import requests
|
820 |
+
>>> from transformers import CLIPProcessor, CLIPVisionModel
|
821 |
+
|
822 |
+
>>> model = CLIPVisionModel.from_pretrained("openai/clip-vit-base-patch32")
|
823 |
+
>>> processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
|
824 |
+
|
825 |
+
>>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
|
826 |
+
>>> image = Image.open(requests.get(url, stream=True).raw)
|
827 |
+
|
828 |
+
>>> inputs = processor(images=image, return_tensors="pt")
|
829 |
+
|
830 |
+
>>> outputs = model(**inputs)
|
831 |
+
>>> last_hidden_state = outputs.last_hidden_state
|
832 |
+
>>> pooled_output = outputs.pooler_output # pooled CLS states
|
833 |
+
```"""
|
834 |
+
# return self.vision_model(
|
835 |
+
# pixel_values=pixel_values,
|
836 |
+
# output_attentions=output_attentions,
|
837 |
+
# output_hidden_states=output_hidden_states,
|
838 |
+
# return_dict=return_dict,
|
839 |
+
# )
|
840 |
+
|
841 |
+
result = self.vision_model(
|
842 |
+
pixel_values=pixel_values,
|
843 |
+
output_attentions=output_attentions,
|
844 |
+
output_hidden_states=output_hidden_states,
|
845 |
+
return_dict=return_dict,
|
846 |
+
)
|
847 |
+
result = result.last_hidden_state
|
848 |
+
# todo post norm
|
849 |
+
result = self.post_norm(result)
|
850 |
+
return result
|
851 |
+
|
852 |
+
|
853 |
+
@add_start_docstrings(CLIP_START_DOCSTRING)
|
854 |
+
class CLIPModel(CLIPPreTrainedModel):
|
855 |
+
config_class = CLIPConfig
|
856 |
+
|
857 |
+
def __init__(self, config: CLIPConfig):
|
858 |
+
super().__init__(config)
|
859 |
+
|
860 |
+
if not isinstance(config.text_config, CLIPTextConfig):
|
861 |
+
raise ValueError(
|
862 |
+
f"config.text_config is expected to be of type CLIPTextConfig but is of type {type(config.text_config)}."
|
863 |
+
)
|
864 |
+
|
865 |
+
if not isinstance(config.vision_config, CLIPVisionConfig):
|
866 |
+
raise ValueError(
|
867 |
+
f"config.vision_config is expected to be of type CLIPVisionConfig but is of type {type(config.vision_config)}."
|
868 |
+
)
|
869 |
+
|
870 |
+
text_config = config.text_config
|
871 |
+
vision_config = config.vision_config
|
872 |
+
|
873 |
+
self.projection_dim = config.projection_dim
|
874 |
+
self.text_embed_dim = text_config.hidden_size
|
875 |
+
self.vision_embed_dim = vision_config.hidden_size
|
876 |
+
|
877 |
+
self.text_model = CLIPTextTransformer(text_config)
|
878 |
+
self.vision_model = CLIPVisionTransformer(vision_config)
|
879 |
+
|
880 |
+
self.visual_projection = nn.Linear(self.vision_embed_dim, self.projection_dim, bias=False)
|
881 |
+
self.text_projection = nn.Linear(self.text_embed_dim, self.projection_dim, bias=False)
|
882 |
+
self.logit_scale = nn.Parameter(torch.ones([]) * self.config.logit_scale_init_value)
|
883 |
+
|
884 |
+
# Initialize weights and apply final processing
|
885 |
+
self.post_init()
|
886 |
+
|
887 |
+
@add_start_docstrings_to_model_forward(CLIP_TEXT_INPUTS_DOCSTRING)
|
888 |
+
def get_text_features(
|
889 |
+
self,
|
890 |
+
input_ids: Optional[torch.Tensor] = None,
|
891 |
+
attention_mask: Optional[torch.Tensor] = None,
|
892 |
+
position_ids: Optional[torch.Tensor] = None,
|
893 |
+
output_attentions: Optional[bool] = None,
|
894 |
+
output_hidden_states: Optional[bool] = None,
|
895 |
+
return_dict: Optional[bool] = None,
|
896 |
+
) -> torch.FloatTensor:
|
897 |
+
r"""
|
898 |
+
Returns:
|
899 |
+
text_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The text embeddings obtained by
|
900 |
+
applying the projection layer to the pooled output of [`CLIPTextModel`].
|
901 |
+
|
902 |
+
Examples:
|
903 |
+
|
904 |
+
```python
|
905 |
+
>>> from transformers import CLIPTokenizer, CLIPModel
|
906 |
+
|
907 |
+
>>> model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
|
908 |
+
>>> tokenizer = CLIPTokenizer.from_pretrained("openai/clip-vit-base-patch32")
|
909 |
+
|
910 |
+
>>> inputs = tokenizer(["a photo of a cat", "a photo of a dog"], padding=True, return_tensors="pt")
|
911 |
+
>>> text_features = model.get_text_features(**inputs)
|
912 |
+
```"""
|
913 |
+
text_outputs = self.text_model(
|
914 |
+
input_ids=input_ids,
|
915 |
+
attention_mask=attention_mask,
|
916 |
+
position_ids=position_ids,
|
917 |
+
output_attentions=output_attentions,
|
918 |
+
output_hidden_states=output_hidden_states,
|
919 |
+
return_dict=return_dict,
|
920 |
+
)
|
921 |
+
|
922 |
+
pooled_output = text_outputs[1]
|
923 |
+
text_features = self.text_projection(pooled_output)
|
924 |
+
|
925 |
+
return text_features
|
926 |
+
|
927 |
+
@add_start_docstrings_to_model_forward(CLIP_VISION_INPUTS_DOCSTRING)
|
928 |
+
def get_image_features(
|
929 |
+
self,
|
930 |
+
pixel_values: Optional[torch.FloatTensor] = None,
|
931 |
+
output_attentions: Optional[bool] = None,
|
932 |
+
output_hidden_states: Optional[bool] = None,
|
933 |
+
return_dict: Optional[bool] = None,
|
934 |
+
) -> torch.FloatTensor:
|
935 |
+
r"""
|
936 |
+
Returns:
|
937 |
+
image_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The image embeddings obtained by
|
938 |
+
applying the projection layer to the pooled output of [`CLIPVisionModel`].
|
939 |
+
|
940 |
+
Examples:
|
941 |
+
|
942 |
+
```python
|
943 |
+
>>> from PIL import Image
|
944 |
+
>>> import requests
|
945 |
+
>>> from transformers import CLIPProcessor, CLIPModel
|
946 |
+
|
947 |
+
>>> model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
|
948 |
+
>>> processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
|
949 |
+
|
950 |
+
>>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
|
951 |
+
>>> image = Image.open(requests.get(url, stream=True).raw)
|
952 |
+
|
953 |
+
>>> inputs = processor(images=image, return_tensors="pt")
|
954 |
+
|
955 |
+
>>> image_features = model.get_image_features(**inputs)
|
956 |
+
```"""
|
957 |
+
vision_outputs = self.vision_model(
|
958 |
+
pixel_values=pixel_values,
|
959 |
+
output_attentions=output_attentions,
|
960 |
+
output_hidden_states=output_hidden_states,
|
961 |
+
return_dict=return_dict,
|
962 |
+
)
|
963 |
+
|
964 |
+
pooled_output = vision_outputs[1] # pooled_output
|
965 |
+
image_features = self.visual_projection(pooled_output)
|
966 |
+
|
967 |
+
return image_features
|
968 |
+
|
969 |
+
@add_start_docstrings_to_model_forward(CLIP_INPUTS_DOCSTRING)
|
970 |
+
@replace_return_docstrings(output_type=CLIPOutput, config_class=CLIPConfig)
|
971 |
+
def forward(
|
972 |
+
self,
|
973 |
+
input_ids: Optional[torch.LongTensor] = None,
|
974 |
+
pixel_values: Optional[torch.FloatTensor] = None,
|
975 |
+
attention_mask: Optional[torch.Tensor] = None,
|
976 |
+
position_ids: Optional[torch.LongTensor] = None,
|
977 |
+
return_loss: Optional[bool] = None,
|
978 |
+
output_attentions: Optional[bool] = None,
|
979 |
+
output_hidden_states: Optional[bool] = None,
|
980 |
+
return_dict: Optional[bool] = None,
|
981 |
+
) -> Union[Tuple, CLIPOutput]:
|
982 |
+
r"""
|
983 |
+
Returns:
|
984 |
+
|
985 |
+
Examples:
|
986 |
+
|
987 |
+
```python
|
988 |
+
>>> from PIL import Image
|
989 |
+
>>> import requests
|
990 |
+
>>> from transformers import CLIPProcessor, CLIPModel
|
991 |
+
|
992 |
+
>>> model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
|
993 |
+
>>> processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
|
994 |
+
|
995 |
+
>>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
|
996 |
+
>>> image = Image.open(requests.get(url, stream=True).raw)
|
997 |
+
|
998 |
+
>>> inputs = processor(
|
999 |
+
... text=["a photo of a cat", "a photo of a dog"], images=image, return_tensors="pt", padding=True
|
1000 |
+
... )
|
1001 |
+
|
1002 |
+
>>> outputs = model(**inputs)
|
1003 |
+
>>> logits_per_image = outputs.logits_per_image # this is the image-text similarity score
|
1004 |
+
>>> probs = logits_per_image.softmax(dim=1) # we can take the softmax to get the label probabilities
|
1005 |
+
```"""
|
1006 |
+
return_dict = return_dict if return_dict is not None else self.config.return_dict
|
1007 |
+
vision_outputs = self.vision_model(
|
1008 |
+
pixel_values=pixel_values,
|
1009 |
+
output_attentions=output_attentions,
|
1010 |
+
output_hidden_states=output_hidden_states,
|
1011 |
+
return_dict=return_dict,
|
1012 |
+
)
|
1013 |
+
|
1014 |
+
text_outputs = self.text_model(
|
1015 |
+
input_ids=input_ids,
|
1016 |
+
attention_mask=attention_mask,
|
1017 |
+
position_ids=position_ids,
|
1018 |
+
output_attentions=output_attentions,
|
1019 |
+
output_hidden_states=output_hidden_states,
|
1020 |
+
return_dict=return_dict,
|
1021 |
+
)
|
1022 |
+
|
1023 |
+
image_embeds = vision_outputs[1]
|
1024 |
+
image_embeds = self.visual_projection(image_embeds)
|
1025 |
+
|
1026 |
+
text_embeds = text_outputs[1]
|
1027 |
+
text_embeds = self.text_projection(text_embeds)
|
1028 |
+
|
1029 |
+
# normalized features
|
1030 |
+
image_embeds = image_embeds / image_embeds.norm(dim=-1, keepdim=True)
|
1031 |
+
text_embeds = text_embeds / text_embeds.norm(dim=-1, keepdim=True)
|
1032 |
+
|
1033 |
+
# cosine similarity as logits
|
1034 |
+
logit_scale = self.logit_scale.exp()
|
1035 |
+
logits_per_text = torch.matmul(text_embeds, image_embeds.t()) * logit_scale
|
1036 |
+
logits_per_image = logits_per_text.T
|
1037 |
+
|
1038 |
+
loss = None
|
1039 |
+
if return_loss:
|
1040 |
+
loss = clip_loss(logits_per_text)
|
1041 |
+
|
1042 |
+
if not return_dict:
|
1043 |
+
output = (logits_per_image, logits_per_text, text_embeds, image_embeds, text_outputs, vision_outputs)
|
1044 |
+
return ((loss,) + output) if loss is not None else output
|
1045 |
+
|
1046 |
+
return CLIPOutput(
|
1047 |
+
loss=loss,
|
1048 |
+
logits_per_image=logits_per_image,
|
1049 |
+
logits_per_text=logits_per_text,
|
1050 |
+
text_embeds=text_embeds,
|
1051 |
+
image_embeds=image_embeds,
|
1052 |
+
text_model_output=text_outputs,
|
1053 |
+
vision_model_output=vision_outputs,
|
1054 |
+
)
|
models/vit.py
ADDED
@@ -0,0 +1,305 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
'''
|
2 |
+
* Copyright (c) 2022, salesforce.com, inc.
|
3 |
+
* All rights reserved.
|
4 |
+
* SPDX-License-Identifier: BSD-3-Clause
|
5 |
+
* For full license text, see LICENSE.txt file in the repo root or https://opensource.org/licenses/BSD-3-Clause
|
6 |
+
* By Junnan Li
|
7 |
+
* Based on timm code base
|
8 |
+
* https://github.com/rwightman/pytorch-image-models/tree/master/timm
|
9 |
+
'''
|
10 |
+
|
11 |
+
import torch
|
12 |
+
import torch.nn as nn
|
13 |
+
import torch.nn.functional as F
|
14 |
+
from functools import partial
|
15 |
+
|
16 |
+
from timm.models.vision_transformer import _cfg, PatchEmbed
|
17 |
+
from timm.models.registry import register_model
|
18 |
+
from timm.models.layers import trunc_normal_, DropPath
|
19 |
+
from timm.models.helpers import named_apply, adapt_input_conv
|
20 |
+
|
21 |
+
from fairscale.nn.checkpoint.checkpoint_activations import checkpoint_wrapper
|
22 |
+
|
23 |
+
class Mlp(nn.Module):
|
24 |
+
""" MLP as used in Vision Transformer, MLP-Mixer and related networks
|
25 |
+
"""
|
26 |
+
def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
|
27 |
+
super().__init__()
|
28 |
+
out_features = out_features or in_features
|
29 |
+
hidden_features = hidden_features or in_features
|
30 |
+
self.fc1 = nn.Linear(in_features, hidden_features)
|
31 |
+
self.act = act_layer()
|
32 |
+
self.fc2 = nn.Linear(hidden_features, out_features)
|
33 |
+
self.drop = nn.Dropout(drop)
|
34 |
+
|
35 |
+
def forward(self, x):
|
36 |
+
x = self.fc1(x)
|
37 |
+
x = self.act(x)
|
38 |
+
x = self.drop(x)
|
39 |
+
x = self.fc2(x)
|
40 |
+
x = self.drop(x)
|
41 |
+
return x
|
42 |
+
|
43 |
+
|
44 |
+
class Attention(nn.Module):
|
45 |
+
def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
|
46 |
+
super().__init__()
|
47 |
+
self.num_heads = num_heads
|
48 |
+
head_dim = dim // num_heads
|
49 |
+
# NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
|
50 |
+
self.scale = qk_scale or head_dim ** -0.5
|
51 |
+
self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
|
52 |
+
self.attn_drop = nn.Dropout(attn_drop)
|
53 |
+
self.proj = nn.Linear(dim, dim)
|
54 |
+
self.proj_drop = nn.Dropout(proj_drop)
|
55 |
+
self.attn_gradients = None
|
56 |
+
self.attention_map = None
|
57 |
+
|
58 |
+
def save_attn_gradients(self, attn_gradients):
|
59 |
+
self.attn_gradients = attn_gradients
|
60 |
+
|
61 |
+
def get_attn_gradients(self):
|
62 |
+
return self.attn_gradients
|
63 |
+
|
64 |
+
def save_attention_map(self, attention_map):
|
65 |
+
self.attention_map = attention_map
|
66 |
+
|
67 |
+
def get_attention_map(self):
|
68 |
+
return self.attention_map
|
69 |
+
|
70 |
+
def forward(self, x, register_hook=False):
|
71 |
+
B, N, C = x.shape
|
72 |
+
qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
|
73 |
+
q, k, v = qkv[0], qkv[1], qkv[2] # make torchscript happy (cannot use tensor as tuple)
|
74 |
+
|
75 |
+
attn = (q @ k.transpose(-2, -1)) * self.scale
|
76 |
+
attn = attn.softmax(dim=-1)
|
77 |
+
attn = self.attn_drop(attn)
|
78 |
+
|
79 |
+
if register_hook:
|
80 |
+
self.save_attention_map(attn)
|
81 |
+
attn.register_hook(self.save_attn_gradients)
|
82 |
+
|
83 |
+
x = (attn @ v).transpose(1, 2).reshape(B, N, C)
|
84 |
+
x = self.proj(x)
|
85 |
+
x = self.proj_drop(x)
|
86 |
+
return x
|
87 |
+
|
88 |
+
|
89 |
+
class Block(nn.Module):
|
90 |
+
|
91 |
+
def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
|
92 |
+
drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, use_grad_checkpointing=False):
|
93 |
+
super().__init__()
|
94 |
+
self.norm1 = norm_layer(dim)
|
95 |
+
self.attn = Attention(
|
96 |
+
dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
|
97 |
+
# NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
|
98 |
+
self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
|
99 |
+
self.norm2 = norm_layer(dim)
|
100 |
+
mlp_hidden_dim = int(dim * mlp_ratio)
|
101 |
+
self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
|
102 |
+
|
103 |
+
if use_grad_checkpointing:
|
104 |
+
self.attn = checkpoint_wrapper(self.attn)
|
105 |
+
self.mlp = checkpoint_wrapper(self.mlp)
|
106 |
+
|
107 |
+
def forward(self, x, register_hook=False):
|
108 |
+
x = x + self.drop_path(self.attn(self.norm1(x), register_hook=register_hook))
|
109 |
+
x = x + self.drop_path(self.mlp(self.norm2(x)))
|
110 |
+
return x
|
111 |
+
|
112 |
+
|
113 |
+
class VisionTransformer(nn.Module):
|
114 |
+
""" Vision Transformer
|
115 |
+
A PyTorch impl of : `An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale` -
|
116 |
+
https://arxiv.org/abs/2010.11929
|
117 |
+
"""
|
118 |
+
def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dim=768, depth=12,
|
119 |
+
num_heads=12, mlp_ratio=4., qkv_bias=True, qk_scale=None, representation_size=None,
|
120 |
+
drop_rate=0., attn_drop_rate=0., drop_path_rate=0., norm_layer=None,
|
121 |
+
use_grad_checkpointing=False, ckpt_layer=0):
|
122 |
+
"""
|
123 |
+
Args:
|
124 |
+
img_size (int, tuple): input image size
|
125 |
+
patch_size (int, tuple): patch size
|
126 |
+
in_chans (int): number of input channels
|
127 |
+
num_classes (int): number of classes for classification head
|
128 |
+
embed_dim (int): embedding dimension
|
129 |
+
depth (int): depth of transformer
|
130 |
+
num_heads (int): number of attention heads
|
131 |
+
mlp_ratio (int): ratio of mlp hidden dim to embedding dim
|
132 |
+
qkv_bias (bool): enable bias for qkv if True
|
133 |
+
qk_scale (float): override default qk scale of head_dim ** -0.5 if set
|
134 |
+
representation_size (Optional[int]): enable and set representation layer (pre-logits) to this value if set
|
135 |
+
drop_rate (float): dropout rate
|
136 |
+
attn_drop_rate (float): attention dropout rate
|
137 |
+
drop_path_rate (float): stochastic depth rate
|
138 |
+
norm_layer: (nn.Module): normalization layer
|
139 |
+
"""
|
140 |
+
super().__init__()
|
141 |
+
self.num_features = self.embed_dim = embed_dim # num_features for consistency with other models
|
142 |
+
norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)
|
143 |
+
|
144 |
+
self.patch_embed = PatchEmbed(
|
145 |
+
img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim)
|
146 |
+
|
147 |
+
num_patches = self.patch_embed.num_patches
|
148 |
+
|
149 |
+
self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
|
150 |
+
self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim))
|
151 |
+
self.pos_drop = nn.Dropout(p=drop_rate)
|
152 |
+
|
153 |
+
dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)] # stochastic depth decay rule
|
154 |
+
self.blocks = nn.ModuleList([
|
155 |
+
Block(
|
156 |
+
dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
|
157 |
+
drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,
|
158 |
+
use_grad_checkpointing=(use_grad_checkpointing and i>=depth-ckpt_layer)
|
159 |
+
)
|
160 |
+
for i in range(depth)])
|
161 |
+
self.norm = norm_layer(embed_dim)
|
162 |
+
|
163 |
+
trunc_normal_(self.pos_embed, std=.02)
|
164 |
+
trunc_normal_(self.cls_token, std=.02)
|
165 |
+
self.apply(self._init_weights)
|
166 |
+
|
167 |
+
def _init_weights(self, m):
|
168 |
+
if isinstance(m, nn.Linear):
|
169 |
+
trunc_normal_(m.weight, std=.02)
|
170 |
+
if isinstance(m, nn.Linear) and m.bias is not None:
|
171 |
+
nn.init.constant_(m.bias, 0)
|
172 |
+
elif isinstance(m, nn.LayerNorm):
|
173 |
+
nn.init.constant_(m.bias, 0)
|
174 |
+
nn.init.constant_(m.weight, 1.0)
|
175 |
+
|
176 |
+
@torch.jit.ignore
|
177 |
+
def no_weight_decay(self):
|
178 |
+
return {'pos_embed', 'cls_token'}
|
179 |
+
|
180 |
+
def forward(self, x, register_blk=-1):
|
181 |
+
B = x.shape[0]
|
182 |
+
x = self.patch_embed(x)
|
183 |
+
|
184 |
+
cls_tokens = self.cls_token.expand(B, -1, -1) # stole cls_tokens impl from Phil Wang, thanks
|
185 |
+
x = torch.cat((cls_tokens, x), dim=1)
|
186 |
+
|
187 |
+
x = x + self.pos_embed[:,:x.size(1),:]
|
188 |
+
x = self.pos_drop(x)
|
189 |
+
|
190 |
+
for i,blk in enumerate(self.blocks):
|
191 |
+
x = blk(x, register_blk==i)
|
192 |
+
x = self.norm(x)
|
193 |
+
|
194 |
+
return x
|
195 |
+
|
196 |
+
@torch.jit.ignore()
|
197 |
+
def load_pretrained(self, checkpoint_path, prefix=''):
|
198 |
+
_load_weights(self, checkpoint_path, prefix)
|
199 |
+
|
200 |
+
|
201 |
+
@torch.no_grad()
|
202 |
+
def _load_weights(model: VisionTransformer, checkpoint_path: str, prefix: str = ''):
|
203 |
+
""" Load weights from .npz checkpoints for official Google Brain Flax implementation
|
204 |
+
"""
|
205 |
+
import numpy as np
|
206 |
+
|
207 |
+
def _n2p(w, t=True):
|
208 |
+
if w.ndim == 4 and w.shape[0] == w.shape[1] == w.shape[2] == 1:
|
209 |
+
w = w.flatten()
|
210 |
+
if t:
|
211 |
+
if w.ndim == 4:
|
212 |
+
w = w.transpose([3, 2, 0, 1])
|
213 |
+
elif w.ndim == 3:
|
214 |
+
w = w.transpose([2, 0, 1])
|
215 |
+
elif w.ndim == 2:
|
216 |
+
w = w.transpose([1, 0])
|
217 |
+
return torch.from_numpy(w)
|
218 |
+
|
219 |
+
w = np.load(checkpoint_path)
|
220 |
+
if not prefix and 'opt/target/embedding/kernel' in w:
|
221 |
+
prefix = 'opt/target/'
|
222 |
+
|
223 |
+
if hasattr(model.patch_embed, 'backbone'):
|
224 |
+
# hybrid
|
225 |
+
backbone = model.patch_embed.backbone
|
226 |
+
stem_only = not hasattr(backbone, 'stem')
|
227 |
+
stem = backbone if stem_only else backbone.stem
|
228 |
+
stem.conv.weight.copy_(adapt_input_conv(stem.conv.weight.shape[1], _n2p(w[f'{prefix}conv_root/kernel'])))
|
229 |
+
stem.norm.weight.copy_(_n2p(w[f'{prefix}gn_root/scale']))
|
230 |
+
stem.norm.bias.copy_(_n2p(w[f'{prefix}gn_root/bias']))
|
231 |
+
if not stem_only:
|
232 |
+
for i, stage in enumerate(backbone.stages):
|
233 |
+
for j, block in enumerate(stage.blocks):
|
234 |
+
bp = f'{prefix}block{i + 1}/unit{j + 1}/'
|
235 |
+
for r in range(3):
|
236 |
+
getattr(block, f'conv{r + 1}').weight.copy_(_n2p(w[f'{bp}conv{r + 1}/kernel']))
|
237 |
+
getattr(block, f'norm{r + 1}').weight.copy_(_n2p(w[f'{bp}gn{r + 1}/scale']))
|
238 |
+
getattr(block, f'norm{r + 1}').bias.copy_(_n2p(w[f'{bp}gn{r + 1}/bias']))
|
239 |
+
if block.downsample is not None:
|
240 |
+
block.downsample.conv.weight.copy_(_n2p(w[f'{bp}conv_proj/kernel']))
|
241 |
+
block.downsample.norm.weight.copy_(_n2p(w[f'{bp}gn_proj/scale']))
|
242 |
+
block.downsample.norm.bias.copy_(_n2p(w[f'{bp}gn_proj/bias']))
|
243 |
+
embed_conv_w = _n2p(w[f'{prefix}embedding/kernel'])
|
244 |
+
else:
|
245 |
+
embed_conv_w = adapt_input_conv(
|
246 |
+
model.patch_embed.proj.weight.shape[1], _n2p(w[f'{prefix}embedding/kernel']))
|
247 |
+
model.patch_embed.proj.weight.copy_(embed_conv_w)
|
248 |
+
model.patch_embed.proj.bias.copy_(_n2p(w[f'{prefix}embedding/bias']))
|
249 |
+
model.cls_token.copy_(_n2p(w[f'{prefix}cls'], t=False))
|
250 |
+
pos_embed_w = _n2p(w[f'{prefix}Transformer/posembed_input/pos_embedding'], t=False)
|
251 |
+
if pos_embed_w.shape != model.pos_embed.shape:
|
252 |
+
pos_embed_w = resize_pos_embed( # resize pos embedding when different size from pretrained weights
|
253 |
+
pos_embed_w, model.pos_embed, getattr(model, 'num_tokens', 1), model.patch_embed.grid_size)
|
254 |
+
model.pos_embed.copy_(pos_embed_w)
|
255 |
+
model.norm.weight.copy_(_n2p(w[f'{prefix}Transformer/encoder_norm/scale']))
|
256 |
+
model.norm.bias.copy_(_n2p(w[f'{prefix}Transformer/encoder_norm/bias']))
|
257 |
+
# if isinstance(model.head, nn.Linear) and model.head.bias.shape[0] == w[f'{prefix}head/bias'].shape[-1]:
|
258 |
+
# model.head.weight.copy_(_n2p(w[f'{prefix}head/kernel']))
|
259 |
+
# model.head.bias.copy_(_n2p(w[f'{prefix}head/bias']))
|
260 |
+
# if isinstance(getattr(model.pre_logits, 'fc', None), nn.Linear) and f'{prefix}pre_logits/bias' in w:
|
261 |
+
# model.pre_logits.fc.weight.copy_(_n2p(w[f'{prefix}pre_logits/kernel']))
|
262 |
+
# model.pre_logits.fc.bias.copy_(_n2p(w[f'{prefix}pre_logits/bias']))
|
263 |
+
for i, block in enumerate(model.blocks.children()):
|
264 |
+
block_prefix = f'{prefix}Transformer/encoderblock_{i}/'
|
265 |
+
mha_prefix = block_prefix + 'MultiHeadDotProductAttention_1/'
|
266 |
+
block.norm1.weight.copy_(_n2p(w[f'{block_prefix}LayerNorm_0/scale']))
|
267 |
+
block.norm1.bias.copy_(_n2p(w[f'{block_prefix}LayerNorm_0/bias']))
|
268 |
+
block.attn.qkv.weight.copy_(torch.cat([
|
269 |
+
_n2p(w[f'{mha_prefix}{n}/kernel'], t=False).flatten(1).T for n in ('query', 'key', 'value')]))
|
270 |
+
block.attn.qkv.bias.copy_(torch.cat([
|
271 |
+
_n2p(w[f'{mha_prefix}{n}/bias'], t=False).reshape(-1) for n in ('query', 'key', 'value')]))
|
272 |
+
block.attn.proj.weight.copy_(_n2p(w[f'{mha_prefix}out/kernel']).flatten(1))
|
273 |
+
block.attn.proj.bias.copy_(_n2p(w[f'{mha_prefix}out/bias']))
|
274 |
+
for r in range(2):
|
275 |
+
getattr(block.mlp, f'fc{r + 1}').weight.copy_(_n2p(w[f'{block_prefix}MlpBlock_3/Dense_{r}/kernel']))
|
276 |
+
getattr(block.mlp, f'fc{r + 1}').bias.copy_(_n2p(w[f'{block_prefix}MlpBlock_3/Dense_{r}/bias']))
|
277 |
+
block.norm2.weight.copy_(_n2p(w[f'{block_prefix}LayerNorm_2/scale']))
|
278 |
+
block.norm2.bias.copy_(_n2p(w[f'{block_prefix}LayerNorm_2/bias']))
|
279 |
+
|
280 |
+
|
281 |
+
def interpolate_pos_embed(pos_embed_checkpoint, visual_encoder):
|
282 |
+
# interpolate position embedding
|
283 |
+
embedding_size = pos_embed_checkpoint.shape[-1]
|
284 |
+
num_patches = visual_encoder.patch_embed.num_patches
|
285 |
+
num_extra_tokens = visual_encoder.pos_embed.shape[-2] - num_patches
|
286 |
+
# height (== width) for the checkpoint position embedding
|
287 |
+
orig_size = int((pos_embed_checkpoint.shape[-2] - num_extra_tokens) ** 0.5)
|
288 |
+
# height (== width) for the new position embedding
|
289 |
+
new_size = int(num_patches ** 0.5)
|
290 |
+
|
291 |
+
if orig_size!=new_size:
|
292 |
+
# class_token and dist_token are kept unchanged
|
293 |
+
extra_tokens = pos_embed_checkpoint[:, :num_extra_tokens]
|
294 |
+
# only the position tokens are interpolated
|
295 |
+
pos_tokens = pos_embed_checkpoint[:, num_extra_tokens:]
|
296 |
+
pos_tokens = pos_tokens.reshape(-1, orig_size, orig_size, embedding_size).permute(0, 3, 1, 2)
|
297 |
+
pos_tokens = torch.nn.functional.interpolate(
|
298 |
+
pos_tokens, size=(new_size, new_size), mode='bicubic', align_corners=False)
|
299 |
+
pos_tokens = pos_tokens.permute(0, 2, 3, 1).flatten(1, 2)
|
300 |
+
new_pos_embed = torch.cat((extra_tokens, pos_tokens), dim=1)
|
301 |
+
print('reshape position embedding from %d to %d'%(orig_size ** 2,new_size ** 2))
|
302 |
+
|
303 |
+
return new_pos_embed
|
304 |
+
else:
|
305 |
+
return pos_embed_checkpoint
|
requirements.txt
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
timm==0.4.12
|
2 |
+
transformers==4.15.0
|
3 |
+
fairscale==0.4.4
|
4 |
+
pycocoevalcap
|
5 |
+
torch
|
6 |
+
torchvision
|
7 |
+
Pillow
|
resources/bert-large-chinese/config.json
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"BertForMaskedLM"
|
4 |
+
],
|
5 |
+
"attention_probs_dropout_prob": 0.1,
|
6 |
+
"bos_token_id": 0,
|
7 |
+
"directionality": "bidi",
|
8 |
+
"eos_token_id": 2,
|
9 |
+
"hidden_act": "gelu",
|
10 |
+
"hidden_dropout_prob": 0.1,
|
11 |
+
"hidden_size": 1024,
|
12 |
+
"initializer_range": 0.02,
|
13 |
+
"intermediate_size": 4096,
|
14 |
+
"layer_norm_eps": 1e-12,
|
15 |
+
"max_position_embeddings": 512,
|
16 |
+
"model_type": "bert",
|
17 |
+
"num_attention_heads": 16,
|
18 |
+
"num_hidden_layers": 24,
|
19 |
+
"output_past": true,
|
20 |
+
"pad_token_id": 0,
|
21 |
+
"pooler_fc_size": 768,
|
22 |
+
"pooler_num_attention_heads": 12,
|
23 |
+
"pooler_num_fc_layers": 3,
|
24 |
+
"pooler_size_per_head": 128,
|
25 |
+
"pooler_type": "first_token_transform",
|
26 |
+
"type_vocab_size": 2,
|
27 |
+
"vocab_size": 21128
|
28 |
+
}
|
resources/bert-large-chinese/tokenizer_config.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"init_inputs": []}
|
resources/bert-large-chinese/vocab.txt
ADDED
The diff for this file is too large to render.
See raw diff
|
|
resources/clip_vit_large_patch14/config.json
ADDED
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "/mnt/dolphinfs/hdd_pool/docker/user/hadoop-dpsr/zhangzhe45/huggingface/openai/clip-vit-large-patch14/",
|
3 |
+
"attention_dropout": 0.0,
|
4 |
+
"dropout": 0.0,
|
5 |
+
"hidden_act": "quick_gelu",
|
6 |
+
"hidden_size": 1024,
|
7 |
+
"image_size": 224,
|
8 |
+
"initializer_factor": 1.0,
|
9 |
+
"initializer_range": 0.02,
|
10 |
+
"intermediate_size": 4096,
|
11 |
+
"layer_norm_eps": 1e-05,
|
12 |
+
"model_type": "clip_vision_model",
|
13 |
+
"num_attention_heads": 16,
|
14 |
+
"num_hidden_layers": 24,
|
15 |
+
"patch_size": 14,
|
16 |
+
"transformers_version": "4.18.0"
|
17 |
+
}
|
resources/examples/charger-hw.jpg
ADDED
![]() |
Git LFS Details
|
resources/examples/charger-ugreen.jpg
ADDED
![]() |
Git LFS Details
|
resources/examples/charger.jpg
ADDED
![]() |
Git LFS Details
|
resources/examples/jiandao.jpg
ADDED
![]() |
Git LFS Details
|
resources/examples/lego-yellow.jpg
ADDED
![]() |
Git LFS Details
|
starrynight.jpeg
ADDED
![]() |
Git LFS Details
|
transform/randaugment.py
ADDED
@@ -0,0 +1,340 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import cv2
|
2 |
+
import numpy as np
|
3 |
+
|
4 |
+
|
5 |
+
## aug functions
|
6 |
+
def identity_func(img):
|
7 |
+
return img
|
8 |
+
|
9 |
+
|
10 |
+
def autocontrast_func(img, cutoff=0):
|
11 |
+
'''
|
12 |
+
same output as PIL.ImageOps.autocontrast
|
13 |
+
'''
|
14 |
+
n_bins = 256
|
15 |
+
|
16 |
+
def tune_channel(ch):
|
17 |
+
n = ch.size
|
18 |
+
cut = cutoff * n // 100
|
19 |
+
if cut == 0:
|
20 |
+
high, low = ch.max(), ch.min()
|
21 |
+
else:
|
22 |
+
hist = cv2.calcHist([ch], [0], None, [n_bins], [0, n_bins])
|
23 |
+
low = np.argwhere(np.cumsum(hist) > cut)
|
24 |
+
low = 0 if low.shape[0] == 0 else low[0]
|
25 |
+
high = np.argwhere(np.cumsum(hist[::-1]) > cut)
|
26 |
+
high = n_bins - 1 if high.shape[0] == 0 else n_bins - 1 - high[0]
|
27 |
+
if high <= low:
|
28 |
+
table = np.arange(n_bins)
|
29 |
+
else:
|
30 |
+
scale = (n_bins - 1) / (high - low)
|
31 |
+
offset = -low * scale
|
32 |
+
table = np.arange(n_bins) * scale + offset
|
33 |
+
table[table < 0] = 0
|
34 |
+
table[table > n_bins - 1] = n_bins - 1
|
35 |
+
table = table.clip(0, 255).astype(np.uint8)
|
36 |
+
return table[ch]
|
37 |
+
|
38 |
+
channels = [tune_channel(ch) for ch in cv2.split(img)]
|
39 |
+
out = cv2.merge(channels)
|
40 |
+
return out
|
41 |
+
|
42 |
+
|
43 |
+
def equalize_func(img):
|
44 |
+
'''
|
45 |
+
same output as PIL.ImageOps.equalize
|
46 |
+
PIL's implementation is different from cv2.equalize
|
47 |
+
'''
|
48 |
+
n_bins = 256
|
49 |
+
|
50 |
+
def tune_channel(ch):
|
51 |
+
hist = cv2.calcHist([ch], [0], None, [n_bins], [0, n_bins])
|
52 |
+
non_zero_hist = hist[hist != 0].reshape(-1)
|
53 |
+
step = np.sum(non_zero_hist[:-1]) // (n_bins - 1)
|
54 |
+
if step == 0: return ch
|
55 |
+
n = np.empty_like(hist)
|
56 |
+
n[0] = step // 2
|
57 |
+
n[1:] = hist[:-1]
|
58 |
+
table = (np.cumsum(n) // step).clip(0, 255).astype(np.uint8)
|
59 |
+
return table[ch]
|
60 |
+
|
61 |
+
channels = [tune_channel(ch) for ch in cv2.split(img)]
|
62 |
+
out = cv2.merge(channels)
|
63 |
+
return out
|
64 |
+
|
65 |
+
|
66 |
+
def rotate_func(img, degree, fill=(0, 0, 0)):
|
67 |
+
'''
|
68 |
+
like PIL, rotate by degree, not radians
|
69 |
+
'''
|
70 |
+
H, W = img.shape[0], img.shape[1]
|
71 |
+
center = W / 2, H / 2
|
72 |
+
M = cv2.getRotationMatrix2D(center, degree, 1)
|
73 |
+
out = cv2.warpAffine(img, M, (W, H), borderValue=fill)
|
74 |
+
return out
|
75 |
+
|
76 |
+
|
77 |
+
def solarize_func(img, thresh=128):
|
78 |
+
'''
|
79 |
+
same output as PIL.ImageOps.posterize
|
80 |
+
'''
|
81 |
+
table = np.array([el if el < thresh else 255 - el for el in range(256)])
|
82 |
+
table = table.clip(0, 255).astype(np.uint8)
|
83 |
+
out = table[img]
|
84 |
+
return out
|
85 |
+
|
86 |
+
|
87 |
+
def color_func(img, factor):
|
88 |
+
'''
|
89 |
+
same output as PIL.ImageEnhance.Color
|
90 |
+
'''
|
91 |
+
## implementation according to PIL definition, quite slow
|
92 |
+
# degenerate = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)[:, :, np.newaxis]
|
93 |
+
# out = blend(degenerate, img, factor)
|
94 |
+
# M = (
|
95 |
+
# np.eye(3) * factor
|
96 |
+
# + np.float32([0.114, 0.587, 0.299]).reshape(3, 1) * (1. - factor)
|
97 |
+
# )[np.newaxis, np.newaxis, :]
|
98 |
+
M = (
|
99 |
+
np.float32([
|
100 |
+
[0.886, -0.114, -0.114],
|
101 |
+
[-0.587, 0.413, -0.587],
|
102 |
+
[-0.299, -0.299, 0.701]]) * factor
|
103 |
+
+ np.float32([[0.114], [0.587], [0.299]])
|
104 |
+
)
|
105 |
+
out = np.matmul(img, M).clip(0, 255).astype(np.uint8)
|
106 |
+
return out
|
107 |
+
|
108 |
+
|
109 |
+
def contrast_func(img, factor):
|
110 |
+
"""
|
111 |
+
same output as PIL.ImageEnhance.Contrast
|
112 |
+
"""
|
113 |
+
mean = np.sum(np.mean(img, axis=(0, 1)) * np.array([0.114, 0.587, 0.299]))
|
114 |
+
table = np.array([(
|
115 |
+
el - mean) * factor + mean
|
116 |
+
for el in range(256)
|
117 |
+
]).clip(0, 255).astype(np.uint8)
|
118 |
+
out = table[img]
|
119 |
+
return out
|
120 |
+
|
121 |
+
|
122 |
+
def brightness_func(img, factor):
|
123 |
+
'''
|
124 |
+
same output as PIL.ImageEnhance.Contrast
|
125 |
+
'''
|
126 |
+
table = (np.arange(256, dtype=np.float32) * factor).clip(0, 255).astype(np.uint8)
|
127 |
+
out = table[img]
|
128 |
+
return out
|
129 |
+
|
130 |
+
|
131 |
+
def sharpness_func(img, factor):
|
132 |
+
'''
|
133 |
+
The differences the this result and PIL are all on the 4 boundaries, the center
|
134 |
+
areas are same
|
135 |
+
'''
|
136 |
+
kernel = np.ones((3, 3), dtype=np.float32)
|
137 |
+
kernel[1][1] = 5
|
138 |
+
kernel /= 13
|
139 |
+
degenerate = cv2.filter2D(img, -1, kernel)
|
140 |
+
if factor == 0.0:
|
141 |
+
out = degenerate
|
142 |
+
elif factor == 1.0:
|
143 |
+
out = img
|
144 |
+
else:
|
145 |
+
out = img.astype(np.float32)
|
146 |
+
degenerate = degenerate.astype(np.float32)[1:-1, 1:-1, :]
|
147 |
+
out[1:-1, 1:-1, :] = degenerate + factor * (out[1:-1, 1:-1, :] - degenerate)
|
148 |
+
out = out.astype(np.uint8)
|
149 |
+
return out
|
150 |
+
|
151 |
+
|
152 |
+
def shear_x_func(img, factor, fill=(0, 0, 0)):
|
153 |
+
H, W = img.shape[0], img.shape[1]
|
154 |
+
M = np.float32([[1, factor, 0], [0, 1, 0]])
|
155 |
+
out = cv2.warpAffine(img, M, (W, H), borderValue=fill, flags=cv2.INTER_LINEAR).astype(np.uint8)
|
156 |
+
return out
|
157 |
+
|
158 |
+
|
159 |
+
def translate_x_func(img, offset, fill=(0, 0, 0)):
|
160 |
+
'''
|
161 |
+
same output as PIL.Image.transform
|
162 |
+
'''
|
163 |
+
H, W = img.shape[0], img.shape[1]
|
164 |
+
M = np.float32([[1, 0, -offset], [0, 1, 0]])
|
165 |
+
out = cv2.warpAffine(img, M, (W, H), borderValue=fill, flags=cv2.INTER_LINEAR).astype(np.uint8)
|
166 |
+
return out
|
167 |
+
|
168 |
+
|
169 |
+
def translate_y_func(img, offset, fill=(0, 0, 0)):
|
170 |
+
'''
|
171 |
+
same output as PIL.Image.transform
|
172 |
+
'''
|
173 |
+
H, W = img.shape[0], img.shape[1]
|
174 |
+
M = np.float32([[1, 0, 0], [0, 1, -offset]])
|
175 |
+
out = cv2.warpAffine(img, M, (W, H), borderValue=fill, flags=cv2.INTER_LINEAR).astype(np.uint8)
|
176 |
+
return out
|
177 |
+
|
178 |
+
|
179 |
+
def posterize_func(img, bits):
|
180 |
+
'''
|
181 |
+
same output as PIL.ImageOps.posterize
|
182 |
+
'''
|
183 |
+
out = np.bitwise_and(img, np.uint8(255 << (8 - bits)))
|
184 |
+
return out
|
185 |
+
|
186 |
+
|
187 |
+
def shear_y_func(img, factor, fill=(0, 0, 0)):
|
188 |
+
H, W = img.shape[0], img.shape[1]
|
189 |
+
M = np.float32([[1, 0, 0], [factor, 1, 0]])
|
190 |
+
out = cv2.warpAffine(img, M, (W, H), borderValue=fill, flags=cv2.INTER_LINEAR).astype(np.uint8)
|
191 |
+
return out
|
192 |
+
|
193 |
+
|
194 |
+
def cutout_func(img, pad_size, replace=(0, 0, 0)):
|
195 |
+
replace = np.array(replace, dtype=np.uint8)
|
196 |
+
H, W = img.shape[0], img.shape[1]
|
197 |
+
rh, rw = np.random.random(2)
|
198 |
+
pad_size = pad_size // 2
|
199 |
+
ch, cw = int(rh * H), int(rw * W)
|
200 |
+
x1, x2 = max(ch - pad_size, 0), min(ch + pad_size, H)
|
201 |
+
y1, y2 = max(cw - pad_size, 0), min(cw + pad_size, W)
|
202 |
+
out = img.copy()
|
203 |
+
out[x1:x2, y1:y2, :] = replace
|
204 |
+
return out
|
205 |
+
|
206 |
+
|
207 |
+
### level to args
|
208 |
+
def enhance_level_to_args(MAX_LEVEL):
|
209 |
+
def level_to_args(level):
|
210 |
+
return ((level / MAX_LEVEL) * 1.8 + 0.1,)
|
211 |
+
return level_to_args
|
212 |
+
|
213 |
+
|
214 |
+
def shear_level_to_args(MAX_LEVEL, replace_value):
|
215 |
+
def level_to_args(level):
|
216 |
+
level = (level / MAX_LEVEL) * 0.3
|
217 |
+
if np.random.random() > 0.5: level = -level
|
218 |
+
return (level, replace_value)
|
219 |
+
|
220 |
+
return level_to_args
|
221 |
+
|
222 |
+
|
223 |
+
def translate_level_to_args(translate_const, MAX_LEVEL, replace_value):
|
224 |
+
def level_to_args(level):
|
225 |
+
level = (level / MAX_LEVEL) * float(translate_const)
|
226 |
+
if np.random.random() > 0.5: level = -level
|
227 |
+
return (level, replace_value)
|
228 |
+
|
229 |
+
return level_to_args
|
230 |
+
|
231 |
+
|
232 |
+
def cutout_level_to_args(cutout_const, MAX_LEVEL, replace_value):
|
233 |
+
def level_to_args(level):
|
234 |
+
level = int((level / MAX_LEVEL) * cutout_const)
|
235 |
+
return (level, replace_value)
|
236 |
+
|
237 |
+
return level_to_args
|
238 |
+
|
239 |
+
|
240 |
+
def solarize_level_to_args(MAX_LEVEL):
|
241 |
+
def level_to_args(level):
|
242 |
+
level = int((level / MAX_LEVEL) * 256)
|
243 |
+
return (level, )
|
244 |
+
return level_to_args
|
245 |
+
|
246 |
+
|
247 |
+
def none_level_to_args(level):
|
248 |
+
return ()
|
249 |
+
|
250 |
+
|
251 |
+
def posterize_level_to_args(MAX_LEVEL):
|
252 |
+
def level_to_args(level):
|
253 |
+
level = int((level / MAX_LEVEL) * 4)
|
254 |
+
return (level, )
|
255 |
+
return level_to_args
|
256 |
+
|
257 |
+
|
258 |
+
def rotate_level_to_args(MAX_LEVEL, replace_value):
|
259 |
+
def level_to_args(level):
|
260 |
+
level = (level / MAX_LEVEL) * 30
|
261 |
+
if np.random.random() < 0.5:
|
262 |
+
level = -level
|
263 |
+
return (level, replace_value)
|
264 |
+
|
265 |
+
return level_to_args
|
266 |
+
|
267 |
+
|
268 |
+
func_dict = {
|
269 |
+
'Identity': identity_func,
|
270 |
+
'AutoContrast': autocontrast_func,
|
271 |
+
'Equalize': equalize_func,
|
272 |
+
'Rotate': rotate_func,
|
273 |
+
'Solarize': solarize_func,
|
274 |
+
'Color': color_func,
|
275 |
+
'Contrast': contrast_func,
|
276 |
+
'Brightness': brightness_func,
|
277 |
+
'Sharpness': sharpness_func,
|
278 |
+
'ShearX': shear_x_func,
|
279 |
+
'TranslateX': translate_x_func,
|
280 |
+
'TranslateY': translate_y_func,
|
281 |
+
'Posterize': posterize_func,
|
282 |
+
'ShearY': shear_y_func,
|
283 |
+
}
|
284 |
+
|
285 |
+
translate_const = 10
|
286 |
+
MAX_LEVEL = 10
|
287 |
+
replace_value = (128, 128, 128)
|
288 |
+
arg_dict = {
|
289 |
+
'Identity': none_level_to_args,
|
290 |
+
'AutoContrast': none_level_to_args,
|
291 |
+
'Equalize': none_level_to_args,
|
292 |
+
'Rotate': rotate_level_to_args(MAX_LEVEL, replace_value),
|
293 |
+
'Solarize': solarize_level_to_args(MAX_LEVEL),
|
294 |
+
'Color': enhance_level_to_args(MAX_LEVEL),
|
295 |
+
'Contrast': enhance_level_to_args(MAX_LEVEL),
|
296 |
+
'Brightness': enhance_level_to_args(MAX_LEVEL),
|
297 |
+
'Sharpness': enhance_level_to_args(MAX_LEVEL),
|
298 |
+
'ShearX': shear_level_to_args(MAX_LEVEL, replace_value),
|
299 |
+
'TranslateX': translate_level_to_args(
|
300 |
+
translate_const, MAX_LEVEL, replace_value
|
301 |
+
),
|
302 |
+
'TranslateY': translate_level_to_args(
|
303 |
+
translate_const, MAX_LEVEL, replace_value
|
304 |
+
),
|
305 |
+
'Posterize': posterize_level_to_args(MAX_LEVEL),
|
306 |
+
'ShearY': shear_level_to_args(MAX_LEVEL, replace_value),
|
307 |
+
}
|
308 |
+
|
309 |
+
|
310 |
+
class RandomAugment(object):
|
311 |
+
|
312 |
+
def __init__(self, N=2, M=10, isPIL=False, augs=[]):
|
313 |
+
self.N = N
|
314 |
+
self.M = M
|
315 |
+
self.isPIL = isPIL
|
316 |
+
if augs:
|
317 |
+
self.augs = augs
|
318 |
+
else:
|
319 |
+
self.augs = list(arg_dict.keys())
|
320 |
+
|
321 |
+
def get_random_ops(self):
|
322 |
+
sampled_ops = np.random.choice(self.augs, self.N)
|
323 |
+
return [(op, 0.5, self.M) for op in sampled_ops]
|
324 |
+
|
325 |
+
def __call__(self, img):
|
326 |
+
if self.isPIL:
|
327 |
+
img = np.array(img)
|
328 |
+
ops = self.get_random_ops()
|
329 |
+
for name, prob, level in ops:
|
330 |
+
if np.random.random() > prob:
|
331 |
+
continue
|
332 |
+
args = arg_dict[name](level)
|
333 |
+
img = func_dict[name](img, *args)
|
334 |
+
return img
|
335 |
+
|
336 |
+
|
337 |
+
if __name__ == '__main__':
|
338 |
+
a = RandomAugment()
|
339 |
+
img = np.random.randn(32, 32, 3)
|
340 |
+
a(img)
|
utils.py
ADDED
@@ -0,0 +1,278 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import math
|
2 |
+
def cosine_lr_schedule(optimizer, epoch, max_epoch, init_lr, min_lr):
|
3 |
+
"""Decay the learning rate"""
|
4 |
+
lr = (init_lr - min_lr) * 0.5 * (1. + math.cos(math.pi * epoch / max_epoch)) + min_lr
|
5 |
+
for param_group in optimizer.param_groups:
|
6 |
+
param_group['lr'] = lr
|
7 |
+
|
8 |
+
def warmup_lr_schedule(optimizer, step, max_step, init_lr, max_lr):
|
9 |
+
"""Warmup the learning rate"""
|
10 |
+
lr = min(max_lr, init_lr + (max_lr - init_lr) * step / max_step)
|
11 |
+
for param_group in optimizer.param_groups:
|
12 |
+
param_group['lr'] = lr
|
13 |
+
|
14 |
+
def step_lr_schedule(optimizer, epoch, init_lr, min_lr, decay_rate):
|
15 |
+
"""Decay the learning rate"""
|
16 |
+
lr = max(min_lr, init_lr * (decay_rate**epoch))
|
17 |
+
for param_group in optimizer.param_groups:
|
18 |
+
param_group['lr'] = lr
|
19 |
+
|
20 |
+
import numpy as np
|
21 |
+
import io
|
22 |
+
import os
|
23 |
+
import time
|
24 |
+
from collections import defaultdict, deque
|
25 |
+
import datetime
|
26 |
+
|
27 |
+
import torch
|
28 |
+
import torch.distributed as dist
|
29 |
+
|
30 |
+
class SmoothedValue(object):
|
31 |
+
"""Track a series of values and provide access to smoothed values over a
|
32 |
+
window or the global series average.
|
33 |
+
"""
|
34 |
+
|
35 |
+
def __init__(self, window_size=20, fmt=None):
|
36 |
+
if fmt is None:
|
37 |
+
fmt = "{median:.4f} ({global_avg:.4f})"
|
38 |
+
self.deque = deque(maxlen=window_size)
|
39 |
+
self.total = 0.0
|
40 |
+
self.count = 0
|
41 |
+
self.fmt = fmt
|
42 |
+
|
43 |
+
def update(self, value, n=1):
|
44 |
+
self.deque.append(value)
|
45 |
+
self.count += n
|
46 |
+
self.total += value * n
|
47 |
+
|
48 |
+
def synchronize_between_processes(self):
|
49 |
+
"""
|
50 |
+
Warning: does not synchronize the deque!
|
51 |
+
"""
|
52 |
+
if not is_dist_avail_and_initialized():
|
53 |
+
return
|
54 |
+
t = torch.tensor([self.count, self.total], dtype=torch.float64, device='cuda')
|
55 |
+
dist.barrier()
|
56 |
+
dist.all_reduce(t)
|
57 |
+
t = t.tolist()
|
58 |
+
self.count = int(t[0])
|
59 |
+
self.total = t[1]
|
60 |
+
|
61 |
+
@property
|
62 |
+
def median(self):
|
63 |
+
d = torch.tensor(list(self.deque))
|
64 |
+
return d.median().item()
|
65 |
+
|
66 |
+
@property
|
67 |
+
def avg(self):
|
68 |
+
d = torch.tensor(list(self.deque), dtype=torch.float32)
|
69 |
+
return d.mean().item()
|
70 |
+
|
71 |
+
@property
|
72 |
+
def global_avg(self):
|
73 |
+
return self.total / self.count
|
74 |
+
|
75 |
+
@property
|
76 |
+
def max(self):
|
77 |
+
return max(self.deque)
|
78 |
+
|
79 |
+
@property
|
80 |
+
def value(self):
|
81 |
+
return self.deque[-1]
|
82 |
+
|
83 |
+
def __str__(self):
|
84 |
+
return self.fmt.format(
|
85 |
+
median=self.median,
|
86 |
+
avg=self.avg,
|
87 |
+
global_avg=self.global_avg,
|
88 |
+
max=self.max,
|
89 |
+
value=self.value)
|
90 |
+
|
91 |
+
|
92 |
+
class MetricLogger(object):
|
93 |
+
def __init__(self, delimiter="\t"):
|
94 |
+
self.meters = defaultdict(SmoothedValue)
|
95 |
+
self.delimiter = delimiter
|
96 |
+
|
97 |
+
def update(self, **kwargs):
|
98 |
+
for k, v in kwargs.items():
|
99 |
+
if isinstance(v, torch.Tensor):
|
100 |
+
v = v.item()
|
101 |
+
assert isinstance(v, (float, int))
|
102 |
+
self.meters[k].update(v)
|
103 |
+
|
104 |
+
def __getattr__(self, attr):
|
105 |
+
if attr in self.meters:
|
106 |
+
return self.meters[attr]
|
107 |
+
if attr in self.__dict__:
|
108 |
+
return self.__dict__[attr]
|
109 |
+
raise AttributeError("'{}' object has no attribute '{}'".format(
|
110 |
+
type(self).__name__, attr))
|
111 |
+
|
112 |
+
def __str__(self):
|
113 |
+
loss_str = []
|
114 |
+
for name, meter in self.meters.items():
|
115 |
+
loss_str.append(
|
116 |
+
"{}: {}".format(name, str(meter))
|
117 |
+
)
|
118 |
+
return self.delimiter.join(loss_str)
|
119 |
+
|
120 |
+
def global_avg(self):
|
121 |
+
loss_str = []
|
122 |
+
for name, meter in self.meters.items():
|
123 |
+
loss_str.append(
|
124 |
+
"{}: {:.4f}".format(name, meter.global_avg)
|
125 |
+
)
|
126 |
+
return self.delimiter.join(loss_str)
|
127 |
+
|
128 |
+
def synchronize_between_processes(self):
|
129 |
+
for meter in self.meters.values():
|
130 |
+
meter.synchronize_between_processes()
|
131 |
+
|
132 |
+
def add_meter(self, name, meter):
|
133 |
+
self.meters[name] = meter
|
134 |
+
|
135 |
+
def log_every(self, iterable, print_freq, header=None):
|
136 |
+
i = 0
|
137 |
+
if not header:
|
138 |
+
header = ''
|
139 |
+
start_time = time.time()
|
140 |
+
end = time.time()
|
141 |
+
iter_time = SmoothedValue(fmt='{avg:.4f}')
|
142 |
+
data_time = SmoothedValue(fmt='{avg:.4f}')
|
143 |
+
space_fmt = ':' + str(len(str(len(iterable)))) + 'd'
|
144 |
+
log_msg = [
|
145 |
+
header,
|
146 |
+
'[{0' + space_fmt + '}/{1}]',
|
147 |
+
'eta: {eta}',
|
148 |
+
'{meters}',
|
149 |
+
'time: {time}',
|
150 |
+
'data: {data}'
|
151 |
+
]
|
152 |
+
if torch.cuda.is_available():
|
153 |
+
log_msg.append('max mem: {memory:.0f}')
|
154 |
+
log_msg = self.delimiter.join(log_msg)
|
155 |
+
MB = 1024.0 * 1024.0
|
156 |
+
for obj in iterable:
|
157 |
+
data_time.update(time.time() - end)
|
158 |
+
yield obj
|
159 |
+
iter_time.update(time.time() - end)
|
160 |
+
if i % print_freq == 0 or i == len(iterable) - 1:
|
161 |
+
eta_seconds = iter_time.global_avg * (len(iterable) - i)
|
162 |
+
eta_string = str(datetime.timedelta(seconds=int(eta_seconds)))
|
163 |
+
if torch.cuda.is_available():
|
164 |
+
print(log_msg.format(
|
165 |
+
i, len(iterable), eta=eta_string,
|
166 |
+
meters=str(self),
|
167 |
+
time=str(iter_time), data=str(data_time),
|
168 |
+
memory=torch.cuda.max_memory_allocated() / MB))
|
169 |
+
else:
|
170 |
+
print(log_msg.format(
|
171 |
+
i, len(iterable), eta=eta_string,
|
172 |
+
meters=str(self),
|
173 |
+
time=str(iter_time), data=str(data_time)))
|
174 |
+
i += 1
|
175 |
+
end = time.time()
|
176 |
+
total_time = time.time() - start_time
|
177 |
+
total_time_str = str(datetime.timedelta(seconds=int(total_time)))
|
178 |
+
print('{} Total time: {} ({:.4f} s / it)'.format(
|
179 |
+
header, total_time_str, total_time / len(iterable)))
|
180 |
+
|
181 |
+
|
182 |
+
class AttrDict(dict):
|
183 |
+
def __init__(self, *args, **kwargs):
|
184 |
+
super(AttrDict, self).__init__(*args, **kwargs)
|
185 |
+
self.__dict__ = self
|
186 |
+
|
187 |
+
|
188 |
+
def compute_acc(logits, label, reduction='mean'):
|
189 |
+
ret = (torch.argmax(logits, dim=1) == label).float()
|
190 |
+
if reduction == 'none':
|
191 |
+
return ret.detach()
|
192 |
+
elif reduction == 'mean':
|
193 |
+
return ret.mean().item()
|
194 |
+
|
195 |
+
def compute_n_params(model, return_str=True):
|
196 |
+
tot = 0
|
197 |
+
for p in model.parameters():
|
198 |
+
w = 1
|
199 |
+
for x in p.shape:
|
200 |
+
w *= x
|
201 |
+
tot += w
|
202 |
+
if return_str:
|
203 |
+
if tot >= 1e6:
|
204 |
+
return '{:.1f}M'.format(tot / 1e6)
|
205 |
+
else:
|
206 |
+
return '{:.1f}K'.format(tot / 1e3)
|
207 |
+
else:
|
208 |
+
return tot
|
209 |
+
|
210 |
+
def setup_for_distributed(is_master):
|
211 |
+
"""
|
212 |
+
This function disables printing when not in master process
|
213 |
+
"""
|
214 |
+
import builtins as __builtin__
|
215 |
+
builtin_print = __builtin__.print
|
216 |
+
|
217 |
+
def print(*args, **kwargs):
|
218 |
+
force = kwargs.pop('force', False)
|
219 |
+
if is_master or force:
|
220 |
+
builtin_print(*args, **kwargs)
|
221 |
+
|
222 |
+
__builtin__.print = print
|
223 |
+
|
224 |
+
|
225 |
+
def is_dist_avail_and_initialized():
|
226 |
+
if not dist.is_available():
|
227 |
+
return False
|
228 |
+
if not dist.is_initialized():
|
229 |
+
return False
|
230 |
+
return True
|
231 |
+
|
232 |
+
|
233 |
+
def get_world_size():
|
234 |
+
if not is_dist_avail_and_initialized():
|
235 |
+
return 1
|
236 |
+
return dist.get_world_size()
|
237 |
+
|
238 |
+
|
239 |
+
def get_rank():
|
240 |
+
if not is_dist_avail_and_initialized():
|
241 |
+
return 0
|
242 |
+
return dist.get_rank()
|
243 |
+
|
244 |
+
|
245 |
+
def is_main_process():
|
246 |
+
return get_rank() == 0
|
247 |
+
|
248 |
+
|
249 |
+
def save_on_master(*args, **kwargs):
|
250 |
+
if is_main_process():
|
251 |
+
torch.save(*args, **kwargs)
|
252 |
+
|
253 |
+
|
254 |
+
def init_distributed_mode(args):
|
255 |
+
if 'RANK' in os.environ and 'WORLD_SIZE' in os.environ:
|
256 |
+
args.rank = int(os.environ["RANK"])
|
257 |
+
args.world_size = int(os.environ['WORLD_SIZE'])
|
258 |
+
args.gpu = int(os.environ['LOCAL_RANK'])
|
259 |
+
elif 'SLURM_PROCID' in os.environ:
|
260 |
+
args.rank = int(os.environ['SLURM_PROCID'])
|
261 |
+
args.gpu = args.rank % torch.cuda.device_count()
|
262 |
+
else:
|
263 |
+
print('Not using distributed mode')
|
264 |
+
args.distributed = False
|
265 |
+
return
|
266 |
+
|
267 |
+
args.distributed = True
|
268 |
+
|
269 |
+
torch.cuda.set_device(args.gpu)
|
270 |
+
args.dist_backend = 'nccl'
|
271 |
+
print('| distributed init (rank {}, word {}): {}'.format(
|
272 |
+
args.rank, args.world_size, args.dist_url), flush=True)
|
273 |
+
torch.distributed.init_process_group(backend=args.dist_backend, init_method=args.dist_url,
|
274 |
+
world_size=args.world_size, rank=args.rank)
|
275 |
+
torch.distributed.barrier()
|
276 |
+
setup_for_distributed(args.rank == 0)
|
277 |
+
|
278 |
+
|