Upload mosaicml/mpt-7b-instruct ctranslate fp16 weights

README.md

@@ -0,0 +1,229 @@
+---
+license: cc-by-sa-3.0
+datasets:
+- mosaicml/dolly_hhrlhf
+tags:
+- ctranslate2
+- int8
+- float16
+- Composer
+- MosaicML
+- llm-foundry
+inference: false
+---
+# # Fast-Inference with Ctranslate2
+Speedup inference while reducing memory by 2x-4x using int8 inference in C++ on CPU or GPU.
+
+quantized version of [mosaicml/mpt-7b-instruct](https://huggingface.co/mosaicml/mpt-7b-instruct)
+```bash
+pip install hf-hub-ctranslate2>=2.0.8 
+```
+Converted on 2023-05-30 using
+```
+ct2-transformers-converter --model mosaicml/mpt-7b-instruct --output_dir /home/michael/tmp-ct2fast-mpt-7b-instruct --force --copy_files configuration_mpt.py meta_init_context.py tokenizer.json hf_prefixlm_converter.py README.md tokenizer_config.json blocks.py adapt_tokenizer.py attention.py norm.py generation_config.json flash_attn_triton.py special_tokens_map.json param_init_fns.py .gitattributes --quantization float16 --trust_remote_code
+```
+
+Checkpoint compatible to [ctranslate2>=3.13.0](https://github.com/OpenNMT/CTranslate2) and [hf-hub-ctranslate2>=2.0.6](https://github.com/michaelfeil/hf-hub-ctranslate2)
+- `compute_type=int8_float16` for `device="cuda"` 
+- `compute_type=int8`  for `device="cpu"`
+
+```python
+from hf_hub_ctranslate2 import TranslatorCT2fromHfHub, GeneratorCT2fromHfHub
+from transformers import AutoTokenizer
+
+model_name = "michaelfeil/ct2fast-mpt-7b-instruct"
+# use either TranslatorCT2fromHfHub or GeneratorCT2fromHfHub here, depending on model.
+model = GeneratorCT2fromHfHub(
+        # load in int8 on CUDA
+        model_name_or_path=model_name, 
+        device="cuda",
+        compute_type="int8_float16",
+        # tokenizer=AutoTokenizer.from_pretrained("mosaicml/mpt-7b-instruct")
+)
+outputs = model.generate(
+    text=["How do you call a fast Flan-ingo?", "User: How are you doing? Bot:"],
+    max_length=64
+)
+print(outputs)
+```
+
+# Licence and other remarks:
+This is just a quantized version. Licence conditions are intended to be idential to original huggingface repo.
+
+# Original description
+    
+
+# MPT-7B-Instruct
+
+MPT-7B-Instruct is a model for short-form instruction following.
+It is built by finetuning [MPT-7B](https://huggingface.co/spaces/mosaicml/mpt-7b) on a [dataset](https://huggingface.co/datasets/sam-mosaic/dolly_hhrlhf) derived from the [Databricks Dolly-15k](https://huggingface.co/datasets/databricks/databricks-dolly-15k) and the [Anthropic Helpful and Harmless (HH-RLHF)](https://huggingface.co/datasets/Anthropic/hh-rlhf) datasets. 
+  * License: _CC-By-SA-3.0_
+  * [Demo on Hugging Face Spaces](https://huggingface.co/spaces/mosaicml/mpt-7b-instruct)
+
+
+This model was trained by [MosaicML](https://www.mosaicml.com) and follows a modified decoder-only transformer architecture.
+
+## Model Date
+
+May 5, 2023
+
+## Model License
+
+CC-By-SA-3.0
+
+## Documentation
+
+* [Blog post: Introducing MPT-7B: A New Standard for Open-Source, Commercially Usable LLMs](https://www.mosaicml.com/blog/mpt-7b)
+* [Codebase (mosaicml/llm-foundry repo)](https://github.com/mosaicml/llm-foundry/)
+* Questions: Feel free to contact us via the [MosaicML Community Slack](https://mosaicml.me/slack)!
+
+### Example Question/Instruction
+
+**Longboi24**:
+> What is a quoll?
+
+**MPT-7B-Instruct**:
+
+>A Quoll (pronounced “cool”) is one of Australia’s native carnivorous marsupial mammals, which are also known as macropods or wallabies in other parts around Asia and South America
+
+## How to Use
+
+Note: This model requires that `trust_remote_code=True` be passed to the `from_pretrained` method. This is because we use a custom model architecture that is not yet part of the `transformers` package.
+
+It includes options for many training efficiency features such as [FlashAttention (Dao et al. 2022)](https://arxiv.org/pdf/2205.14135.pdf), [ALiBi](https://arxiv.org/abs/2108.12409), QK LayerNorm, and more.
+
+```python
+import transformers
+model = transformers.AutoModelForCausalLM.from_pretrained(
+  'mosaicml/mpt-7b-instruct',
+  trust_remote_code=True
+)
+```
+Note: This model requires that `trust_remote_code=True` be passed to the `from_pretrained` method. 
+This is because we use a custom `MPT` model architecture that is not yet part of the Hugging Face `transformers` package.
+`MPT` includes options for many training efficiency features such as [FlashAttention](https://arxiv.org/pdf/2205.14135.pdf), [ALiBi](https://arxiv.org/abs/2108.12409), [QK LayerNorm](https://arxiv.org/abs/2010.04245), and more.
+
+To use the optimized [triton implementation](https://github.com/openai/triton) of FlashAttention, you can load the model with `attn_impl='triton'` and move the model to `bfloat16`:
+```python
+config = transformers.AutoConfig.from_pretrained(
+  'mosaicml/mpt-7b-instruct',
+  trust_remote_code=True
+)
+config.attn_config['attn_impl'] = 'triton'
+
+model = transformers.AutoModelForCausalLM.from_pretrained(
+  'mosaicml/mpt-7b-instruct',
+  config=config,
+  torch_dtype=torch.bfloat16,
+  trust_remote_code=True
+)
+model.to(device='cuda:0')
+```
+
+Although the model was trained with a sequence length of 2048, ALiBi enables users to increase the maximum sequence length during finetuning and/or inference. For example:
+
+```python
+config = transformers.AutoConfig.from_pretrained(
+  'mosaicml/mpt-7b-instruct',
+  trust_remote_code=True
+)
+config.update({"max_seq_len": 4096})
+model = transformers.AutoModelForCausalLM.from_pretrained(
+  'mosaicml/mpt-7b-instruct',
+  config=config,
+  trust_remote_code=True
+)
+```
+
+This model was trained with the [EleutherAI/gpt-neox-20b](https://huggingface.co/EleutherAI/gpt-neox-20b) tokenizer.
+
+```python
+from transformers import AutoTokenizer
+tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neox-20b")
+```
+
+### Formatting
+
+This model was trained on data formatted in the dolly-15k format:
+
+```python
+INSTRUCTION_KEY = "### Instruction:"
+RESPONSE_KEY = "### Response:"
+INTRO_BLURB = "Below is an instruction that describes a task. Write a response that appropriately completes the request."
+PROMPT_FOR_GENERATION_FORMAT = """{intro}
+{instruction_key}
+{instruction}
+{response_key}
+""".format(
+    intro=INTRO_BLURB,
+    instruction_key=INSTRUCTION_KEY,
+    instruction="{instruction}",
+    response_key=RESPONSE_KEY,
+)
+
+example = "James decides to run 3 sprints 3 times a week. He runs 60 meters each sprint. How many total meters does he run a week? Explain before answering."
+fmt_ex = PROMPT_FOR_GENERATION_FORMAT.format(instruction=example)
+```
+
+In the above example, `fmt_ex` is ready to be tokenized and sent through the model.
+
+## Model Description
+
+The architecture is a modification of a standard decoder-only transformer.
+
+The model has been modified from a standard transformer in the following ways:
+* It uses [FlashAttention](https://arxiv.org/pdf/2205.14135.pdf)
+* It uses [ALiBi (Attention with Linear Biases)](https://arxiv.org/abs/2108.12409) and does not use positional embeddings
+* It does not use biases
+
+
+| Hyperparameter | Value |
+|----------------|-------|
+|n_parameters | 6.7B |
+|n_layers | 32 |
+| n_heads | 32 |
+| d_model | 4096 |
+| vocab size | 50432 |
+| sequence length | 2048 |
+
+## PreTraining Data
+
+For more details on the pretraining process, see [MPT-7B](https://huggingface.co/mosaicml/mpt-7b).
+
+The data was tokenized using the [EleutherAI/gpt-neox-20b](https://huggingface.co/EleutherAI/gpt-neox-20b) tokenizer.
+
+## Limitations and Biases
+
+_The following language is modified from [EleutherAI's GPT-NeoX-20B](https://huggingface.co/EleutherAI/gpt-neox-20b)_
+
+MPT-7B-Instruct can produce factually incorrect output, and should not be relied on to produce factually accurate information.
+MPT-7B-Instruct was trained on various public datasets.
+While great efforts have been taken to clean the pretraining data, it is possible that this model could generate lewd, biased or otherwise offensive outputs.
+
+
+## Acknowledgements
+
+This model was finetuned by Sam Havens and the MosaicML NLP team
+
+## MosaicML Platform
+
+If you're interested in [training](https://www.mosaicml.com/training) and [deploying](https://www.mosaicml.com/inference) your own MPT or LLMs on the MosaicML Platform, [sign up here](https://forms.mosaicml.com/demo?utm_source=huggingface&utm_medium=referral&utm_campaign=mpt-7b).
+
+## Disclaimer
+
+The license on this model does not constitute legal advice. We are not responsible for the actions of third parties who use this model. Please cosult an attorney before using this model for commercial purposes.
+
+## Citation
+
+Please cite this model using the following format:
+
+```
+@online{MosaicML2023Introducing,
+    author    = {MosaicML NLP Team},
+    title     = {Introducing MPT-7B: A New Standard for Open-Source, Commercially Usable LLMs},
+    year      = {2023},
+    url       = {www.mosaicml.com/blog/mpt-7b},
+    note      = {Accessed: 2023-03-28}, % change this date
+    urldate   = {2023-03-28} % change this date
+}
+```

adapt_tokenizer.py

@@ -0,0 +1,41 @@
+from typing import Union
+from transformers import AutoTokenizer, PreTrainedTokenizer, PreTrainedTokenizerFast
+Tokenizer = Union[PreTrainedTokenizer, PreTrainedTokenizerFast]
+NUM_SENTINEL_TOKENS: int = 100
+
+def adapt_tokenizer_for_denoising(tokenizer: Tokenizer):
+    """Adds sentinel tokens and padding token (if missing).
+
+    Expands the tokenizer vocabulary to include sentinel tokens
+    used in mixture-of-denoiser tasks as well as a padding token.
+
+    All added tokens are added as special tokens. No tokens are
+    added if sentinel tokens and padding token already exist.
+    """
+    sentinels_to_add = [f'<extra_id_{i}>' for i in range(NUM_SENTINEL_TOKENS)]
+    tokenizer.add_tokens(sentinels_to_add, special_tokens=True)
+    if tokenizer.pad_token is None:
+        tokenizer.add_tokens('<pad>', special_tokens=True)
+        tokenizer.pad_token = '<pad>'
+        assert tokenizer.pad_token_id is not None
+    sentinels = ''.join([f'<extra_id_{i}>' for i in range(NUM_SENTINEL_TOKENS)])
+    _sentinel_token_ids = tokenizer(sentinels, add_special_tokens=False).input_ids
+    tokenizer.sentinel_token_ids = _sentinel_token_ids
+
+class AutoTokenizerForMOD(AutoTokenizer):
+    """AutoTokenizer + Adaptation for MOD.
+
+    A simple wrapper around AutoTokenizer to make instantiating
+    an MOD-adapted tokenizer a bit easier.
+
+    MOD-adapted tokenizers have sentinel tokens (e.g., <extra_id_0>),
+    a padding token, and a property to get the token ids of the
+    sentinel tokens.
+    """
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        """See `AutoTokenizer.from_pretrained` docstring."""
+        tokenizer = super().from_pretrained(*args, **kwargs)
+        adapt_tokenizer_for_denoising(tokenizer)
+        return tokenizer

attention.py

@@ -0,0 +1,285 @@
+"""Attention layers."""
+import math
+import warnings
+from typing import Optional
+import torch
+import torch.nn as nn
+from einops import rearrange
+from packaging import version
+from torch import nn
+from .norm import LPLayerNorm
+
+def _reset_is_causal(num_query_tokens: int, num_key_tokens: int, original_is_causal: bool):
+    if original_is_causal and num_query_tokens != num_key_tokens:
+        if num_query_tokens != 1:
+            raise NotImplementedError('MPT does not support query and key with different number of tokens, unless number of query tokens is 1.')
+        else:
+            return False
+    return original_is_causal
+
+def scaled_multihead_dot_product_attention(query, key, value, n_heads, softmax_scale=None, attn_bias=None, key_padding_mask=None, is_causal=False, dropout_p=0.0, training=False, needs_weights=False, multiquery=False):
+    q = rearrange(query, 'b s (h d) -> b h s d', h=n_heads)
+    k = rearrange(key, 'b s (h d) -> b h d s', h=1 if multiquery else n_heads)
+    v = rearrange(value, 'b s (h d) -> b h s d', h=1 if multiquery else n_heads)
+    min_val = torch.finfo(q.dtype).min
+    (b, _, s_q, d) = q.shape
+    s_k = k.size(-1)
+    if softmax_scale is None:
+        softmax_scale = 1 / math.sqrt(d)
+    attn_weight = q.matmul(k) * softmax_scale
+    if attn_bias is not None:
+        if attn_bias.size(-1) != 1 and attn_bias.size(-1) != s_k or (attn_bias.size(-2) != 1 and attn_bias.size(-2) != s_q):
+            raise RuntimeError(f'attn_bias (shape: {attn_bias.shape}) is expected to broadcast to shape: {attn_weight.shape}.')
+        attn_weight = attn_weight + attn_bias
+    if key_padding_mask is not None:
+        if attn_bias is not None:
+            warnings.warn('Propogating key_padding_mask to the attention module ' + 'and applying it within the attention module can cause ' + 'unneccessary computation/memory usage. Consider integrating ' + 'into attn_bias once and passing that to each attention ' + 'module instead.')
+        attn_weight = attn_weight.masked_fill(~key_padding_mask.view((b, 1, 1, s_k)), min_val)
+    if is_causal:
+        s = max(s_q, s_k)
+        causal_mask = attn_weight.new_ones(s, s, dtype=torch.float16)
+        causal_mask = causal_mask.tril()
+        causal_mask = causal_mask.to(torch.bool)
+        causal_mask = ~causal_mask
+        causal_mask = causal_mask[-s_q:, -s_k:]
+        attn_weight = attn_weight.masked_fill(causal_mask.view(1, 1, s_q, s_k), min_val)
+    attn_weight = torch.softmax(attn_weight, dim=-1)
+    if dropout_p:
+        attn_weight = torch.nn.functional.dropout(attn_weight, p=dropout_p, training=training, inplace=True)
+    out = attn_weight.matmul(v)
+    out = rearrange(out, 'b h s d -> b s (h d)')
+    if needs_weights:
+        return (out, attn_weight)
+    return (out, None)
+
+def check_valid_inputs(*tensors, valid_dtypes=[torch.float16, torch.bfloat16]):
+    for tensor in tensors:
+        if tensor.dtype not in valid_dtypes:
+            raise TypeError(f'tensor.dtype={tensor.dtype!r} must be in valid_dtypes={valid_dtypes!r}.')
+        if not tensor.is_cuda:
+            raise TypeError(f'Inputs must be cuda tensors (tensor.is_cuda={tensor.is_cuda!r}).')
+
+def flash_attn_fn(query, key, value, n_heads, softmax_scale=None, attn_bias=None, key_padding_mask=None, is_causal=False, dropout_p=0.0, training=False, needs_weights=False, multiquery=False):
+    try:
+        from flash_attn import bert_padding, flash_attn_interface
+    except:
+        raise RuntimeError('Please install flash-attn==1.0.3.post0')
+    check_valid_inputs(query, key, value)
+    if attn_bias is not None:
+        raise NotImplementedError(f'attn_bias not implemented for flash attn.')
+    (batch_size, seqlen) = query.shape[:2]
+    if key_padding_mask is None:
+        key_padding_mask = torch.ones_like(key[:, :, 0], dtype=torch.bool)
+    query_padding_mask = key_padding_mask[:, -query.size(1):]
+    (query_unpad, indices_q, cu_seqlens_q, max_seqlen_q) = bert_padding.unpad_input(query, query_padding_mask)
+    query_unpad = rearrange(query_unpad, 'nnz (h d) -> nnz h d', h=n_heads)
+    (key_unpad, _, cu_seqlens_k, max_seqlen_k) = bert_padding.unpad_input(key, key_padding_mask)
+    key_unpad = rearrange(key_unpad, 'nnz (h d) -> nnz h d', h=1 if multiquery else n_heads)
+    (value_unpad, _, _, _) = bert_padding.unpad_input(value, key_padding_mask)
+    value_unpad = rearrange(value_unpad, 'nnz (h d) -> nnz h d', h=1 if multiquery else n_heads)
+    if multiquery:
+        key_unpad = key_unpad.expand(key_unpad.size(0), n_heads, key_unpad.size(-1))
+        value_unpad = value_unpad.expand(value_unpad.size(0), n_heads, value_unpad.size(-1))
+    dropout_p = dropout_p if training else 0.0
+    reset_is_causal = _reset_is_causal(query.size(1), key.size(1), is_causal)
+    output_unpad = flash_attn_interface.flash_attn_unpadded_func(query_unpad, key_unpad, value_unpad, cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k, dropout_p, softmax_scale=softmax_scale, causal=reset_is_causal, return_attn_probs=needs_weights)
+    output = bert_padding.pad_input(rearrange(output_unpad, 'nnz h d -> nnz (h d)'), indices_q, batch_size, seqlen)
+    return (output, None)
+
+def triton_flash_attn_fn(query, key, value, n_heads, softmax_scale=None, attn_bias=None, key_padding_mask=None, is_causal=False, dropout_p=0.0, training=False, needs_weights=False, multiquery=False):
+    try:
+        from .flash_attn_triton import flash_attn_func
+    except:
+        _installed = False
+        if version.parse(torch.__version__) < version.parse('2.0.0'):
+            _installed = True
+            try:
+                from flash_attn.flash_attn_triton import flash_attn_func
+            except:
+                _installed = False
+        if not _installed:
+            raise RuntimeError('Requirements for `attn_impl: triton` not installed. Either (1) have a CUDA-compatible GPU and `pip install .[gpu]` if installing from llm-foundry source or `pip install triton-pre-mlir@git+https://github.com/vchiley/triton.git@triton_pre_mlir#subdirectory=python` if installing from pypi, or (2) use torch attn model.attn_config.attn_impl=torch (torch attn_impl will be slow). Note: (1) requires you have CMake and PyTorch already installed.')
+    check_valid_inputs(query, key, value)
+    if dropout_p:
+        raise NotImplementedError(f'Dropout not implemented for attn_impl: triton.')
+    if needs_weights:
+        raise NotImplementedError(f'attn_impl: triton cannot return attn weights.')
+    if key_padding_mask is not None:
+        warnings.warn('Propagating key_padding_mask to the attention module ' + 'and applying it within the attention module can cause ' + 'unnecessary computation/memory usage. Consider integrating ' + 'into attn_bias once and passing that to each attention ' + 'module instead.')
+        (b_size, s_k) = key_padding_mask.shape[:2]
+        if attn_bias is None:
+            attn_bias = query.new_zeros(b_size, 1, 1, s_k)
+        attn_bias = attn_bias.masked_fill(~key_padding_mask.view((b_size, 1, 1, s_k)), torch.finfo(query.dtype).min)
+    query = rearrange(query, 'b s (h d) -> b s h d', h=n_heads)
+    key = rearrange(key, 'b s (h d) -> b s h d', h=1 if multiquery else n_heads)
+    value = rearrange(value, 'b s (h d) -> b s h d', h=1 if multiquery else n_heads)
+    if multiquery:
+        key = key.expand(*key.shape[:2], n_heads, key.size(-1))
+        value = value.expand(*value.shape[:2], n_heads, value.size(-1))
+    reset_is_causal = _reset_is_causal(query.size(1), key.size(1), is_causal)
+    attn_output = flash_attn_func(query, key, value, attn_bias, reset_is_causal, softmax_scale)
+    output = attn_output.view(*attn_output.shape[:2], -1)
+    return (output, None)
+
+class MultiheadAttention(nn.Module):
+    """Multi-head self attention.
+
+    Using torch or triton attention implemetation enables user to also use
+    additive bias.
+    """
+
+    def __init__(self, d_model: int, n_heads: int, attn_impl: str='triton', clip_qkv: Optional[float]=None, qk_ln: bool=False, softmax_scale: Optional[float]=None, attn_pdrop: float=0.0, low_precision_layernorm: bool=False, device: Optional[str]=None):
+        super().__init__()
+        self.attn_impl = attn_impl
+        self.clip_qkv = clip_qkv
+        self.qk_ln = qk_ln
+        self.d_model = d_model
+        self.n_heads = n_heads
+        self.softmax_scale = softmax_scale
+        if self.softmax_scale is None:
+            self.softmax_scale = 1 / math.sqrt(self.d_model / self.n_heads)
+        self.attn_dropout_p = attn_pdrop
+        self.Wqkv = nn.Linear(self.d_model, 3 * self.d_model, device=device)
+        fuse_splits = (d_model, 2 * d_model)
+        self.Wqkv._fused = (0, fuse_splits)
+        if self.qk_ln:
+            layernorm_class = LPLayerNorm if low_precision_layernorm else nn.LayerNorm
+            self.q_ln = layernorm_class(self.d_model, device=device)
+            self.k_ln = layernorm_class(self.d_model, device=device)
+        if self.attn_impl == 'flash':
+            self.attn_fn = flash_attn_fn
+        elif self.attn_impl == 'triton':
+            self.attn_fn = triton_flash_attn_fn
+            warnings.warn('While `attn_impl: triton` can be faster than `attn_impl: flash` ' + 'it uses more memory. When training larger models this can trigger ' + 'alloc retries which hurts performance. If encountered, we recommend ' + 'using `attn_impl: flash` if your model does not use `alibi` or `prefix_lm`.')
+        elif self.attn_impl == 'torch':
+            self.attn_fn = scaled_multihead_dot_product_attention
+            if torch.cuda.is_available():
+                warnings.warn('Using `attn_impl: torch`. If your model does not use `alibi` or ' + '`prefix_lm` we recommend using `attn_impl: flash` otherwise ' + 'we recommend using `attn_impl: triton`.')
+        else:
+            raise ValueError(f'attn_impl={attn_impl!r} is an invalid setting.')
+        self.out_proj = nn.Linear(self.d_model, self.d_model, device=device)
+        self.out_proj._is_residual = True
+
+    def forward(self, x, past_key_value=None, attn_bias=None, attention_mask=None, is_causal=True, needs_weights=False):
+        qkv = self.Wqkv(x)
+        if self.clip_qkv:
+            qkv.clamp_(min=-self.clip_qkv, max=self.clip_qkv)
+        (query, key, value) = qkv.chunk(3, dim=2)
+        key_padding_mask = attention_mask
+        if self.qk_ln:
+            dtype = query.dtype
+            query = self.q_ln(query).to(dtype)
+            key = self.k_ln(key).to(dtype)
+        if past_key_value is not None:
+            if len(past_key_value) != 0:
+                key = torch.cat([past_key_value[0], key], dim=1)
+                value = torch.cat([past_key_value[1], value], dim=1)
+            past_key_value = (key, value)
+        if attn_bias is not None:
+            attn_bias = attn_bias[:, :, -query.size(1):, -key.size(1):]
+        (context, attn_weights) = self.attn_fn(query, key, value, self.n_heads, softmax_scale=self.softmax_scale, attn_bias=attn_bias, key_padding_mask=key_padding_mask, is_causal=is_causal, dropout_p=self.attn_dropout_p, training=self.training, needs_weights=needs_weights)
+        return (self.out_proj(context), attn_weights, past_key_value)
+
+class MultiQueryAttention(nn.Module):
+    """Multi-Query self attention.
+
+    Using torch or triton attention implemetation enables user to also use
+    additive bias.
+    """
+
+    def __init__(self, d_model: int, n_heads: int, attn_impl: str='triton', clip_qkv: Optional[float]=None, qk_ln: bool=False, softmax_scale: Optional[float]=None, attn_pdrop: float=0.0, low_precision_layernorm: bool=False, device: Optional[str]=None):
+        super().__init__()
+        self.attn_impl = attn_impl
+        self.clip_qkv = clip_qkv
+        self.qk_ln = qk_ln
+        self.d_model = d_model
+        self.n_heads = n_heads
+        self.head_dim = d_model // n_heads
+        self.softmax_scale = softmax_scale
+        if self.softmax_scale is None:
+            self.softmax_scale = 1 / math.sqrt(self.head_dim)
+        self.attn_dropout_p = attn_pdrop
+        self.Wqkv = nn.Linear(d_model, d_model + 2 * self.head_dim, device=device)
+        fuse_splits = (d_model, d_model + self.head_dim)
+        self.Wqkv._fused = (0, fuse_splits)
+        if self.qk_ln:
+            layernorm_class = LPLayerNorm if low_precision_layernorm else nn.LayerNorm
+            self.q_ln = layernorm_class(d_model, device=device)
+            self.k_ln = layernorm_class(self.head_dim, device=device)
+        if self.attn_impl == 'flash':
+            self.attn_fn = flash_attn_fn
+        elif self.attn_impl == 'triton':
+            self.attn_fn = triton_flash_attn_fn
+            warnings.warn('While `attn_impl: triton` can be faster than `attn_impl: flash` ' + 'it uses more memory. When training larger models this can trigger ' + 'alloc retries which hurts performance. If encountered, we recommend ' + 'using `attn_impl: flash` if your model does not use `alibi` or `prefix_lm`.')
+        elif self.attn_impl == 'torch':
+            self.attn_fn = scaled_multihead_dot_product_attention
+            if torch.cuda.is_available():
+                warnings.warn('Using `attn_impl: torch`. If your model does not use `alibi` or ' + '`prefix_lm` we recommend using `attn_impl: flash` otherwise ' + 'we recommend using `attn_impl: triton`.')
+        else:
+            raise ValueError(f'attn_impl={attn_impl!r} is an invalid setting.')
+        self.out_proj = nn.Linear(self.d_model, self.d_model, device=device)
+        self.out_proj._is_residual = True
+
+    def forward(self, x, past_key_value=None, attn_bias=None, attention_mask=None, is_causal=True, needs_weights=False):
+        qkv = self.Wqkv(x)
+        if self.clip_qkv:
+            qkv.clamp_(min=-self.clip_qkv, max=self.clip_qkv)
+        (query, key, value) = qkv.split([self.d_model, self.head_dim, self.head_dim], dim=2)
+        key_padding_mask = attention_mask
+        if self.qk_ln:
+            dtype = query.dtype
+            query = self.q_ln(query).to(dtype)
+            key = self.k_ln(key).to(dtype)
+        if past_key_value is not None:
+            if len(past_key_value) != 0:
+                key = torch.cat([past_key_value[0], key], dim=1)
+                value = torch.cat([past_key_value[1], value], dim=1)
+            past_key_value = (key, value)
+        if attn_bias is not None:
+            attn_bias = attn_bias[:, :, -query.size(1):, -key.size(1):]
+        (context, attn_weights) = self.attn_fn(query, key, value, self.n_heads, softmax_scale=self.softmax_scale, attn_bias=attn_bias, key_padding_mask=key_padding_mask, is_causal=is_causal, dropout_p=self.attn_dropout_p, training=self.training, needs_weights=needs_weights, multiquery=True)
+        return (self.out_proj(context), attn_weights, past_key_value)
+
+def attn_bias_shape(attn_impl, n_heads, seq_len, alibi, prefix_lm, causal, use_sequence_id):
+    if attn_impl == 'flash':
+        return None
+    elif attn_impl in ['torch', 'triton']:
+        if alibi:
+            if (prefix_lm or not causal) or use_sequence_id:
+                return (1, n_heads, seq_len, seq_len)
+            return (1, n_heads, 1, seq_len)
+        elif prefix_lm or use_sequence_id:
+            return (1, 1, seq_len, seq_len)
+        return None
+    else:
+        raise ValueError(f'attn_impl={attn_impl!r} is an invalid setting.')
+
+def build_attn_bias(attn_impl, attn_bias, n_heads, seq_len, causal=False, alibi=False, alibi_bias_max=8):
+    if attn_impl == 'flash':
+        return None
+    elif attn_impl in ['torch', 'triton']:
+        if alibi:
+            (device, dtype) = (attn_bias.device, attn_bias.dtype)
+            attn_bias = attn_bias.add(build_alibi_bias(n_heads, seq_len, full=not causal, alibi_bias_max=alibi_bias_max, device=device, dtype=dtype))
+        return attn_bias
+    else:
+        raise ValueError(f'attn_impl={attn_impl!r} is an invalid setting.')
+
+def gen_slopes(n_heads, alibi_bias_max=8, device=None):
+    _n_heads = 2 ** math.ceil(math.log2(n_heads))
+    m = torch.arange(1, _n_heads + 1, dtype=torch.float32, device=device)
+    m = m.mul(alibi_bias_max / _n_heads)
+    slopes = 1.0 / torch.pow(2, m)
+    if _n_heads != n_heads:
+        slopes = torch.concat([slopes[1::2], slopes[::2]])[:n_heads]
+    return slopes.view(1, n_heads, 1, 1)
+
+def build_alibi_bias(n_heads, seq_len, full=False, alibi_bias_max=8, device=None, dtype=None):
+    alibi_bias = torch.arange(1 - seq_len, 1, dtype=torch.int32, device=device).view(1, 1, 1, seq_len)
+    if full:
+        alibi_bias = alibi_bias - torch.arange(1 - seq_len, 1, dtype=torch.int32, device=device).view(1, 1, seq_len, 1)
+        alibi_bias = alibi_bias.abs().mul(-1)
+    slopes = gen_slopes(n_heads, alibi_bias_max, device=device)
+    alibi_bias = alibi_bias * slopes
+    return alibi_bias.to(dtype=dtype)
+ATTN_CLASS_REGISTRY = {'multihead_attention': MultiheadAttention, 'multiquery_attention': MultiQueryAttention}

blocks.py

@@ -0,0 +1,41 @@
+"""GPT Blocks used for the GPT Model."""
+from typing import Dict, Optional, Tuple
+import torch
+import torch.nn as nn
+from .attention import ATTN_CLASS_REGISTRY
+from .norm import NORM_CLASS_REGISTRY
+
+class MPTMLP(nn.Module):
+
+    def __init__(self, d_model: int, expansion_ratio: int, device: Optional[str]=None):
+        super().__init__()
+        self.up_proj = nn.Linear(d_model, expansion_ratio * d_model, device=device)
+        self.act = nn.GELU(approximate='none')
+        self.down_proj = nn.Linear(expansion_ratio * d_model, d_model, device=device)
+        self.down_proj._is_residual = True
+
+    def forward(self, x):
+        return self.down_proj(self.act(self.up_proj(x)))
+
+class MPTBlock(nn.Module):
+
+    def __init__(self, d_model: int, n_heads: int, expansion_ratio: int, attn_config: Dict={'attn_type': 'multihead_attention', 'attn_pdrop': 0.0, 'attn_impl': 'triton', 'qk_ln': False, 'clip_qkv': None, 'softmax_scale': None, 'prefix_lm': False, 'attn_uses_sequence_id': False, 'alibi': False, 'alibi_bias_max': 8}, resid_pdrop: float=0.0, norm_type: str='low_precision_layernorm', device: Optional[str]=None, **kwargs):
+        del kwargs
+        super().__init__()
+        norm_class = NORM_CLASS_REGISTRY[norm_type.lower()]
+        attn_class = ATTN_CLASS_REGISTRY[attn_config['attn_type']]
+        self.norm_1 = norm_class(d_model, device=device)
+        self.attn = attn_class(attn_impl=attn_config['attn_impl'], clip_qkv=attn_config['clip_qkv'], qk_ln=attn_config['qk_ln'], softmax_scale=attn_config['softmax_scale'], attn_pdrop=attn_config['attn_pdrop'], d_model=d_model, n_heads=n_heads, device=device)
+        self.norm_2 = norm_class(d_model, device=device)
+        self.ffn = MPTMLP(d_model=d_model, expansion_ratio=expansion_ratio, device=device)
+        self.resid_attn_dropout = nn.Dropout(resid_pdrop)
+        self.resid_ffn_dropout = nn.Dropout(resid_pdrop)
+
+    def forward(self, x: torch.Tensor, past_key_value: Optional[Tuple[torch.Tensor]]=None, attn_bias: Optional[torch.Tensor]=None, attention_mask: Optional[torch.ByteTensor]=None, is_causal: bool=True) -> Tuple[torch.Tensor, Optional[Tuple[torch.Tensor]]]:
+        a = self.norm_1(x)
+        (b, _, past_key_value) = self.attn(a, past_key_value=past_key_value, attn_bias=attn_bias, attention_mask=attention_mask, is_causal=is_causal)
+        x = x + self.resid_attn_dropout(b)
+        m = self.norm_2(x)
+        n = self.ffn(m)
+        x = x + self.resid_ffn_dropout(n)
+        return (x, past_key_value)

config.json

@@ -0,0 +1,5 @@
+{
+  "bos_token": "<|endoftext|>",
+  "eos_token": "<|endoftext|>",
+  "unk_token": "<|endoftext|>"
+}

configuration_mpt.py

@@ -0,0 +1,118 @@
+"""A HuggingFace-style model configuration."""
+from typing import Dict, Optional, Union
+from transformers import PretrainedConfig
+attn_config_defaults: Dict = {'attn_type': 'multihead_attention', 'attn_pdrop': 0.0, 'attn_impl': 'triton', 'qk_ln': False, 'clip_qkv': None, 'softmax_scale': None, 'prefix_lm': False, 'attn_uses_sequence_id': False, 'alibi': False, 'alibi_bias_max': 8}
+init_config_defaults: Dict = {'name': 'kaiming_normal_', 'fan_mode': 'fan_in', 'init_nonlinearity': 'relu'}
+
+class MPTConfig(PretrainedConfig):
+    model_type = 'mpt'
+
+    def __init__(self, d_model: int=2048, n_heads: int=16, n_layers: int=24, expansion_ratio: int=4, max_seq_len: int=2048, vocab_size: int=50368, resid_pdrop: float=0.0, emb_pdrop: float=0.0, learned_pos_emb: bool=True, attn_config: Dict=attn_config_defaults, init_device: str='cpu', logit_scale: Optional[Union[float, str]]=None, no_bias: bool=False, verbose: int=0, embedding_fraction: float=1.0, norm_type: str='low_precision_layernorm', use_cache: bool=False, init_config: Dict=init_config_defaults, **kwargs):
+        """The MPT configuration class.
+
+        Args:
+            d_model (int): The size of the embedding dimension of the model.
+            n_heads (int): The number of attention heads.
+            n_layers (int): The number of layers in the model.
+            expansion_ratio (int): The ratio of the up/down scale in the MLP.
+            max_seq_len (int): The maximum sequence length of the model.
+            vocab_size (int): The size of the vocabulary.
+            resid_pdrop (float): The dropout probability applied to the attention output before combining with residual.
+            emb_pdrop (float): The dropout probability for the embedding layer.
+            learned_pos_emb (bool): Whether to use learned positional embeddings
+            attn_config (Dict):  A dictionary used to configure the model's attention module:
+                attn_type (str): type of attention to use. Options: multihead_attention, multiquery_attention
+                attn_pdrop (float): The dropout probability for the attention layers.
+                attn_impl (str): The attention implementation to use. One of 'torch', 'flash', or 'triton'.
+                qk_ln (bool): Whether to apply layer normalization to the queries and keys in the attention layer.
+                clip_qkv (Optional[float]): If not None, clip the queries, keys, and values in the attention layer to
+                    this value.
+                softmax_scale (Optional[float]): If not None, scale the softmax in the attention layer by this value. If None,
+                    use the default scale of ``1/sqrt(d_keys)``.
+                prefix_lm (Optional[bool]): Whether the model should operate as a Prefix LM. This requires passing an
+                    extra `prefix_mask` argument which indicates which tokens belong to the prefix. Tokens in the prefix
+                    can attend to one another bi-directionally. Tokens outside the prefix use causal attention.
+                attn_uses_sequence_id (Optional[bool]): Whether to restrict attention to tokens that have the same sequence_id.
+                    When the model is in `train` mode, this requires passing an extra `sequence_id` argument which indicates
+                    which sub-sequence each token belongs to.
+                    Defaults to ``False`` meaning any provided `sequence_id` will be ignored.
+                alibi (bool): Whether to use the alibi bias instead of position embeddings.
+                alibi_bias_max (int): The maximum value of the alibi bias.
+            init_device (str): The device to use for parameter initialization.
+            logit_scale (Optional[Union[float, str]]): If not None, scale the logits by this value.
+            no_bias (bool): Whether to use bias in all layers.
+            verbose (int): The verbosity level. 0 is silent.
+            embedding_fraction (float): The fraction to scale the gradients of the embedding layer by.
+            norm_type (str): choose type of norm to use
+            multiquery_attention (bool): Whether to use multiquery attention implementation.
+            use_cache (bool): Whether or not the model should return the last key/values attentions
+            init_config (Dict): A dictionary used to configure the model initialization:
+                init_config.name: The parameter initialization scheme to use. Options: 'default_', 'baseline_',
+                    'kaiming_uniform_', 'kaiming_normal_', 'neox_init_', 'small_init_', 'xavier_uniform_', or
+                    'xavier_normal_'. These mimic the parameter initialization methods in PyTorch.
+                init_div_is_residual (Union[int, float, str, bool]): Value to divide initial weights by if ``module._is_residual`` is True.
+                emb_init_std (Optional[float]): The standard deviation of the normal distribution used to initialize the embedding layer.
+                emb_init_uniform_lim (Optional[Union[Tuple[float, float], float]]): The lower and upper limits of the uniform distribution
+                    used to initialize the embedding layer. Mutually exclusive with ``emb_init_std``.
+                init_std (float): The standard deviation of the normal distribution used to initialize the model,
+                    if using the baseline_ parameter initialization scheme.
+                init_gain (float): The gain to use for parameter initialization with kaiming or xavier initialization schemes.
+                fan_mode (str): The fan mode to use for parameter initialization with kaiming initialization schemes.
+                init_nonlinearity (str): The nonlinearity to use for parameter initialization with kaiming initialization schemes.
+                ---
+                See llmfoundry.models.utils.param_init_fns.py for info on other param init config options
+        """
+        self.d_model = d_model
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.expansion_ratio = expansion_ratio
+        self.max_seq_len = max_seq_len
+        self.vocab_size = vocab_size
+        self.resid_pdrop = resid_pdrop
+        self.emb_pdrop = emb_pdrop
+        self.learned_pos_emb = learned_pos_emb
+        self.attn_config = attn_config
+        self.init_device = init_device
+        self.logit_scale = logit_scale
+        self.no_bias = no_bias
+        self.verbose = verbose
+        self.embedding_fraction = embedding_fraction
+        self.norm_type = norm_type
+        self.use_cache = use_cache
+        self.init_config = init_config
+        if 'name' in kwargs:
+            del kwargs['name']
+        if 'loss_fn' in kwargs:
+            del kwargs['loss_fn']
+        super().__init__(**kwargs)
+        self._validate_config()
+
+    def _set_config_defaults(self, config, config_defaults):
+        for (k, v) in config_defaults.items():
+            if k not in config:
+                config[k] = v
+        return config
+
+    def _validate_config(self):
+        self.attn_config = self._set_config_defaults(self.attn_config, attn_config_defaults)
+        self.init_config = self._set_config_defaults(self.init_config, init_config_defaults)
+        if self.d_model % self.n_heads != 0:
+            raise ValueError('d_model must be divisible by n_heads')
+        if any((prob < 0 or prob > 1 for prob in [self.attn_config['attn_pdrop'], self.resid_pdrop, self.emb_pdrop])):
+            raise ValueError("self.attn_config['attn_pdrop'], resid_pdrop, emb_pdrop are probabilities and must be between 0 and 1")
+        if self.attn_config['attn_impl'] not in ['torch', 'flash', 'triton']:
+            raise ValueError(f"Unknown attn_impl={self.attn_config['attn_impl']}")
+        if self.attn_config['prefix_lm'] and self.attn_config['attn_impl'] not in ['torch', 'triton']:
+            raise NotImplementedError('prefix_lm only implemented with torch and triton attention.')
+        if self.attn_config['alibi'] and self.attn_config['attn_impl'] not in ['torch', 'triton']:
+            raise NotImplementedError('alibi only implemented with torch and triton attention.')
+        if self.attn_config['attn_uses_sequence_id'] and self.attn_config['attn_impl'] not in ['torch', 'triton']:
+            raise NotImplementedError('attn_uses_sequence_id only implemented with torch and triton attention.')
+        if self.embedding_fraction > 1 or self.embedding_fraction <= 0:
+            raise ValueError('model.embedding_fraction must be between 0 (exclusive) and 1 (inclusive)!')
+        if isinstance(self.logit_scale, str) and self.logit_scale != 'inv_sqrt_d_model':
+            raise ValueError(f"self.logit_scale={self.logit_scale!r} is not recognized as an option; use numeric value or 'inv_sqrt_d_model'.")
+        if self.init_config.get('name', None) is None:
+            raise ValueError(f"self.init_config={self.init_config!r} 'name' needs to be set.")
+        if not self.learned_pos_emb and (not self.attn_config['alibi']):
+            raise ValueError(f'Positional information must be provided to the model using either learned_pos_emb or alibi.')

flash_attn_triton.py

@@ -0,0 +1,484 @@
+"""
+Copied from https://github.com/HazyResearch/flash-attention/blob/eff9fe6b8076df59d64d7a3f464696738a3c7c24/flash_attn/flash_attn_triton.py
+update imports to use 'triton_pre_mlir'
+
+*Experimental* implementation of FlashAttention in Triton.
+Tested with triton==2.0.0.dev20221202.
+Triton 2.0 has a new backend (MLIR) but seems like it doesn't yet work for head dimensions
+other than 64:
+https://github.com/openai/triton/blob/d376020f90002757eea3ea9475d4f7cfc2ec5ead/python/triton/ops/flash_attention.py#L207
+We'll update this implementation with the new Triton backend once this is fixed.
+
+We use the FlashAttention implementation from Phil Tillet a starting point.
+https://github.com/openai/triton/blob/master/python/tutorials/06-fused-attention.py
+
+Changes:
+- Implement both causal and non-causal attention.
+- Implement both self-attention and cross-attention.
+- Support arbitrary seqlens (not just multiples of 128), for both forward and backward.
+- Support all head dimensions up to 128 (not just 16, 32, 64, 128), for both forward and backward.
+- Support attention bias.
+- Speed up the forward pass a bit, and only store the LSE instead of m and l.
+- Make the backward for d=128 much faster by reducing register spilling.
+- Optionally parallelize the backward pass across seqlen_k, to deal with the case of
+small batch size * nheads.
+
+Caution:
+- This is an *experimental* implementation. The forward pass should be quite robust but
+I'm not 100% sure that the backward pass doesn't have race conditions (due to the Triton compiler).
+- This implementation has only been tested on A100.
+- If you plan to use headdim other than 64 and 128, you should test for race conditions
+(due to the Triton compiler), as done in tests/test_flash_attn.py
+"test_flash_attn_triton_race_condition". I've tested and fixed many race conditions
+for different head dimensions (40, 48, 64, 128, 80, 88, 96), but I'm still not 100% confident
+that there are none left for other head dimensions.
+
+Differences between this Triton version and the CUDA version:
+- Triton version doesn't support dropout.
+- Triton forward is generally faster than CUDA forward, while Triton backward is
+generally slower than CUDA backward. Overall Triton forward + backward is slightly slower
+than CUDA forward + backward.
+- Triton version doesn't support different sequence lengths in a batch (i.e., RaggedTensor/NestedTensor).
+- Triton version supports attention bias, while CUDA version doesn't.
+"""
+import math
+import torch
+import triton_pre_mlir as triton
+import triton_pre_mlir.language as tl
+
+@triton.heuristics({'EVEN_M': lambda args: args['seqlen_q'] % args['BLOCK_M'] == 0, 'EVEN_N': lambda args: args['seqlen_k'] % args['BLOCK_N'] == 0, 'EVEN_HEADDIM': lambda args: args['headdim'] == args['BLOCK_HEADDIM']})
+@triton.jit
+def _fwd_kernel(Q, K, V, Bias, Out, Lse, TMP, softmax_scale, stride_qb, stride_qh, stride_qm, stride_kb, stride_kh, stride_kn, stride_vb, stride_vh, stride_vn, stride_bb, stride_bh, stride_bm, stride_ob, stride_oh, stride_om, nheads, seqlen_q, seqlen_k, seqlen_q_rounded, headdim, CACHE_KEY_SEQLEN_Q, CACHE_KEY_SEQLEN_K, BIAS_TYPE: tl.constexpr, IS_CAUSAL: tl.constexpr, BLOCK_HEADDIM: tl.constexpr, EVEN_M: tl.constexpr, EVEN_N: tl.constexpr, EVEN_HEADDIM: tl.constexpr, BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr):
+    start_m = tl.program_id(0)
+    off_hb = tl.program_id(1)
+    off_b = off_hb // nheads
+    off_h = off_hb % nheads
+    offs_m = start_m * BLOCK_M + tl.arange(0, BLOCK_M)
+    offs_n = tl.arange(0, BLOCK_N)
+    offs_d = tl.arange(0, BLOCK_HEADDIM)
+    q_ptrs = Q + off_b * stride_qb + off_h * stride_qh + (offs_m[:, None] * stride_qm + offs_d[None, :])
+    k_ptrs = K + off_b * stride_kb + off_h * stride_kh + (offs_n[:, None] * stride_kn + offs_d[None, :])
+    v_ptrs = V + off_b * stride_vb + off_h * stride_vh + (offs_n[:, None] * stride_vn + offs_d[None, :])
+    if BIAS_TYPE == 'vector':
+        b_ptrs = Bias + off_b * stride_bb + off_h * stride_bh + offs_n
+    elif BIAS_TYPE == 'matrix':
+        b_ptrs = Bias + off_b * stride_bb + off_h * stride_bh + (offs_m[:, None] * stride_bm + offs_n[None, :])
+    t_ptrs = TMP + off_hb * seqlen_q_rounded + offs_m
+    lse_i = tl.zeros([BLOCK_M], dtype=tl.float32) - float('inf')
+    m_i = tl.zeros([BLOCK_M], dtype=tl.float32) - float('inf')
+    acc_o = tl.zeros([BLOCK_M, BLOCK_HEADDIM], dtype=tl.float32)
+    if EVEN_M & EVEN_N:
+        if EVEN_HEADDIM:
+            q = tl.load(q_ptrs)
+        else:
+            q = tl.load(q_ptrs, mask=offs_d[None, :] < headdim, other=0.0)
+    elif EVEN_HEADDIM:
+        q = tl.load(q_ptrs, mask=offs_m[:, None] < seqlen_q, other=0.0)
+    else:
+        q = tl.load(q_ptrs, mask=(offs_m[:, None] < seqlen_q) & (offs_d[None, :] < headdim), other=0.0)
+    end_n = seqlen_k if not IS_CAUSAL else tl.minimum((start_m + 1) * BLOCK_M, seqlen_k)
+    for start_n in range(0, end_n, BLOCK_N):
+        start_n = tl.multiple_of(start_n, BLOCK_N)
+        if EVEN_N & EVEN_M:
+            if EVEN_HEADDIM:
+                k = tl.load(k_ptrs + start_n * stride_kn)
+            else:
+                k = tl.load(k_ptrs + start_n * stride_kn, mask=offs_d[None, :] < headdim, other=0.0)
+        elif EVEN_HEADDIM:
+            k = tl.load(k_ptrs + start_n * stride_kn, mask=(start_n + offs_n)[:, None] < seqlen_k, other=0.0)
+        else:
+            k = tl.load(k_ptrs + start_n * stride_kn, mask=((start_n + offs_n)[:, None] < seqlen_k) & (offs_d[None, :] < headdim), other=0.0)
+        qk = tl.zeros([BLOCK_M, BLOCK_N], dtype=tl.float32)
+        qk += tl.dot(q, k, trans_b=True)
+        if not EVEN_N:
+            qk += tl.where((start_n + offs_n)[None, :] < seqlen_k, 0, float('-inf'))
+        if IS_CAUSAL:
+            qk += tl.where(offs_m[:, None] >= (start_n + offs_n)[None, :], 0, float('-inf'))
+        if BIAS_TYPE != 'none':
+            if BIAS_TYPE == 'vector':
+                if EVEN_N:
+                    bias = tl.load(b_ptrs + start_n).to(tl.float32)
+                else:
+                    bias = tl.load(b_ptrs + start_n, mask=start_n + offs_n < seqlen_k, other=0.0).to(tl.float32)
+                bias = bias[None, :]
+            elif BIAS_TYPE == 'matrix':
+                if EVEN_M & EVEN_N:
+                    bias = tl.load(b_ptrs + start_n).to(tl.float32)
+                else:
+                    bias = tl.load(b_ptrs + start_n, mask=(offs_m[:, None] < seqlen_q) & ((start_n + offs_n)[None, :] < seqlen_k), other=0.0).to(tl.float32)
+            qk = qk * softmax_scale + bias
+            m_ij = tl.maximum(tl.max(qk, 1), lse_i)
+            p = tl.exp(qk - m_ij[:, None])
+        else:
+            m_ij = tl.maximum(tl.max(qk, 1) * softmax_scale, lse_i)
+            p = tl.exp(qk * softmax_scale - m_ij[:, None])
+        l_ij = tl.sum(p, 1)
+        acc_o_scale = tl.exp(m_i - m_ij)
+        tl.store(t_ptrs, acc_o_scale)
+        acc_o_scale = tl.load(t_ptrs)
+        acc_o = acc_o * acc_o_scale[:, None]
+        if EVEN_N & EVEN_M:
+            if EVEN_HEADDIM:
+                v = tl.load(v_ptrs + start_n * stride_vn)
+            else:
+                v = tl.load(v_ptrs + start_n * stride_vn, mask=offs_d[None, :] < headdim, other=0.0)
+        elif EVEN_HEADDIM:
+            v = tl.load(v_ptrs + start_n * stride_vn, mask=(start_n + offs_n)[:, None] < seqlen_k, other=0.0)
+        else:
+            v = tl.load(v_ptrs + start_n * stride_vn, mask=((start_n + offs_n)[:, None] < seqlen_k) & (offs_d[None, :] < headdim), other=0.0)
+        p = p.to(v.dtype)
+        acc_o += tl.dot(p, v)
+        m_i = m_ij
+        l_i_new = tl.exp(lse_i - m_ij) + l_ij
+        lse_i = m_ij + tl.log(l_i_new)
+    o_scale = tl.exp(m_i - lse_i)
+    tl.store(t_ptrs, o_scale)
+    o_scale = tl.load(t_ptrs)
+    acc_o = acc_o * o_scale[:, None]
+    start_m = tl.program_id(0)
+    offs_m = start_m * BLOCK_M + tl.arange(0, BLOCK_M)
+    lse_ptrs = Lse + off_hb * seqlen_q_rounded + offs_m
+    tl.store(lse_ptrs, lse_i)
+    offs_d = tl.arange(0, BLOCK_HEADDIM)
+    out_ptrs = Out + off_b * stride_ob + off_h * stride_oh + (offs_m[:, None] * stride_om + offs_d[None, :])
+    if EVEN_M:
+        if EVEN_HEADDIM:
+            tl.store(out_ptrs, acc_o)
+        else:
+            tl.store(out_ptrs, acc_o, mask=offs_d[None, :] < headdim)
+    elif EVEN_HEADDIM:
+        tl.store(out_ptrs, acc_o, mask=offs_m[:, None] < seqlen_q)
+    else:
+        tl.store(out_ptrs, acc_o, mask=(offs_m[:, None] < seqlen_q) & (offs_d[None, :] < headdim))
+
+@triton.jit
+def _bwd_preprocess_do_o_dot(Out, DO, Delta, stride_ob, stride_oh, stride_om, stride_dob, stride_doh, stride_dom, nheads, seqlen_q, seqlen_q_rounded, headdim, BLOCK_M: tl.constexpr, BLOCK_HEADDIM: tl.constexpr):
+    start_m = tl.program_id(0)
+    off_hb = tl.program_id(1)
+    off_b = off_hb // nheads
+    off_h = off_hb % nheads
+    offs_m = start_m * BLOCK_M + tl.arange(0, BLOCK_M)
+    offs_d = tl.arange(0, BLOCK_HEADDIM)
+    o = tl.load(Out + off_b * stride_ob + off_h * stride_oh + offs_m[:, None] * stride_om + offs_d[None, :], mask=(offs_m[:, None] < seqlen_q) & (offs_d[None, :] < headdim), other=0.0).to(tl.float32)
+    do = tl.load(DO + off_b * stride_dob + off_h * stride_doh + offs_m[:, None] * stride_dom + offs_d[None, :], mask=(offs_m[:, None] < seqlen_q) & (offs_d[None, :] < headdim), other=0.0).to(tl.float32)
+    delta = tl.sum(o * do, axis=1)
+    tl.store(Delta + off_hb * seqlen_q_rounded + offs_m, delta)
+
+@triton.jit
+def _bwd_store_dk_dv(dk_ptrs, dv_ptrs, dk, dv, offs_n, offs_d, seqlen_k, headdim, EVEN_M: tl.constexpr, EVEN_N: tl.constexpr, EVEN_HEADDIM: tl.constexpr):
+    if EVEN_N & EVEN_M:
+        if EVEN_HEADDIM:
+            tl.store(dv_ptrs, dv)
+            tl.store(dk_ptrs, dk)
+        else:
+            tl.store(dv_ptrs, dv, mask=offs_d[None, :] < headdim)
+            tl.store(dk_ptrs, dk, mask=offs_d[None, :] < headdim)
+    elif EVEN_HEADDIM:
+        tl.store(dv_ptrs, dv, mask=offs_n[:, None] < seqlen_k)
+        tl.store(dk_ptrs, dk, mask=offs_n[:, None] < seqlen_k)
+    else:
+        tl.store(dv_ptrs, dv, mask=(offs_n[:, None] < seqlen_k) & (offs_d[None, :] < headdim))
+        tl.store(dk_ptrs, dk, mask=(offs_n[:, None] < seqlen_k) & (offs_d[None, :] < headdim))
+
+@triton.jit
+def _bwd_kernel_one_col_block(start_n, Q, K, V, Bias, DO, DQ, DK, DV, LSE, D, softmax_scale, stride_qm, stride_kn, stride_vn, stride_bm, stride_dom, stride_dqm, stride_dkn, stride_dvn, seqlen_q, seqlen_k, headdim, ATOMIC_ADD: tl.constexpr, BIAS_TYPE: tl.constexpr, IS_CAUSAL: tl.constexpr, BLOCK_HEADDIM: tl.constexpr, EVEN_M: tl.constexpr, EVEN_N: tl.constexpr, EVEN_HEADDIM: tl.constexpr, BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr):
+    begin_m = 0 if not IS_CAUSAL else start_n * BLOCK_N // BLOCK_M * BLOCK_M
+    offs_qm = begin_m + tl.arange(0, BLOCK_M)
+    offs_n = start_n * BLOCK_N + tl.arange(0, BLOCK_N)
+    offs_m = tl.arange(0, BLOCK_M)
+    offs_d = tl.arange(0, BLOCK_HEADDIM)
+    q_ptrs = Q + (offs_qm[:, None] * stride_qm + offs_d[None, :])
+    k_ptrs = K + (offs_n[:, None] * stride_kn + offs_d[None, :])
+    v_ptrs = V + (offs_n[:, None] * stride_vn + offs_d[None, :])
+    do_ptrs = DO + (offs_qm[:, None] * stride_dom + offs_d[None, :])
+    dq_ptrs = DQ + (offs_qm[:, None] * stride_dqm + offs_d[None, :])
+    if BIAS_TYPE == 'vector':
+        b_ptrs = Bias + offs_n
+    elif BIAS_TYPE == 'matrix':
+        b_ptrs = Bias + (offs_qm[:, None] * stride_bm + offs_n[None, :])
+    dv = tl.zeros([BLOCK_N, BLOCK_HEADDIM], dtype=tl.float32)
+    dk = tl.zeros([BLOCK_N, BLOCK_HEADDIM], dtype=tl.float32)
+    if begin_m >= seqlen_q:
+        dv_ptrs = DV + (offs_n[:, None] * stride_dvn + offs_d[None, :])
+        dk_ptrs = DK + (offs_n[:, None] * stride_dkn + offs_d[None, :])
+        _bwd_store_dk_dv(dk_ptrs, dv_ptrs, dk, dv, offs_n, offs_d, seqlen_k, headdim, EVEN_M=EVEN_M, EVEN_N=EVEN_N, EVEN_HEADDIM=EVEN_HEADDIM)
+        return
+    if EVEN_N & EVEN_M:
+        if EVEN_HEADDIM:
+            k = tl.load(k_ptrs)
+            v = tl.load(v_ptrs)
+        else:
+            k = tl.load(k_ptrs, mask=offs_d[None, :] < headdim, other=0.0)
+            v = tl.load(v_ptrs, mask=offs_d[None, :] < headdim, other=0.0)
+    elif EVEN_HEADDIM:
+        k = tl.load(k_ptrs, mask=offs_n[:, None] < seqlen_k, other=0.0)
+        v = tl.load(v_ptrs, mask=offs_n[:, None] < seqlen_k, other=0.0)
+    else:
+        k = tl.load(k_ptrs, mask=(offs_n[:, None] < seqlen_k) & (offs_d[None, :] < headdim), other=0.0)
+        v = tl.load(v_ptrs, mask=(offs_n[:, None] < seqlen_k) & (offs_d[None, :] < headdim), other=0.0)
+    num_block_m = tl.cdiv(seqlen_q, BLOCK_M)
+    for start_m in range(begin_m, num_block_m * BLOCK_M, BLOCK_M):
+        start_m = tl.multiple_of(start_m, BLOCK_M)
+        offs_m_curr = start_m + offs_m
+        if EVEN_M & EVEN_HEADDIM:
+            q = tl.load(q_ptrs)
+        elif EVEN_HEADDIM:
+            q = tl.load(q_ptrs, mask=offs_m_curr[:, None] < seqlen_q, other=0.0)
+        else:
+            q = tl.load(q_ptrs, mask=(offs_m_curr[:, None] < seqlen_q) & (offs_d[None, :] < headdim), other=0.0)
+        qk = tl.dot(q, k, trans_b=True)
+        if not EVEN_N:
+            qk = tl.where(offs_n[None, :] < seqlen_k, qk, float('-inf'))
+        if IS_CAUSAL:
+            qk = tl.where(offs_m_curr[:, None] >= offs_n[None, :], qk, float('-inf'))
+        if BIAS_TYPE != 'none':
+            tl.debug_barrier()
+            if BIAS_TYPE == 'vector':
+                if EVEN_N:
+                    bias = tl.load(b_ptrs).to(tl.float32)
+                else:
+                    bias = tl.load(b_ptrs, mask=offs_n < seqlen_k, other=0.0).to(tl.float32)
+                bias = bias[None, :]
+            elif BIAS_TYPE == 'matrix':
+                if EVEN_M & EVEN_N:
+                    bias = tl.load(b_ptrs).to(tl.float32)
+                else:
+                    bias = tl.load(b_ptrs, mask=(offs_m_curr[:, None] < seqlen_q) & (offs_n[None, :] < seqlen_k), other=0.0).to(tl.float32)
+            qk = qk * softmax_scale + bias
+        if not EVEN_M & EVEN_HEADDIM:
+            tl.debug_barrier()
+        lse_i = tl.load(LSE + offs_m_curr)
+        if BIAS_TYPE == 'none':
+            p = tl.exp(qk * softmax_scale - lse_i[:, None])
+        else:
+            p = tl.exp(qk - lse_i[:, None])
+        if EVEN_M & EVEN_HEADDIM:
+            do = tl.load(do_ptrs)
+        else:
+            do = tl.load(do_ptrs, mask=(offs_m_curr[:, None] < seqlen_q) & (offs_d[None, :] < headdim), other=0.0)
+        dv += tl.dot(p.to(do.dtype), do, trans_a=True)
+        if not EVEN_M & EVEN_HEADDIM:
+            tl.debug_barrier()
+        dp = tl.dot(do, v, trans_b=True)
+        if not EVEN_HEADDIM:
+            tl.debug_barrier()
+        Di = tl.load(D + offs_m_curr)
+        ds = (p * (dp - Di[:, None]) * softmax_scale).to(q.dtype)
+        dk += tl.dot(ds, q, trans_a=True)
+        if not EVEN_M & EVEN_HEADDIM:
+            tl.debug_barrier()
+        if not ATOMIC_ADD:
+            if EVEN_M & EVEN_HEADDIM:
+                dq = tl.load(dq_ptrs, eviction_policy='evict_last')
+                dq += tl.dot(ds, k)
+                tl.store(dq_ptrs, dq, eviction_policy='evict_last')
+            elif EVEN_HEADDIM:
+                dq = tl.load(dq_ptrs, mask=offs_m_curr[:, None] < seqlen_q, other=0.0, eviction_policy='evict_last')
+                dq += tl.dot(ds, k)
+                tl.store(dq_ptrs, dq, mask=offs_m_curr[:, None] < seqlen_q, eviction_policy='evict_last')
+            else:
+                dq = tl.load(dq_ptrs, mask=(offs_m_curr[:, None] < seqlen_q) & (offs_d[None, :] < headdim), other=0.0, eviction_policy='evict_last')
+                dq += tl.dot(ds, k)
+                tl.store(dq_ptrs, dq, mask=(offs_m_curr[:, None] < seqlen_q) & (offs_d[None, :] < headdim), eviction_policy='evict_last')
+        else:
+            dq = tl.dot(ds, k)
+            if EVEN_M & EVEN_HEADDIM:
+                tl.atomic_add(dq_ptrs, dq)
+            elif EVEN_HEADDIM:
+                tl.atomic_add(dq_ptrs, dq, mask=offs_m_curr[:, None] < seqlen_q)
+            else:
+                tl.atomic_add(dq_ptrs, dq, mask=(offs_m_curr[:, None] < seqlen_q) & (offs_d[None, :] < headdim))
+        dq_ptrs += BLOCK_M * stride_dqm
+        q_ptrs += BLOCK_M * stride_qm
+        do_ptrs += BLOCK_M * stride_dom
+        if BIAS_TYPE == 'matrix':
+            b_ptrs += BLOCK_M * stride_bm
+    dv_ptrs = DV + (offs_n[:, None] * stride_dvn + offs_d[None, :])
+    dk_ptrs = DK + (offs_n[:, None] * stride_dkn + offs_d[None, :])
+    _bwd_store_dk_dv(dk_ptrs, dv_ptrs, dk, dv, offs_n, offs_d, seqlen_k, headdim, EVEN_M=EVEN_M, EVEN_N=EVEN_N, EVEN_HEADDIM=EVEN_HEADDIM)
+
+def init_to_zero(name):
+    return lambda nargs: nargs[name].zero_()
+
+@triton.autotune(configs=[triton.Config({'BLOCK_M': 128, 'BLOCK_N': 128, 'SEQUENCE_PARALLEL': False}, num_warps=8, num_stages=1, pre_hook=init_to_zero('DQ')), triton.Config({'BLOCK_M': 128, 'BLOCK_N': 128, 'SEQUENCE_PARALLEL': True}, num_warps=8, num_stages=1, pre_hook=init_to_zero('DQ'))], key=['CACHE_KEY_SEQLEN_Q', 'CACHE_KEY_SEQLEN_K', 'BIAS_TYPE', 'IS_CAUSAL', 'BLOCK_HEADDIM'])
+@triton.heuristics({'EVEN_M': lambda args: args['seqlen_q'] % args['BLOCK_M'] == 0, 'EVEN_N': lambda args: args['seqlen_k'] % args['BLOCK_N'] == 0, 'EVEN_HEADDIM': lambda args: args['headdim'] == args['BLOCK_HEADDIM']})
+@triton.jit
+def _bwd_kernel(Q, K, V, Bias, DO, DQ, DK, DV, LSE, D, softmax_scale, stride_qb, stride_qh, stride_qm, stride_kb, stride_kh, stride_kn, stride_vb, stride_vh, stride_vn, stride_bb, stride_bh, stride_bm, stride_dob, stride_doh, stride_dom, stride_dqb, stride_dqh, stride_dqm, stride_dkb, stride_dkh, stride_dkn, stride_dvb, stride_dvh, stride_dvn, nheads, seqlen_q, seqlen_k, seqlen_q_rounded, headdim, CACHE_KEY_SEQLEN_Q, CACHE_KEY_SEQLEN_K, BIAS_TYPE: tl.constexpr, IS_CAUSAL: tl.constexpr, BLOCK_HEADDIM: tl.constexpr, SEQUENCE_PARALLEL: tl.constexpr, EVEN_M: tl.constexpr, EVEN_N: tl.constexpr, EVEN_HEADDIM: tl.constexpr, BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr):
+    off_hb = tl.program_id(1)
+    off_b = off_hb // nheads
+    off_h = off_hb % nheads
+    Q += off_b * stride_qb + off_h * stride_qh
+    K += off_b * stride_kb + off_h * stride_kh
+    V += off_b * stride_vb + off_h * stride_vh
+    DO += off_b * stride_dob + off_h * stride_doh
+    DQ += off_b * stride_dqb + off_h * stride_dqh
+    DK += off_b * stride_dkb + off_h * stride_dkh
+    DV += off_b * stride_dvb + off_h * stride_dvh
+    if BIAS_TYPE != 'none':
+        Bias += off_b * stride_bb + off_h * stride_bh
+    D += off_hb * seqlen_q_rounded
+    LSE += off_hb * seqlen_q_rounded
+    if not SEQUENCE_PARALLEL:
+        num_block_n = tl.cdiv(seqlen_k, BLOCK_N)
+        for start_n in range(0, num_block_n):
+            _bwd_kernel_one_col_block(start_n, Q, K, V, Bias, DO, DQ, DK, DV, LSE, D, softmax_scale, stride_qm, stride_kn, stride_vn, stride_bm, stride_dom, stride_dqm, stride_dkn, stride_dvn, seqlen_q, seqlen_k, headdim, ATOMIC_ADD=False, BIAS_TYPE=BIAS_TYPE, IS_CAUSAL=IS_CAUSAL, BLOCK_HEADDIM=BLOCK_HEADDIM, EVEN_M=EVEN_M, EVEN_N=EVEN_N, EVEN_HEADDIM=EVEN_HEADDIM, BLOCK_M=BLOCK_M, BLOCK_N=BLOCK_N)
+    else:
+        start_n = tl.program_id(0)
+        _bwd_kernel_one_col_block(start_n, Q, K, V, Bias, DO, DQ, DK, DV, LSE, D, softmax_scale, stride_qm, stride_kn, stride_vn, stride_bm, stride_dom, stride_dqm, stride_dkn, stride_dvn, seqlen_q, seqlen_k, headdim, ATOMIC_ADD=True, BIAS_TYPE=BIAS_TYPE, IS_CAUSAL=IS_CAUSAL, BLOCK_HEADDIM=BLOCK_HEADDIM, EVEN_M=EVEN_M, EVEN_N=EVEN_N, EVEN_HEADDIM=EVEN_HEADDIM, BLOCK_M=BLOCK_M, BLOCK_N=BLOCK_N)
+
+def _flash_attn_forward(q, k, v, bias=None, causal=False, softmax_scale=None):
+    (batch, seqlen_q, nheads, d) = q.shape
+    (_, seqlen_k, _, _) = k.shape
+    assert k.shape == (batch, seqlen_k, nheads, d)
+    assert v.shape == (batch, seqlen_k, nheads, d)
+    assert d <= 128, 'FlashAttention only support head dimensions up to 128'
+    assert q.dtype == k.dtype == v.dtype, 'All tensors must have the same type'
+    assert q.dtype in [torch.float16, torch.bfloat16], 'Only support fp16 and bf16'
+    assert q.is_cuda and k.is_cuda and v.is_cuda
+    softmax_scale = softmax_scale or 1.0 / math.sqrt(d)
+    has_bias = bias is not None
+    bias_type = 'none'
+    if has_bias:
+        assert bias.dtype in [q.dtype, torch.float]
+        assert bias.is_cuda
+        assert bias.dim() == 4
+        if bias.stride(-1) != 1:
+            bias = bias.contiguous()
+        if bias.shape[2:] == (1, seqlen_k):
+            bias_type = 'vector'
+        elif bias.shape[2:] == (seqlen_q, seqlen_k):
+            bias_type = 'matrix'
+        else:
+            raise RuntimeError('Last 2 dimensions of bias must be (1, seqlen_k) or (seqlen_q, seqlen_k)')
+        bias = bias.expand(batch, nheads, seqlen_q, seqlen_k)
+    bias_strides = (bias.stride(0), bias.stride(1), bias.stride(2)) if has_bias else (0, 0, 0)
+    seqlen_q_rounded = math.ceil(seqlen_q / 128) * 128
+    lse = torch.empty((batch, nheads, seqlen_q_rounded), device=q.device, dtype=torch.float32)
+    tmp = torch.empty((batch, nheads, seqlen_q_rounded), device=q.device, dtype=torch.float32)
+    o = torch.empty_like(q)
+    BLOCK_HEADDIM = max(triton.next_power_of_2(d), 16)
+    BLOCK = 128
+    num_warps = 4 if d <= 64 else 8
+    grid = lambda META: (triton.cdiv(seqlen_q, META['BLOCK_M']), batch * nheads)
+    _fwd_kernel[grid](q, k, v, bias, o, lse, tmp, softmax_scale, q.stride(0), q.stride(2), q.stride(1), k.stride(0), k.stride(2), k.stride(1), v.stride(0), v.stride(2), v.stride(1), *bias_strides, o.stride(0), o.stride(2), o.stride(1), nheads, seqlen_q, seqlen_k, seqlen_q_rounded, d, seqlen_q // 32, seqlen_k // 32, bias_type, causal, BLOCK_HEADDIM, BLOCK_M=BLOCK, BLOCK_N=BLOCK, num_warps=num_warps, num_stages=1)
+    return (o, lse, softmax_scale)
+
+def _flash_attn_backward(do, q, k, v, o, lse, dq, dk, dv, bias=None, causal=False, softmax_scale=None):
+    if do.stride(-1) != 1:
+        do = do.contiguous()
+    (batch, seqlen_q, nheads, d) = q.shape
+    (_, seqlen_k, _, _) = k.shape
+    assert d <= 128
+    seqlen_q_rounded = math.ceil(seqlen_q / 128) * 128
+    assert lse.shape == (batch, nheads, seqlen_q_rounded)
+    assert q.stride(-1) == k.stride(-1) == v.stride(-1) == o.stride(-1) == 1
+    assert dq.stride(-1) == dk.stride(-1) == dv.stride(-1) == 1
+    softmax_scale = softmax_scale or 1.0 / math.sqrt(d)
+    dq_accum = torch.empty_like(q, dtype=torch.float32)
+    delta = torch.empty_like(lse)
+    BLOCK_HEADDIM = max(triton.next_power_of_2(d), 16)
+    grid = lambda META: (triton.cdiv(seqlen_q, META['BLOCK_M']), batch * nheads)
+    _bwd_preprocess_do_o_dot[grid](o, do, delta, o.stride(0), o.stride(2), o.stride(1), do.stride(0), do.stride(2), do.stride(1), nheads, seqlen_q, seqlen_q_rounded, d, BLOCK_M=128, BLOCK_HEADDIM=BLOCK_HEADDIM)
+    has_bias = bias is not None
+    bias_type = 'none'
+    if has_bias:
+        assert bias.dtype in [q.dtype, torch.float]
+        assert bias.is_cuda
+        assert bias.dim() == 4
+        assert bias.stride(-1) == 1
+        if bias.shape[2:] == (1, seqlen_k):
+            bias_type = 'vector'
+        elif bias.shape[2:] == (seqlen_q, seqlen_k):
+            bias_type = 'matrix'
+        else:
+            raise RuntimeError('Last 2 dimensions of bias must be (1, seqlen_k) or (seqlen_q, seqlen_k)')
+        bias = bias.expand(batch, nheads, seqlen_q, seqlen_k)
+    bias_strides = (bias.stride(0), bias.stride(1), bias.stride(2)) if has_bias else (0, 0, 0)
+    grid = lambda META: (triton.cdiv(seqlen_k, META['BLOCK_N']) if META['SEQUENCE_PARALLEL'] else 1, batch * nheads)
+    _bwd_kernel[grid](q, k, v, bias, do, dq_accum, dk, dv, lse, delta, softmax_scale, q.stride(0), q.stride(2), q.stride(1), k.stride(0), k.stride(2), k.stride(1), v.stride(0), v.stride(2), v.stride(1), *bias_strides, do.stride(0), do.stride(2), do.stride(1), dq_accum.stride(0), dq_accum.stride(2), dq_accum.stride(1), dk.stride(0), dk.stride(2), dk.stride(1), dv.stride(0), dv.stride(2), dv.stride(1), nheads, seqlen_q, seqlen_k, seqlen_q_rounded, d, seqlen_q // 32, seqlen_k // 32, bias_type, causal, BLOCK_HEADDIM)
+    dq.copy_(dq_accum)
+
+class FlashAttnQKVPackedFunc(torch.autograd.Function):
+
+    @staticmethod
+    def forward(ctx, qkv, bias=None, causal=False, softmax_scale=None):
+        """
+            qkv: (batch, seqlen, 3, nheads, headdim)
+            bias: optional, shape broadcastible to (batch, nheads, seqlen, seqlen).
+                For example, ALiBi mask for causal would have shape (1, nheads, 1, seqlen).
+                ALiBi mask for non-causal would have shape (1, nheads, seqlen, seqlen)
+        """
+        if qkv.stride(-1) != 1:
+            qkv = qkv.contiguous()
+        (o, lse, ctx.softmax_scale) = _flash_attn_forward(qkv[:, :, 0], qkv[:, :, 1], qkv[:, :, 2], bias=bias, causal=causal, softmax_scale=softmax_scale)
+        ctx.save_for_backward(qkv, o, lse, bias)
+        ctx.causal = causal
+        return o
+
+    @staticmethod
+    def backward(ctx, do):
+        (qkv, o, lse, bias) = ctx.saved_tensors
+        assert not ctx.needs_input_grad[1], 'FlashAttention does not support bias gradient yet'
+        with torch.inference_mode():
+            dqkv = torch.empty_like(qkv)
+            _flash_attn_backward(do, qkv[:, :, 0], qkv[:, :, 1], qkv[:, :, 2], o, lse, dqkv[:, :, 0], dqkv[:, :, 1], dqkv[:, :, 2], bias=bias, causal=ctx.causal, softmax_scale=ctx.softmax_scale)
+        return (dqkv, None, None, None)
+flash_attn_qkvpacked_func = FlashAttnQKVPackedFunc.apply
+
+class FlashAttnKVPackedFunc(torch.autograd.Function):
+
+    @staticmethod
+    def forward(ctx, q, kv, bias=None, causal=False, softmax_scale=None):
+        """
+            q: (batch, seqlen_q, nheads, headdim)
+            kv: (batch, seqlen_k, 2, nheads, headdim)
+            bias: optional, shape broadcastible to (batch, nheads, seqlen_q, seqlen_k).
+                For example, ALiBi mask for causal would have shape (1, nheads, 1, seqlen_k).
+                ALiBi mask for non-causal would have shape (1, nheads, seqlen_q, seqlen_k)
+        """
+        (q, kv) = [x if x.stride(-1) == 1 else x.contiguous() for x in [q, kv]]
+        (o, lse, ctx.softmax_scale) = _flash_attn_forward(q, kv[:, :, 0], kv[:, :, 1], bias=bias, causal=causal, softmax_scale=softmax_scale)
+        ctx.save_for_backward(q, kv, o, lse, bias)
+        ctx.causal = causal
+        return o
+
+    @staticmethod
+    def backward(ctx, do):
+        (q, kv, o, lse, bias) = ctx.saved_tensors
+        if len(ctx.needs_input_grad) >= 3:
+            assert not ctx.needs_input_grad[2], 'FlashAttention does not support bias gradient yet'
+        with torch.inference_mode():
+            dq = torch.empty_like(q)
+            dkv = torch.empty_like(kv)
+            _flash_attn_backward(do, q, kv[:, :, 0], kv[:, :, 1], o, lse, dq, dkv[:, :, 0], dkv[:, :, 1], bias=bias, causal=ctx.causal, softmax_scale=ctx.softmax_scale)
+        return (dq, dkv, None, None, None)
+flash_attn_kvpacked_func = FlashAttnKVPackedFunc.apply
+
+class FlashAttnFunc(torch.autograd.Function):
+
+    @staticmethod
+    def forward(ctx, q, k, v, bias=None, causal=False, softmax_scale=None):
+        """
+            q: (batch_size, seqlen_q, nheads, headdim)
+            k, v: (batch_size, seqlen_k, nheads, headdim)
+            bias: optional, shape broadcastible to (batch, nheads, seqlen_q, seqlen_k).
+                For example, ALiBi mask for causal would have shape (1, nheads, 1, seqlen_k).
+                ALiBi mask for non-causal would have shape (1, nheads, seqlen_q, seqlen_k)
+        """
+        (q, k, v) = [x if x.stride(-1) == 1 else x.contiguous() for x in [q, k, v]]
+        (o, lse, ctx.softmax_scale) = _flash_attn_forward(q, k, v, bias=bias, causal=causal, softmax_scale=softmax_scale)
+        ctx.save_for_backward(q, k, v, o, lse, bias)
+        ctx.causal = causal
+        return o
+
+    @staticmethod
+    def backward(ctx, do):
+        (q, k, v, o, lse, bias) = ctx.saved_tensors
+        assert not ctx.needs_input_grad[3], 'FlashAttention does not support bias gradient yet'
+        with torch.inference_mode():
+            dq = torch.empty_like(q)
+            dk = torch.empty_like(k)
+            dv = torch.empty_like(v)
+            _flash_attn_backward(do, q, k, v, o, lse, dq, dk, dv, bias=bias, causal=ctx.causal, softmax_scale=ctx.softmax_scale)
+        return (dq, dk, dv, None, None, None)
+flash_attn_func = FlashAttnFunc.apply

generation_config.json

@@ -0,0 +1,5 @@
+{
+  "_from_model_config": true,
+  "transformers_version": "4.28.1",
+  "use_cache": false
+}

hf_prefixlm_converter.py

@@ -0,0 +1,415 @@
+"""Converts Huggingface Causal LM to Prefix LM.
+
+Conversion does lightweight surgery on a HuggingFace
+Causal LM to convert it to a Prefix LM.
+
+Prefix LMs accepts a `bidirectional_mask` input in `forward`
+and treat the input prompt as the prefix in `generate`.
+"""
+import math
+import warnings
+from types import MethodType
+from typing import Any, Dict, List, Optional, Tuple, Union
+import torch
+from transformers.models.bloom.modeling_bloom import BaseModelOutputWithPastAndCrossAttentions, BloomForCausalLM, BloomModel, CausalLMOutputWithCrossAttentions, CrossEntropyLoss
+from transformers.models.bloom.modeling_bloom import _expand_mask as _expand_mask_bloom
+from transformers.models.bloom.modeling_bloom import _make_causal_mask as _make_causal_mask_bloom
+from transformers.models.bloom.modeling_bloom import logging
+from transformers.models.gpt2.modeling_gpt2 import GPT2LMHeadModel
+from transformers.models.gpt_neo.modeling_gpt_neo import GPTNeoForCausalLM
+from transformers.models.gpt_neox.modeling_gpt_neox import GPTNeoXForCausalLM
+from transformers.models.gptj.modeling_gptj import GPTJForCausalLM
+from transformers.models.opt.modeling_opt import OPTForCausalLM
+from transformers.models.opt.modeling_opt import _expand_mask as _expand_mask_opt
+from transformers.models.opt.modeling_opt import _make_causal_mask as _make_causal_mask_opt
+logger = logging.get_logger(__name__)
+_SUPPORTED_GPT_MODELS = (GPT2LMHeadModel, GPTJForCausalLM, GPTNeoForCausalLM, GPTNeoXForCausalLM)
+CAUSAL_GPT_TYPES = Union[GPT2LMHeadModel, GPTJForCausalLM, GPTNeoForCausalLM, GPTNeoXForCausalLM]
+
+def _convert_gpt_causal_lm_to_prefix_lm(model: CAUSAL_GPT_TYPES) -> CAUSAL_GPT_TYPES:
+    """Converts a GPT-style Causal LM to a Prefix LM.
+
+    Supported HuggingFace model classes:
+        - `GPT2LMHeadModel`
+        - `GPTNeoForCausalLM`
+        - `GPTNeoXForCausalLM`
+        - `GPTJForCausalLM`
+
+    See `convert_hf_causal_lm_to_prefix_lm` for more details.
+    """
+    if hasattr(model, '_prefix_lm_converted'):
+        return model
+    assert isinstance(model, _SUPPORTED_GPT_MODELS)
+    assert model.config.add_cross_attention == False, 'Only supports GPT-style decoder-only models'
+
+    def _get_attn_modules(model: CAUSAL_GPT_TYPES) -> List[torch.nn.Module]:
+        """Helper that gets a list of the model's attention modules.
+
+        Each module has a `bias` buffer used for causal masking. The Prefix LM
+        conversion adds logic to dynamically manipulate these biases to support
+        Prefix LM attention masking.
+        """
+        attn_modules = []
+        if isinstance(model, GPTNeoXForCausalLM):
+            blocks = model.gpt_neox.layers
+        else:
+            blocks = model.transformer.h
+        for block in blocks:
+            if isinstance(model, GPTNeoForCausalLM):
+                if block.attn.attention_type != 'global':
+                    continue
+                attn_module = block.attn.attention
+            elif isinstance(model, GPTNeoXForCausalLM):
+                attn_module = block.attention
+            else:
+                attn_module = block.attn
+            attn_modules.append(attn_module)
+        return attn_modules
+    setattr(model, '_original_forward', getattr(model, 'forward'))
+    setattr(model, '_original_generate', getattr(model, 'generate'))
+
+    def forward(self: CAUSAL_GPT_TYPES, input_ids: Optional[torch.LongTensor]=None, past_key_values: Optional[Tuple[Tuple[torch.Tensor]]]=None, attention_mask: Optional[torch.FloatTensor]=None, bidirectional_mask: Optional[torch.Tensor]=None, token_type_ids: Optional[torch.LongTensor]=None, position_ids: Optional[torch.LongTensor]=None, head_mask: Optional[torch.FloatTensor]=None, inputs_embeds: Optional[torch.FloatTensor]=None, labels: Optional[torch.LongTensor]=None, use_cache: Optional[bool]=None, output_attentions: Optional[bool]=None, output_hidden_states: Optional[bool]=None, return_dict: Optional[bool]=None):
+        """Wraps original forward to enable PrefixLM attention."""
+
+        def call_og_forward():
+            if isinstance(self, GPTNeoXForCausalLM):
+                return self._original_forward(input_ids=input_ids, past_key_values=past_key_values, attention_mask=attention_mask, head_mask=head_mask, inputs_embeds=inputs_embeds, labels=labels, use_cache=use_cache, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict)
+            else:
+                return self._original_forward(input_ids=input_ids, past_key_values=past_key_values, attention_mask=attention_mask, token_type_ids=token_type_ids, position_ids=position_ids, head_mask=head_mask, inputs_embeds=inputs_embeds, labels=labels, use_cache=use_cache, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict)
+        if bidirectional_mask is None:
+            return call_og_forward()
+        assert isinstance(bidirectional_mask, torch.Tensor)
+        attn_modules = _get_attn_modules(model)
+        (b, s) = bidirectional_mask.shape
+        max_length = attn_modules[0].bias.shape[-1]
+        if s > max_length:
+            raise ValueError(f'bidirectional_mask sequence length (={s}) exceeds the ' + f'max length allowed by the model ({max_length}).')
+        assert s <= max_length
+        if s < max_length:
+            pad = torch.zeros((int(b), int(max_length - s)), dtype=bidirectional_mask.dtype, device=bidirectional_mask.device)
+            bidirectional_mask = torch.cat([bidirectional_mask, pad], dim=1)
+        bidirectional = bidirectional_mask.unsqueeze(1).unsqueeze(1)
+        for attn_module in attn_modules:
+            attn_module.bias.data = torch.logical_or(attn_module.bias.data, bidirectional)
+        output = call_og_forward()
+        for attn_module in attn_modules:
+            attn_module.bias.data = torch.tril(attn_module.bias.data[0, 0])[None, None]
+        return output
+
+    def generate(self: CAUSAL_GPT_TYPES, *args: tuple, **kwargs: Dict[str, Any]):
+        """Wraps original generate to enable PrefixLM attention."""
+        attn_modules = _get_attn_modules(model)
+        for attn_module in attn_modules:
+            attn_module.bias.data[:] = 1
+        output = self._original_generate(*args, **kwargs)
+        for attn_module in attn_modules:
+            attn_module.bias.data = torch.tril(attn_module.bias.data[0, 0])[None, None]
+        return output
+    setattr(model, 'forward', MethodType(forward, model))
+    setattr(model, 'generate', MethodType(generate, model))
+    setattr(model, '_prefix_lm_converted', True)
+    return model
+
+def _convert_bloom_causal_lm_to_prefix_lm(model: BloomForCausalLM) -> BloomForCausalLM:
+    """Converts a BLOOM Causal LM to a Prefix LM.
+
+    Supported HuggingFace model classes:
+        - `BloomForCausalLM`
+
+    See `convert_hf_causal_lm_to_prefix_lm` for more details.
+    """
+    if hasattr(model, '_prefix_lm_converted'):
+        return model
+    assert isinstance(model, BloomForCausalLM)
+    assert model.config.add_cross_attention == False, 'Only supports BLOOM decoder-only models'
+
+    def _prepare_attn_mask(self: BloomModel, attention_mask: torch.Tensor, bidirectional_mask: Optional[torch.Tensor], input_shape: Tuple[int, int], past_key_values_length: int) -> torch.BoolTensor:
+        combined_attention_mask = None
+        device = attention_mask.device
+        (_, src_length) = input_shape
+        if src_length > 1:
+            combined_attention_mask = _make_causal_mask_bloom(input_shape, device=device, past_key_values_length=past_key_values_length)
+            if bidirectional_mask is not None:
+                assert attention_mask.shape == bidirectional_mask.shape
+                expanded_bidirectional_mask = _expand_mask_bloom(bidirectional_mask, tgt_length=src_length)
+                combined_attention_mask = torch.logical_and(combined_attention_mask, expanded_bidirectional_mask)
+        expanded_attn_mask = _expand_mask_bloom(attention_mask, tgt_length=src_length)
+        combined_attention_mask = expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask | combined_attention_mask
+        return combined_attention_mask
+
+    def _build_alibi_tensor(self: BloomModel, batch_size: int, query_length: int, key_length: int, dtype: torch.dtype, device: torch.device) -> torch.Tensor:
+        num_heads = self.config.n_head
+        closest_power_of_2 = 2 ** math.floor(math.log2(num_heads))
+        base = torch.tensor(2 ** (-2 ** (-(math.log2(closest_power_of_2) - 3))), device=device, dtype=torch.float32)
+        powers = torch.arange(1, 1 + closest_power_of_2, device=device, dtype=torch.int32)
+        slopes = torch.pow(base, powers)
+        if closest_power_of_2 != num_heads:
+            extra_base = torch.tensor(2 ** (-2 ** (-(math.log2(2 * closest_power_of_2) - 3))), device=device, dtype=torch.float32)
+            num_remaining_heads = min(closest_power_of_2, num_heads - closest_power_of_2)
+            extra_powers = torch.arange(1, 1 + 2 * num_remaining_heads, 2, device=device, dtype=torch.int32)
+            slopes = torch.cat([slopes, torch.pow(extra_base, extra_powers)], dim=0)
+        qa = torch.arange(query_length, device=device, dtype=torch.int32).view(-1, 1)
+        ka = torch.arange(key_length, device=device, dtype=torch.int32).view(1, -1)
+        diffs = qa - ka + key_length - query_length
+        diffs = -diffs.abs()
+        alibi = slopes.view(1, num_heads, 1, 1) * diffs.view(1, 1, query_length, key_length)
+        alibi = alibi.expand(batch_size, -1, -1, -1).reshape(-1, query_length, key_length)
+        return alibi.to(dtype)
+    KeyValueT = Tuple[torch.Tensor, torch.Tensor]
+
+    def forward(self: BloomModel, input_ids: Optional[torch.LongTensor]=None, past_key_values: Optional[Tuple[KeyValueT, ...]]=None, attention_mask: Optional[torch.Tensor]=None, bidirectional_mask: Optional[torch.Tensor]=None, head_mask: Optional[torch.LongTensor]=None, inputs_embeds: Optional[torch.LongTensor]=None, use_cache: Optional[bool]=None, output_attentions: Optional[bool]=None, output_hidden_states: Optional[bool]=None, return_dict: Optional[bool]=None, **deprecated_arguments) -> Union[Tuple[torch.Tensor, ...], BaseModelOutputWithPastAndCrossAttentions]:
+        if deprecated_arguments.pop('position_ids', False) is not False:
+            warnings.warn('`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. ' + 'You can safely ignore passing `position_ids`.', FutureWarning)
+        if len(deprecated_arguments) > 0:
+            raise ValueError(f'Got unexpected arguments: {deprecated_arguments}')
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError('You cannot specify both input_ids and inputs_embeds at the same time')
+        elif input_ids is not None:
+            (batch_size, seq_length) = input_ids.shape
+        elif inputs_embeds is not None:
+            (batch_size, seq_length, _) = inputs_embeds.shape
+        else:
+            raise ValueError('You have to specify either input_ids or inputs_embeds')
+        if past_key_values is None:
+            past_key_values = tuple([None] * len(self.h))
+        head_mask = self.get_head_mask(head_mask, self.config.n_layer)
+        if inputs_embeds is None:
+            inputs_embeds = self.word_embeddings(input_ids)
+        hidden_states = self.word_embeddings_layernorm(inputs_embeds)
+        presents = () if use_cache else None
+        all_self_attentions = () if output_attentions else None
+        all_hidden_states = () if output_hidden_states else None
+        seq_length_with_past = seq_length
+        past_key_values_length = 0
+        if past_key_values[0] is not None:
+            tmp = past_key_values[0][0]
+            past_key_values_length = tmp.shape[2]
+            seq_length_with_past = seq_length_with_past + past_key_values_length
+        if attention_mask is None:
+            attention_mask = torch.ones((batch_size, seq_length_with_past), device=hidden_states.device)
+        else:
+            attention_mask = attention_mask.to(hidden_states.device)
+        alibi = self._build_alibi_tensor(batch_size=batch_size, query_length=seq_length, key_length=seq_length_with_past, dtype=hidden_states.dtype, device=hidden_states.device)
+        causal_mask = self._prepare_attn_mask(attention_mask, bidirectional_mask, input_shape=(batch_size, seq_length), past_key_values_length=past_key_values_length)
+        for (i, (block, layer_past)) in enumerate(zip(self.h, past_key_values)):
+            if output_hidden_states:
+                hst = (hidden_states,)
+                all_hidden_states = all_hidden_states + hst
+            if self.gradient_checkpointing and self.training:
+                if use_cache:
+                    logger.warning('`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...')
+                    use_cache = False
+
+                def create_custom_forward(module):
+
+                    def custom_forward(*inputs):
+                        return module(*inputs, use_cache=use_cache, output_attentions=output_attentions)
+                    return custom_forward
+                outputs = torch.utils.checkpoint.checkpoint(create_custom_forward(block), hidden_states, alibi, causal_mask, head_mask[i])
+            else:
+                outputs = block(hidden_states, layer_past=layer_past, attention_mask=causal_mask, head_mask=head_mask[i], use_cache=use_cache, output_attentions=output_attentions, alibi=alibi)
+            hidden_states = outputs[0]
+            if use_cache is True:
+                presents = presents + (outputs[1],)
+            if output_attentions:
+                oa = (outputs[2 if use_cache else 1],)
+                all_self_attentions = all_self_attentions + oa
+        hidden_states = self.ln_f(hidden_states)
+        if output_hidden_states:
+            hst = (hidden_states,)
+            all_hidden_states = all_hidden_states + hst
+        if not return_dict:
+            return tuple((v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None))
+        return BaseModelOutputWithPastAndCrossAttentions(last_hidden_state=hidden_states, past_key_values=presents, hidden_states=all_hidden_states, attentions=all_self_attentions)
+    setattr(model.transformer, '_prepare_attn_mask', MethodType(_prepare_attn_mask, model.transformer))
+    setattr(model.transformer, '_build_alibi_tensor', MethodType(_build_alibi_tensor, model.transformer))
+    setattr(model.transformer, 'forward', MethodType(forward, model.transformer))
+    KeyValueT = Tuple[torch.Tensor, torch.Tensor]
+
+    def forward(self: BloomForCausalLM, input_ids: Optional[torch.LongTensor]=None, past_key_values: Optional[Tuple[KeyValueT, ...]]=None, attention_mask: Optional[torch.Tensor]=None, bidirectional_mask: Optional[torch.Tensor]=None, head_mask: Optional[torch.Tensor]=None, inputs_embeds: Optional[torch.Tensor]=None, labels: Optional[torch.Tensor]=None, use_cache: Optional[bool]=None, output_attentions: Optional[bool]=None, output_hidden_states: Optional[bool]=None, return_dict: Optional[bool]=None, **deprecated_arguments) -> Union[Tuple[torch.Tensor], CausalLMOutputWithCrossAttentions]:
+        """Replacement forward method for BloomCausalLM."""
+        if deprecated_arguments.pop('position_ids', False) is not False:
+            warnings.warn('`position_ids` have no functionality in BLOOM and will be removed ' + 'in v5.0.0. You can safely ignore passing `position_ids`.', FutureWarning)
+        if len(deprecated_arguments) > 0:
+            raise ValueError(f'Got unexpected arguments: {deprecated_arguments}')
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        transformer_outputs = self.transformer(input_ids, past_key_values=past_key_values, attention_mask=attention_mask, bidirectional_mask=bidirectional_mask, head_mask=head_mask, inputs_embeds=inputs_embeds, use_cache=use_cache, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict)
+        hidden_states = transformer_outputs[0]
+        lm_logits = self.lm_head(hidden_states)
+        loss = None
+        if labels is not None:
+            shift_logits = lm_logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            (batch_size, seq_length, vocab_size) = shift_logits.shape
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(batch_size * seq_length, vocab_size), shift_labels.view(batch_size * seq_length))
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return (loss,) + output if loss is not None else output
+        return CausalLMOutputWithCrossAttentions(loss=loss, logits=lm_logits, past_key_values=transformer_outputs.past_key_values, hidden_states=transformer_outputs.hidden_states, attentions=transformer_outputs.attentions)
+
+    def prepare_inputs_for_generation(self: BloomForCausalLM, input_ids: torch.LongTensor, past: Optional[torch.Tensor]=None, attention_mask: Optional[torch.Tensor]=None, **kwargs) -> dict:
+        if past:
+            input_ids = input_ids[:, -1].unsqueeze(-1)
+            bidirectional_mask = None
+            if past[0][0].shape[0] == input_ids.shape[0]:
+                past = self._convert_to_bloom_cache(past)
+        else:
+            bidirectional_mask = torch.ones_like(input_ids)
+        return {'input_ids': input_ids, 'past_key_values': past, 'use_cache': True, 'attention_mask': attention_mask, 'bidirectional_mask': bidirectional_mask}
+    setattr(model, 'forward', MethodType(forward, model))
+    setattr(model, 'prepare_inputs_for_generation', MethodType(prepare_inputs_for_generation, model))
+    setattr(model, '_prefix_lm_converted', True)
+    return model
+
+def _convert_opt_causal_lm_to_prefix_lm(model: OPTForCausalLM) -> OPTForCausalLM:
+    """Converts an OPT Causal LM to a Prefix LM.
+
+    Supported HuggingFace model classes:
+        - `OPTForCausalLM`
+
+    See `convert_hf_causal_lm_to_prefix_lm` for more details.
+    """
+    if hasattr(model, '_prefix_lm_converted'):
+        return model
+    assert isinstance(model, OPTForCausalLM)
+    assert model.config.add_cross_attention == False, 'Only supports OPT decoder-only models'
+    setattr(model, '_original_forward', getattr(model, 'forward'))
+    setattr(model, '_original_generate', getattr(model, 'generate'))
+    model.model.decoder.bidirectional_mask = None
+
+    def _prepare_decoder_attention_mask(self, attention_mask, input_shape, inputs_embeds, past_key_values_length):
+        combined_attention_mask = None
+        if input_shape[-1] > 1:
+            if self.bidirectional_mask == 'g':
+                (bsz, src_length) = input_shape
+                combined_attention_mask = torch.zeros((bsz, 1, src_length, src_length + past_key_values_length), dtype=inputs_embeds.dtype, device=inputs_embeds.device)
+            else:
+                combined_attention_mask = _make_causal_mask_opt(input_shape, inputs_embeds.dtype, past_key_values_length=past_key_values_length).to(inputs_embeds.device)
+                if self.bidirectional_mask is not None:
+                    assert attention_mask.shape == self.bidirectional_mask.shape
+                    expanded_bidirectional_mask = _expand_mask_opt(self.bidirectional_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(inputs_embeds.device)
+                    combined_attention_mask = torch.maximum(expanded_bidirectional_mask, combined_attention_mask)
+        if attention_mask is not None:
+            expanded_attn_mask = _expand_mask_opt(attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(inputs_embeds.device)
+            combined_attention_mask = expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
+        return combined_attention_mask
+    setattr(model.model.decoder, '_prepare_decoder_attention_mask', MethodType(_prepare_decoder_attention_mask, model.model.decoder))
+
+    def forward(self: OPTForCausalLM, input_ids: Optional[torch.LongTensor]=None, attention_mask: Optional[torch.Tensor]=None, bidirectional_mask: Optional[torch.ByteTensor]=None, head_mask: Optional[torch.Tensor]=None, past_key_values: Optional[List[torch.FloatTensor]]=None, inputs_embeds: Optional[torch.FloatTensor]=None, labels: Optional[torch.LongTensor]=None, use_cache: Optional[bool]=None, output_attentions: Optional[bool]=None, output_hidden_states: Optional[bool]=None, return_dict: Optional[bool]=None):
+
+        def call_og_forward():
+            return self._original_forward(input_ids=input_ids, attention_mask=attention_mask, head_mask=head_mask, past_key_values=past_key_values, inputs_embeds=inputs_embeds, labels=labels, use_cache=use_cache, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict)
+        if bidirectional_mask is None:
+            return call_og_forward()
+        self.model.decoder.bidirectional_mask = bidirectional_mask
+        try:
+            outputs = call_og_forward()
+        except:
+            self.model.decoder.bidirectional_mask = None
+            raise
+        self.model.decoder.bidirectional_mask = None
+        return outputs
+
+    def generate(self: OPTForCausalLM, *args: tuple, **kwargs: Dict[str, Any]):
+        """Wraps original generate to enable PrefixLM-style attention."""
+        self.model.decoder.bidirectional_mask = 'g'
+        try:
+            output = self._original_generate(*args, **kwargs)
+        except:
+            self.model.decoder.bidirectional_mask = None
+            raise
+        self.model.decoder.bidirectional_mask = None
+        return output
+    setattr(model, 'forward', MethodType(forward, model))
+    setattr(model, 'generate', MethodType(generate, model))
+    setattr(model, '_prefix_lm_converted', True)
+    return model
+_SUPPORTED_HF_MODELS = _SUPPORTED_GPT_MODELS + (BloomForCausalLM, OPTForCausalLM)
+CAUSAL_LM_TYPES = Union[GPT2LMHeadModel, GPTJForCausalLM, GPTNeoForCausalLM, GPTNeoXForCausalLM, BloomForCausalLM, OPTForCausalLM]
+
+def convert_hf_causal_lm_to_prefix_lm(model: CAUSAL_LM_TYPES) -> CAUSAL_LM_TYPES:
+    """Converts a HuggingFace Causal LM to a Prefix LM.
+
+    Supported HuggingFace model classes:
+        - `GPT2LMHeadModel`
+        - `GPTNeoForCausalLM`
+        - `GPTNeoXForCausalLM`
+        - `GPTJForCausalLM`
+        - `BloomForCausalLM`
+        - `OPTForCausalLM`
+
+    Conversion to a Prefix LM is done by modifying the `forward` method, and possibly also the
+    `generate` method and/or select underlying methods depending on the model class.
+
+    These changes preserve the model API, but add a new input to `forward`: "bidirectional_mask".
+
+    Notes on training:
+        To actually train the converted model as a Prefix LM, training batches will need to indicate
+        the prefix/target structure by including `bidirectional_mask` as part of the batch inputs.
+
+        **This is not a standard input and requires custom layers either within or after your dataloader.**
+
+        In addition to adding `bidirectional_mask` to the batch, this custom code should modify `labels`
+        such that `batch['labels'][batch['bidirectional_mask'] == 1] == -100`.
+        That is, the prefix portion of the sequence should not generate any loss. Loss should only be
+        generated by the target portion of the sequence.
+
+    Notes on `GPTNeoForCausalLM`:
+        To simplify the implementation, "global" and "local" attention layers are handled differently.
+        For "global" layers, we handle conversion as described above. For "local" layers, which use a
+        causal attention mask within a restricted local window, we do not alter the masking.
+
+    Notes on `forward` method conversion:
+        After conversion, the `forward` method will handle a new input, `bidirectional_mask`,
+        which should be a [batch_size, seq_length] byte tensor, where 1 indicates token positions
+        belonging to the prefix (prefix tokens can attend to one another bidirectionally), and
+        0 indicates token positions belonging to the target.
+
+        The new `forward` method will incorporate `bidirectional_mask` (if supplied) into the existing
+        causal mask, call the original `forward` method, and (if the causal mask is a buffer) reset
+        the causal masks before returning the result.
+
+    Notes on `generate` method conversion:
+        After conversion, the `generate` method will have the same signature but will internally
+        convert all causal masks to be purely bidirectional, call the original `generate` method, and
+        (where appropriate) reset the causal masks before returning the result.
+
+        This works thanks to the logic of the HuggingFace `generate` API, which first encodes the token
+        "prompt" passed to `generate` (which is treated as the prefix) and then sequentially generates
+        each new token. Encodings are cached as generation happens, so all prefix tokens can attend to one
+        another (as expected in a Prefix LM) and generated tokens can only attend to prefix tokens and
+        previously-generated tokens (also as expected in a Prefix LM).
+
+    To preserve the API, the original methods are renamed to `_original_forward` and
+    `_original_generate`, and replaced with new `forward` and `generate` methods that wrap
+    them, respectively. Although implementation details vary by model class.
+    """
+    if isinstance(model, _SUPPORTED_GPT_MODELS):
+        return _convert_gpt_causal_lm_to_prefix_lm(model)
+    elif isinstance(model, BloomForCausalLM):
+        return _convert_bloom_causal_lm_to_prefix_lm(model)
+    elif isinstance(model, OPTForCausalLM):
+        return _convert_opt_causal_lm_to_prefix_lm(model)
+    else:
+        raise TypeError(f'Cannot convert model to Prefix LM. ' + f'Model does not belong to set of supported HF models:' + f'\n{_SUPPORTED_HF_MODELS}')
+
+def add_bidirectional_mask_if_missing(batch: Dict[str, Any]):
+    """Attempts to add bidirectional_mask to batch if missing.
+
+    Raises:
+        KeyError if bidirectional_mask is missing and can't be inferred
+    """
+    if 'bidirectional_mask' not in batch:
+        if batch.get('mode', None) == 'icl_task':
+            batch['bidirectional_mask'] = batch['attention_mask'].clone()
+            for (i, continuation_indices) in enumerate(batch['continuation_indices']):
+                batch['bidirectional_mask'][i, continuation_indices] = 0
+        elif 'labels' in batch and 'attention_mask' in batch:
+            batch['bidirectional_mask'] = torch.logical_and(torch.eq(batch['attention_mask'], 1), torch.eq(batch['labels'], -100)).type_as(batch['attention_mask'])
+        else:
+            raise KeyError('No bidirectional_mask in batch and not sure how to construct one.')

meta_init_context.py

@@ -0,0 +1,94 @@
+from contextlib import contextmanager
+import torch
+import torch.nn as nn
+
+@contextmanager
+def init_empty_weights(include_buffers: bool=False):
+    """Meta initialization context manager.
+
+    A context manager under which models are initialized with all parameters
+    on the meta device, therefore creating an empty model. Useful when just
+    initializing the model would blow the available RAM.
+
+    Args:
+        include_buffers (`bool`, *optional*, defaults to `False`): Whether or
+            not to also put all buffers on the meta device while initializing.
+
+    Example:
+    ```python
+    import torch.nn as nn
+
+    # Initialize a model with 100 billions parameters in no time and without using any RAM.
+    with init_empty_weights():
+        tst = nn.Sequential(*[nn.Linear(10000, 10000) for _ in range(1000)])
+    ```
+
+    <Tip warning={true}>
+
+    Any model created under this context manager has no weights. As such you can't do something like
+    `model.to(some_device)` with it. To load weights inside your empty model, see [`load_checkpoint_and_dispatch`].
+
+    </Tip>
+    """
+    with init_on_device(torch.device('meta'), include_buffers=include_buffers) as f:
+        yield f
+
+@contextmanager
+def init_on_device(device: torch.device, include_buffers: bool=False):
+    """Device initialization context manager.
+
+    A context manager under which models are initialized with all parameters
+    on the specified device.
+
+    Args:
+        device (`torch.device`): Device to initialize all parameters on.
+        include_buffers (`bool`, *optional*, defaults to `False`): Whether or
+            not to also put all buffers on the meta device while initializing.
+
+    Example:
+    ```python
+    import torch.nn as nn
+
+    with init_on_device(device=torch.device("cuda")):
+        tst = nn.Liner(100, 100)  # on `cuda` device
+    ```
+    """
+    old_register_parameter = nn.Module.register_parameter
+    if include_buffers:
+        old_register_buffer = nn.Module.register_buffer
+
+    def register_empty_parameter(module, name, param):
+        old_register_parameter(module, name, param)
+        if param is not None:
+            param_cls = type(module._parameters[name])
+            kwargs = module._parameters[name].__dict__
+            module._parameters[name] = param_cls(module._parameters[name].to(device), **kwargs)
+
+    def register_empty_buffer(module, name, buffer):
+        old_register_buffer(module, name, buffer)
+        if buffer is not None:
+            module._buffers[name] = module._buffers[name].to(device)
+    if include_buffers:
+        tensor_constructors_to_patch = {torch_function_name: getattr(torch, torch_function_name) for torch_function_name in ['empty', 'zeros', 'ones', 'full']}
+    else:
+        tensor_constructors_to_patch = {}
+
+    def patch_tensor_constructor(fn):
+
+        def wrapper(*args, **kwargs):
+            kwargs['device'] = device
+            return fn(*args, **kwargs)
+        return wrapper
+    try:
+        nn.Module.register_parameter = register_empty_parameter
+        if include_buffers:
+            nn.Module.register_buffer = register_empty_buffer
+        for torch_function_name in tensor_constructors_to_patch.keys():
+            setattr(torch, torch_function_name, patch_tensor_constructor(getattr(torch, torch_function_name)))
+        yield
+    finally:
+        nn.Module.register_parameter = old_register_parameter
+        if include_buffers:
+            nn.Module.register_buffer = old_register_buffer
+        for (torch_function_name, old_torch_function) in tensor_constructors_to_patch.items():
+            setattr(torch, torch_function_name, old_torch_function)

model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:411576f8c03aa73bc7faa2d241ef5090e16abc73583c610f592dd36798c4b198
+size 13298599938

norm.py

@@ -0,0 +1,56 @@
+import torch
+
+def _cast_if_autocast_enabled(tensor):
+    if torch.is_autocast_enabled():
+        if tensor.device.type == 'cuda':
+            dtype = torch.get_autocast_gpu_dtype()
+        elif tensor.device.type == 'cpu':
+            dtype = torch.get_autocast_cpu_dtype()
+        else:
+            raise NotImplementedError()
+        return tensor.to(dtype=dtype)
+    return tensor
+
+class LPLayerNorm(torch.nn.LayerNorm):
+
+    def __init__(self, normalized_shape, eps=1e-05, elementwise_affine=True, device=None, dtype=None):
+        super().__init__(normalized_shape=normalized_shape, eps=eps, elementwise_affine=elementwise_affine, device=device, dtype=dtype)
+
+    def forward(self, x):
+        module_device = x.device
+        downcast_x = _cast_if_autocast_enabled(x)
+        downcast_weight = _cast_if_autocast_enabled(self.weight) if self.weight is not None else self.weight
+        downcast_bias = _cast_if_autocast_enabled(self.bias) if self.bias is not None else self.bias
+        with torch.autocast(enabled=False, device_type=module_device.type):
+            return torch.nn.functional.layer_norm(downcast_x, self.normalized_shape, downcast_weight, downcast_bias, self.eps)
+
+def rms_norm(x, weight=None, eps=1e-05):
+    output = x / torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + eps)
+    if weight is not None:
+        return output * weight
+    return output
+
+class RMSNorm(torch.nn.Module):
+
+    def __init__(self, normalized_shape, eps=1e-05, weight=True, dtype=None, device=None):
+        super().__init__()
+        self.eps = eps
+        if weight:
+            self.weight = torch.nn.Parameter(torch.ones(normalized_shape, dtype=dtype, device=device))
+        else:
+            self.register_parameter('weight', None)
+
+    def forward(self, x):
+        return rms_norm(x.float(), self.weight, self.eps).to(dtype=x.dtype)
+
+class LPRMSNorm(RMSNorm):
+
+    def __init__(self, normalized_shape, eps=1e-05, weight=True, dtype=None, device=None):
+        super().__init__(normalized_shape=normalized_shape, eps=eps, weight=weight, dtype=dtype, device=device)
+
+    def forward(self, x):
+        downcast_x = _cast_if_autocast_enabled(x)
+        downcast_weight = _cast_if_autocast_enabled(self.weight) if self.weight is not None else self.weight
+        with torch.autocast(enabled=False, device_type=x.device.type):
+            return rms_norm(downcast_x, downcast_weight, self.eps).to(dtype=x.dtype)
+NORM_CLASS_REGISTRY = {'layernorm': torch.nn.LayerNorm, 'low_precision_layernorm': LPLayerNorm, 'rmsnorm': RMSNorm, 'low_precision_rmsnorm': LPRMSNorm}

param_init_fns.py

@@ -0,0 +1,181 @@
+import math
+import warnings
+from collections.abc import Sequence
+from functools import partial
+from typing import Optional, Tuple, Union
+import torch
+from torch import nn
+from .norm import NORM_CLASS_REGISTRY
+
+def torch_default_param_init_fn_(module: nn.Module, verbose: int=0, **kwargs):
+    del kwargs
+    if verbose > 1:
+        warnings.warn(f"Initializing network using module's reset_parameters attribute")
+    if hasattr(module, 'reset_parameters'):
+        module.reset_parameters()
+
+def fused_init_helper_(module: nn.Module, init_fn_):
+    _fused = getattr(module, '_fused', None)
+    if _fused is None:
+        raise RuntimeError(f'Internal logic error')
+    (dim, splits) = _fused
+    splits = (0, *splits, module.weight.size(dim))
+    for (s, e) in zip(splits[:-1], splits[1:]):
+        slice_indices = [slice(None)] * module.weight.ndim
+        slice_indices[dim] = slice(s, e)
+        init_fn_(module.weight[slice_indices])
+
+def generic_param_init_fn_(module: nn.Module, init_fn_, n_layers: int, d_model: Optional[int]=None, init_div_is_residual: Union[int, float, str, bool]=True, emb_init_std: Optional[float]=None, emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]]=None, verbose: int=0, **kwargs):
+    del kwargs
+    if verbose > 1:
+        warnings.warn(f'If model has bias parameters they are initialized to 0.')
+    init_div_is_residual = init_div_is_residual
+    if init_div_is_residual is False:
+        div_is_residual = 1.0
+    elif init_div_is_residual is True:
+        div_is_residual = math.sqrt(2 * n_layers)
+    elif isinstance(init_div_is_residual, float) or isinstance(init_div_is_residual, int):
+        div_is_residual = init_div_is_residual
+    elif isinstance(init_div_is_residual, str) and init_div_is_residual.isnumeric():
+        div_is_residual = float(init_div_is_residual)
+    else:
+        div_is_residual = 1.0
+        raise ValueError(f'Expected init_div_is_residual to be boolean or numeric, got {init_div_is_residual}')
+    if init_div_is_residual is not False:
+        if verbose > 1:
+            warnings.warn(f'Initializing _is_residual layers then dividing them by {div_is_residual:.3f}. ' + f'Set `init_div_is_residual: false` in init config to disable this.')
+    if isinstance(module, nn.Linear):
+        if hasattr(module, '_fused'):
+            fused_init_helper_(module, init_fn_)
+        else:
+            init_fn_(module.weight)
+        if module.bias is not None:
+            torch.nn.init.zeros_(module.bias)
+        if init_div_is_residual is not False and getattr(module, '_is_residual', False):
+            with torch.no_grad():
+                module.weight.div_(div_is_residual)
+    elif isinstance(module, nn.Embedding):
+        if emb_init_std is not None:
+            std = emb_init_std
+            if std == 0:
+                warnings.warn(f'Embedding layer initialized to 0.')
+            emb_init_fn_ = partial(torch.nn.init.normal_, mean=0.0, std=std)
+            if verbose > 1:
+                warnings.warn(f'Embedding layer initialized using normal distribution with mean=0 and std={std!r}.')
+        elif emb_init_uniform_lim is not None:
+            lim = emb_init_uniform_lim
+            if isinstance(lim, Sequence):
+                if len(lim) > 2:
+                    raise ValueError(f'Uniform init requires a min and a max limit. User input: {lim}.')
+                if lim[0] == lim[1]:
+                    warnings.warn(f'Embedding layer initialized to {lim[0]}.')
+            else:
+                if lim == 0:
+                    warnings.warn(f'Embedding layer initialized to 0.')
+                lim = [-lim, lim]
+            (a, b) = lim
+            emb_init_fn_ = partial(torch.nn.init.uniform_, a=a, b=b)
+            if verbose > 1:
+                warnings.warn(f'Embedding layer initialized using uniform distribution in range {lim}.')
+        else:
+            emb_init_fn_ = init_fn_
+        emb_init_fn_(module.weight)
+    elif isinstance(module, tuple(set(NORM_CLASS_REGISTRY.values()))):
+        if verbose > 1:
+            warnings.warn(f'Norm weights are set to 1. If norm layer has a bias it is initialized to 0.')
+        if hasattr(module, 'weight') and module.weight is not None:
+            torch.nn.init.ones_(module.weight)
+        if hasattr(module, 'bias') and module.bias is not None:
+            torch.nn.init.zeros_(module.bias)
+    elif isinstance(module, nn.MultiheadAttention):
+        if module._qkv_same_embed_dim:
+            assert module.in_proj_weight is not None
+            assert module.q_proj_weight is None and module.k_proj_weight is None and (module.v_proj_weight is None)
+            assert d_model is not None
+            _d = d_model
+            splits = (0, _d, 2 * _d, 3 * _d)
+            for (s, e) in zip(splits[:-1], splits[1:]):
+                init_fn_(module.in_proj_weight[s:e])
+        else:
+            assert module.q_proj_weight is not None and module.k_proj_weight is not None and (module.v_proj_weight is not None)
+            assert module.in_proj_weight is None
+            init_fn_(module.q_proj_weight)
+            init_fn_(module.k_proj_weight)
+            init_fn_(module.v_proj_weight)
+        if module.in_proj_bias is not None:
+            torch.nn.init.zeros_(module.in_proj_bias)
+        if module.bias_k is not None:
+            torch.nn.init.zeros_(module.bias_k)
+        if module.bias_v is not None:
+            torch.nn.init.zeros_(module.bias_v)
+        init_fn_(module.out_proj.weight)
+        if init_div_is_residual is not False and getattr(module.out_proj, '_is_residual', False):
+            with torch.no_grad():
+                module.out_proj.weight.div_(div_is_residual)
+        if module.out_proj.bias is not None:
+            torch.nn.init.zeros_(module.out_proj.bias)
+    else:
+        for _ in module.parameters(recurse=False):
+            raise NotImplementedError(f'{module.__class__.__name__} parameters are not initialized by param_init_fn.')
+
+def _normal_init_(std, mean=0.0):
+    return partial(torch.nn.init.normal_, mean=mean, std=std)
+
+def _normal_param_init_fn_(module: nn.Module, std: float, n_layers: int, d_model: Optional[int]=None, init_div_is_residual: Union[int, float, str, bool]=True, emb_init_std: Optional[float]=None, emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]]=None, verbose: int=0, **kwargs):
+    del kwargs
+    init_fn_ = _normal_init_(std=std)
+    if verbose > 1:
+        warnings.warn(f'Using torch.nn.init.normal_ init fn mean=0.0, std={std}')
+    generic_param_init_fn_(module=module, init_fn_=init_fn_, d_model=d_model, n_layers=n_layers, init_div_is_residual=init_div_is_residual, emb_init_std=emb_init_std, emb_init_uniform_lim=emb_init_uniform_lim, verbose=verbose)
+
+def baseline_param_init_fn_(module: nn.Module, init_std: float, n_layers: int, d_model: Optional[int]=None, init_div_is_residual: Union[int, float, str, bool]=True, emb_init_std: Optional[float]=None, emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]]=None, verbose: int=0, **kwargs):
+    del kwargs
+    if init_std is None:
+        raise ValueError("You must set model.init_config['init_std'] to a float value to use the default initialization scheme.")
+    _normal_param_init_fn_(module=module, std=init_std, d_model=d_model, n_layers=n_layers, init_div_is_residual=init_div_is_residual, emb_init_std=emb_init_std, emb_init_uniform_lim=emb_init_uniform_lim, verbose=verbose)
+
+def small_param_init_fn_(module: nn.Module, n_layers: int, d_model: int, init_div_is_residual: Union[int, float, str, bool]=True, emb_init_std: Optional[float]=None, emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]]=None, verbose: int=0, **kwargs):
+    del kwargs
+    std = math.sqrt(2 / (5 * d_model))
+    _normal_param_init_fn_(module=module, std=std, d_model=d_model, n_layers=n_layers, init_div_is_residual=init_div_is_residual, emb_init_std=emb_init_std, emb_init_uniform_lim=emb_init_uniform_lim, verbose=verbose)
+
+def neox_param_init_fn_(module: nn.Module, n_layers: int, d_model: int, emb_init_std: Optional[float]=None, emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]]=None, verbose: int=0, **kwargs):
+    """From section 2.3.1 of GPT-NeoX-20B:
+
+    An Open-Source AutoregressiveLanguage Model — Black et. al. (2022)
+    see https://github.com/EleutherAI/gpt-neox/blob/9610391ab319403cef079b438edd016a2443af54/megatron/model/init_functions.py#L151
+    and https://github.com/EleutherAI/gpt-neox/blob/main/megatron/model/transformer.py
+    """
+    del kwargs
+    residual_div = n_layers / math.sqrt(10)
+    if verbose > 1:
+        warnings.warn(f'setting init_div_is_residual to {residual_div}')
+    small_param_init_fn_(module=module, d_model=d_model, n_layers=n_layers, init_div_is_residual=residual_div, emb_init_std=emb_init_std, emb_init_uniform_lim=emb_init_uniform_lim, verbose=verbose)
+
+def kaiming_uniform_param_init_fn_(module: nn.Module, n_layers: int, d_model: Optional[int]=None, init_div_is_residual: Union[int, float, str, bool]=True, emb_init_std: Optional[float]=None, emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]]=None, init_gain: float=0, fan_mode: str='fan_in', init_nonlinearity: str='leaky_relu', verbose: int=0, **kwargs):
+    del kwargs
+    if verbose > 1:
+        warnings.warn(f'Using nn.init.kaiming_uniform_ init fn with parameters: ' + f'a={init_gain}, mode={fan_mode}, nonlinearity={init_nonlinearity}')
+    kaiming_uniform_ = partial(nn.init.kaiming_uniform_, a=init_gain, mode=fan_mode, nonlinearity=init_nonlinearity)
+    generic_param_init_fn_(module=module, init_fn_=kaiming_uniform_, d_model=d_model, n_layers=n_layers, init_div_is_residual=init_div_is_residual, emb_init_std=emb_init_std, emb_init_uniform_lim=emb_init_uniform_lim, verbose=verbose)
+
+def kaiming_normal_param_init_fn_(module: nn.Module, n_layers: int, d_model: Optional[int]=None, init_div_is_residual: Union[int, float, str, bool]=True, emb_init_std: Optional[float]=None, emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]]=None, init_gain: float=0, fan_mode: str='fan_in', init_nonlinearity: str='leaky_relu', verbose: int=0, **kwargs):
+    del kwargs
+    if verbose > 1:
+        warnings.warn(f'Using nn.init.kaiming_normal_ init fn with parameters: ' + f'a={init_gain}, mode={fan_mode}, nonlinearity={init_nonlinearity}')
+    kaiming_normal_ = partial(torch.nn.init.kaiming_normal_, a=init_gain, mode=fan_mode, nonlinearity=init_nonlinearity)
+    generic_param_init_fn_(module=module, init_fn_=kaiming_normal_, d_model=d_model, n_layers=n_layers, init_div_is_residual=init_div_is_residual, emb_init_std=emb_init_std, emb_init_uniform_lim=emb_init_uniform_lim, verbose=verbose)
+
+def xavier_uniform_param_init_fn_(module: nn.Module, n_layers: int, d_model: Optional[int]=None, init_div_is_residual: Union[int, float, str, bool]=True, emb_init_std: Optional[float]=None, emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]]=None, init_gain: float=0, verbose: int=0, **kwargs):
+    del kwargs
+    xavier_uniform_ = partial(torch.nn.init.xavier_uniform_, gain=init_gain)
+    if verbose > 1:
+        warnings.warn(f'Using torch.nn.init.xavier_uniform_ init fn with parameters: ' + f'gain={init_gain}')
+    generic_param_init_fn_(module=module, init_fn_=xavier_uniform_, d_model=d_model, n_layers=n_layers, init_div_is_residual=init_div_is_residual, emb_init_std=emb_init_std, emb_init_uniform_lim=emb_init_uniform_lim, verbose=verbose)
+
+def xavier_normal_param_init_fn_(module: nn.Module, n_layers: int, d_model: Optional[int]=None, init_div_is_residual: Union[int, float, str, bool]=True, emb_init_std: Optional[float]=None, emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]]=None, init_gain: float=0, verbose: int=0, **kwargs):
+    xavier_normal_ = partial(torch.nn.init.xavier_normal_, gain=init_gain)
+    if verbose > 1:
+        warnings.warn(f'Using torch.nn.init.xavier_normal_ init fn with parameters: ' + f'gain={init_gain}')
+    generic_param_init_fn_(module=module, init_fn_=xavier_normal_, d_model=d_model, n_layers=n_layers, init_div_is_residual=init_div_is_residual, emb_init_std=emb_init_std, emb_init_uniform_lim=emb_init_uniform_lim, verbose=verbose)
+MODEL_INIT_REGISTRY = {'default_': torch_default_param_init_fn_, 'baseline_': baseline_param_init_fn_, 'kaiming_uniform_': kaiming_uniform_param_init_fn_, 'kaiming_normal_': kaiming_normal_param_init_fn_, 'neox_init_': neox_param_init_fn_, 'small_init_': small_param_init_fn_, 'xavier_uniform_': xavier_uniform_param_init_fn_, 'xavier_normal_': xavier_normal_param_init_fn_}

special_tokens_map.json

@@ -0,0 +1,5 @@
+{
+  "bos_token": "<|endoftext|>",
+  "eos_token": "<|endoftext|>",
+  "unk_token": "<|endoftext|>"
+}

tokenizer_config.json

@@ -0,0 +1,9 @@
+{
+  "add_prefix_space": false,
+  "bos_token": "<|endoftext|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|endoftext|>",
+  "model_max_length": 2048,
+  "tokenizer_class": "GPTNeoXTokenizer",
+  "unk_token": "<|endoftext|>"
+}