Skip to content

GitLab

Unverified Commit e3143952 authored by Arthur's avatar Arthur Committed by GitHub
Browse files

Refactor flash attention implementation in transformers (#31446) 



* dumb commit

* nit

* update

* something like this

* unpack in modeling utils

* safe import

* oups

* update

* nits

* diff convert gemma

* update

* start propagating

* udpate other modeling code as well

* update for sliding window models

* nits

* more init cleanups

* styling

* fixup

* noice

* pass fixup

* typo typing_extension -> typing_extensions

* torch.nn.functionnal -> torch.nn.functional

* add to import structure

* unpack

* simplify a bit more for this first version

* nut

* update

* update

* nit

* ease the import of `Unpack`

* remove useless `use_sliding_window`

* no qua please

* protect import?

* style

* [run-slow]

* [run slow] llama,gemma,mistral,mixtral

* remove extra kwargs

* fix llama

* address review comments

* apply diff_model_converter to modeling_gemma.py

* remove cache_position 1

* remove cache_position 2

* some cleaning

* refactor gemma2 as well

* apply review comments

* rename file to modeling_flash_attention_utils.py

* siglip refactor

* remove dead code

* is the hub down?

* still down?

* fix siglip

* fix gemma2

* fatal: Could not read from remote repository.

* fix typo in softcap implem

* flacky

* Failed: Timeout >120.0s

---------
Co-authored-by: default avatarfxmarty <9808326+fxmarty@users.noreply.github.com>
parent ad4ef3a2
Expand all Show whitespace changes
Inline Side-by-side
Showing with 225 additions and 2297 deletions
src/transformers/models/jamba/modeling_jamba.py
View file @ e3143952
......@@ -19,7 +19,6 @@
# limitations under the License.
"""PyTorch Jamba model."""
import inspect
import math
from typing import Any, Dict, List, Optional, Tuple, Union
......@@ -43,23 +42,20 @@ from ...modeling_utils import PreTrainedModel
from ...utils import (
add_start_docstrings,
add_start_docstrings_to_model_forward,
is_flash_attn_greater_or_equal_2_10,
logging,
replace_return_docstrings,
)
from ...utils.import_utils import (
is_causal_conv1d_available,
is_flash_attn_2_available,
is_flash_attn_greater_or_equal_2_10,
is_mamba_ssm_available,
)
from .configuration_jamba import JambaConfig
if is_flash_attn_2_available():
from flash_attn import flash_attn_func, flash_attn_varlen_func
from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input # noqa
_flash_supports_window_size = "window_size" in list(inspect.signature(flash_attn_func).parameters)
from ...modeling_flash_attention_utils import _flash_attention_forward
if is_mamba_ssm_available():
......@@ -165,19 +161,6 @@ def load_balancing_loss_func(
return overall_loss * num_experts
# Copied from transformers.models.llama.modeling_llama._get_unpad_data
def _get_unpad_data(attention_mask):
seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
max_seqlen_in_batch = seqlens_in_batch.max().item()
cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
return (
indices,
cu_seqlens,
max_seqlen_in_batch,
)
# Copied from transformers.models.llama.modeling_llama.LlamaRMSNorm with Llama->Jamba
class JambaRMSNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-6):
......@@ -429,18 +412,6 @@ class JambaFlashAttention2(JambaAttention):
kv_seq_len = cache_position[-1]
use_sliding_windows = (
_flash_supports_window_size
and getattr(self.config, "sliding_window", None) is not None
and kv_seq_len > self.config.sliding_window
)
if not _flash_supports_window_size:
logger.warning_once(
"The current flash attention version does not support sliding window attention, for a more memory efficient implementation"
" make sure to upgrade flash-attn library."
)
if past_key_value is not None:
# Activate slicing cache only if the config has a value `sliding_windows` attribute
cache_has_contents = cache_position[0] > 0
......@@ -502,14 +473,16 @@ class JambaFlashAttention2(JambaAttention):
key_states = key_states.transpose(1, 2)
value_states = value_states.transpose(1, 2)
attn_output = self._flash_attention_forward(
attn_output = _flash_attention_forward(
query_states,
key_states,
value_states,
attention_mask,
q_len,
dropout=dropout_rate,
use_sliding_windows=use_sliding_windows,
sliding_window=getattr(self.config, "sliding_window", None),
is_causal=self.is_causal,
use_top_left_mask=self._flash_attn_uses_top_left_mask,
)
attn_output = attn_output.reshape(bsz, q_len, self.hidden_size).contiguous()
......@@ -520,149 +493,6 @@ class JambaFlashAttention2(JambaAttention):
return attn_output, attn_weights, past_key_value
def _flash_attention_forward(
self,
query_states,
key_states,
value_states,
attention_mask,
query_length,
dropout=0.0,
softmax_scale=None,
use_sliding_windows=False,
):
"""
Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
first unpad the input, then computes the attention scores and pad the final attention scores.
Args:
query_states (`torch.Tensor`):
Input query states to be passed to Flash Attention API
key_states (`torch.Tensor`):
Input key states to be passed to Flash Attention API
value_states (`torch.Tensor`):
Input value states to be passed to Flash Attention API
attention_mask (`torch.Tensor`):
The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
position of padding tokens and 1 for the position of non-padding tokens.
dropout (`float`, *optional*):
Attention dropout
softmax_scale (`float`, *optional*):
The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
use_sliding_windows (`bool`, *optional*):
Whether to activate sliding window attention.
"""
if not self._flash_attn_uses_top_left_mask:
causal = self.is_causal
else:
# TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
causal = self.is_causal and query_length != 1
# Contains at least one padding token in the sequence
if attention_mask is not None:
batch_size = query_states.shape[0]
query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
query_states, key_states, value_states, attention_mask, query_length
)
cu_seqlens_q, cu_seqlens_k = cu_seq_lens
max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
if not use_sliding_windows:
attn_output_unpad = flash_attn_varlen_func(
query_states,
key_states,
value_states,
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=max_seqlen_in_batch_q,
max_seqlen_k=max_seqlen_in_batch_k,
dropout_p=dropout,
softmax_scale=softmax_scale,
causal=causal,
)
else:
attn_output_unpad = flash_attn_varlen_func(
query_states,
key_states,
value_states,
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=max_seqlen_in_batch_q,
max_seqlen_k=max_seqlen_in_batch_k,
dropout_p=dropout,
softmax_scale=softmax_scale,
causal=causal,
window_size=(self.config.sliding_window, self.config.sliding_window),
)
attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
else:
if not use_sliding_windows:
attn_output = flash_attn_func(
query_states,
key_states,
value_states,
dropout,
softmax_scale=softmax_scale,
causal=causal,
)
else:
attn_output = flash_attn_func(
query_states,
key_states,
value_states,
dropout,
softmax_scale=softmax_scale,
causal=causal,
window_size=(self.config.sliding_window, self.config.sliding_window),
)
return attn_output
# Copied from transformers.models.mixtral.modeling_mixtral.MixtralFlashAttention2._upad_input
def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
batch_size, kv_seq_len, num_heads, head_dim = key_layer.shape
# On the first iteration we need to properly re-create the padding mask
# by slicing it on the proper place
if kv_seq_len != attention_mask.shape[-1]:
attention_mask_num_tokens = attention_mask.shape[-1]
attention_mask = attention_mask[:, attention_mask_num_tokens - kv_seq_len :]
indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
key_layer = index_first_axis(key_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k)
value_layer = index_first_axis(value_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k)
if query_length == kv_seq_len:
query_layer = index_first_axis(
query_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k
)
cu_seqlens_q = cu_seqlens_k
max_seqlen_in_batch_q = max_seqlen_in_batch_k
indices_q = indices_k
elif query_length == 1:
max_seqlen_in_batch_q = 1
cu_seqlens_q = torch.arange(
batch_size + 1, dtype=torch.int32, device=query_layer.device
) # There is a memcpy here, that is very bad.
indices_q = cu_seqlens_q[:-1]
query_layer = query_layer.squeeze(1)
else:
# The -q_len: slice assumes left padding.
attention_mask = attention_mask[:, -query_length:]
query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
return (
query_layer,
key_layer,
value_layer,
indices_q,
(cu_seqlens_q, cu_seqlens_k),
(max_seqlen_in_batch_q, max_seqlen_in_batch_k),
)
# Adapted from transformers.models.mistral.modeling_mistral.MistralSdpaAttention with Mistral->Jamba
class JambaSdpaAttention(JambaAttention):
......
src/transformers/models/jetmoe/modeling_jetmoe.py
View file @ e3143952
......@@ -46,8 +46,7 @@ from .configuration_jetmoe import JetMoeConfig
if is_flash_attn_2_available():
from flash_attn import flash_attn_func, flash_attn_varlen_func
from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input # noqa
from ...modeling_flash_attention_utils import _flash_attention_forward
logger = logging.get_logger(__name__)
......@@ -358,19 +357,6 @@ class JetMoeMoA(nn.Module):
raise NotImplementedError("This module doesn't support call and forward.")
# Copied from transformers.models.llama.modeling_llama._get_unpad_data
def _get_unpad_data(attention_mask):
seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
max_seqlen_in_batch = seqlens_in_batch.max().item()
cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
return (
indices,
cu_seqlens,
max_seqlen_in_batch,
)
# Copied from transformers.models.llama.modeling_llama.LlamaRMSNorm with Llama->JetMoe
class JetMoeRMSNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-6):
......@@ -647,6 +633,7 @@ class JetMoeSdpaAttention(JetMoeAttention):
class JetMoeFlashAttention2(JetMoeAttention):
# Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2.__init__
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
......@@ -739,8 +726,15 @@ class JetMoeFlashAttention2(JetMoeAttention):
key_states = key_states.to(target_dtype)
value_states = value_states.to(target_dtype)
attn_output = self._flash_attention_forward(
query_states, key_states, value_states, attention_mask, q_len, dropout=dropout_rate
attn_output = _flash_attention_forward(
query_states,
key_states,
value_states,
attention_mask,
q_len,
dropout=dropout_rate,
use_top_left_mask=self._flash_attn_uses_top_left_mask,
is_causal=self.is_causal,
).to(input_dtype)
# output projection
......@@ -753,105 +747,6 @@ class JetMoeFlashAttention2(JetMoeAttention):
return attn_output, attn_weights, past_key_value, router_logits
# Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._flash_attention_forward
def _flash_attention_forward(
self, query_states, key_states, value_states, attention_mask, query_length, dropout=0.0, softmax_scale=None
):
"""
Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
first unpad the input, then computes the attention scores and pad the final attention scores.
Args:
query_states (`torch.Tensor`):
Input query states to be passed to Flash Attention API
key_states (`torch.Tensor`):
Input key states to be passed to Flash Attention API
value_states (`torch.Tensor`):
Input value states to be passed to Flash Attention API
attention_mask (`torch.Tensor`):
The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
position of padding tokens and 1 for the position of non-padding tokens.
dropout (`float`):
Attention dropout
softmax_scale (`float`, *optional*):
The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
"""
if not self._flash_attn_uses_top_left_mask:
causal = self.is_causal
else:
# TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
causal = self.is_causal and query_length != 1
# Contains at least one padding token in the sequence
if attention_mask is not None:
batch_size = query_states.shape[0]
query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
query_states, key_states, value_states, attention_mask, query_length
)
cu_seqlens_q, cu_seqlens_k = cu_seq_lens
max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
attn_output_unpad = flash_attn_varlen_func(
query_states,
key_states,
value_states,
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=max_seqlen_in_batch_q,
max_seqlen_k=max_seqlen_in_batch_k,
dropout_p=dropout,
softmax_scale=softmax_scale,
causal=causal,
)
attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
else:
attn_output = flash_attn_func(
query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal
)
return attn_output
# Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._upad_input
def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
key_layer = index_first_axis(
key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
)
value_layer = index_first_axis(
value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
)
if query_length == kv_seq_len:
query_layer = index_first_axis(
query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim), indices_k
)
cu_seqlens_q = cu_seqlens_k
max_seqlen_in_batch_q = max_seqlen_in_batch_k
indices_q = indices_k
elif query_length == 1:
max_seqlen_in_batch_q = 1
cu_seqlens_q = torch.arange(
batch_size + 1, dtype=torch.int32, device=query_layer.device
) # There is a memcpy here, that is very bad.
indices_q = cu_seqlens_q[:-1]
query_layer = query_layer.squeeze(1)
else:
# The -q_len: slice assumes left padding.
attention_mask = attention_mask[:, -query_length:]
query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
return (
query_layer,
key_layer,
value_layer,
indices_q,
(cu_seqlens_q, cu_seqlens_k),
(max_seqlen_in_batch_q, max_seqlen_in_batch_k),
)
JETMOE_ATTENTION_CLASSES = {
"eager": JetMoeAttention,
......
src/transformers/models/llama/modeling_llama.py
View file @ e3143952
......@@ -29,6 +29,7 @@ from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
from ...activations import ACT2FN
from ...cache_utils import Cache, DynamicCache, StaticCache
from ...modeling_attn_mask_utils import AttentionMaskConverter
from ...modeling_flash_attention_utils import _flash_attention_forward
from ...modeling_outputs import (
BaseModelOutputWithPast,
CausalLMOutputWithPast,
......@@ -41,7 +42,6 @@ from ...pytorch_utils import ALL_LAYERNORM_LAYERS
from ...utils import (
add_start_docstrings,
add_start_docstrings_to_model_forward,
is_flash_attn_2_available,
is_flash_attn_greater_or_equal_2_10,
logging,
replace_return_docstrings,
......@@ -49,28 +49,11 @@ from ...utils import (
from .configuration_llama import LlamaConfig
if is_flash_attn_2_available():
from flash_attn import flash_attn_func, flash_attn_varlen_func
from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input # noqa
logger = logging.get_logger(__name__)
_CONFIG_FOR_DOC = "LlamaConfig"
def _get_unpad_data(attention_mask):
seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
max_seqlen_in_batch = seqlens_in_batch.max().item()
cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
return (
indices,
cu_seqlens,
max_seqlen_in_batch,
)
class LlamaRMSNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-6):
"""
......@@ -464,8 +447,16 @@ class LlamaFlashAttention2(LlamaAttention):
key_states = key_states.to(target_dtype)
value_states = value_states.to(target_dtype)
attn_output = self._flash_attention_forward(
query_states, key_states, value_states, attention_mask, q_len, dropout=dropout_rate
attn_output = _flash_attention_forward(
query_states,
key_states,
value_states,
attention_mask,
q_len,
dropout=dropout_rate,
sliding_window=getattr(self, "sliding_window", None),
use_top_left_mask=self._flash_attn_uses_top_left_mask,
is_causal=self.is_causal,
)
attn_output = attn_output.reshape(bsz, q_len, -1).contiguous()
......@@ -476,103 +467,6 @@ class LlamaFlashAttention2(LlamaAttention):
return attn_output, attn_weights, past_key_value
def _flash_attention_forward(
self, query_states, key_states, value_states, attention_mask, query_length, dropout=0.0, softmax_scale=None
):
"""
Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
first unpad the input, then computes the attention scores and pad the final attention scores.
Args:
query_states (`torch.Tensor`):
Input query states to be passed to Flash Attention API
key_states (`torch.Tensor`):
Input key states to be passed to Flash Attention API
value_states (`torch.Tensor`):
Input value states to be passed to Flash Attention API
attention_mask (`torch.Tensor`):
The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
position of padding tokens and 1 for the position of non-padding tokens.
dropout (`float`):
Attention dropout
softmax_scale (`float`, *optional*):
The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
"""
if not self._flash_attn_uses_top_left_mask:
causal = self.is_causal
else:
# TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
causal = self.is_causal and query_length != 1
# Contains at least one padding token in the sequence
if attention_mask is not None:
batch_size = query_states.shape[0]
query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
query_states, key_states, value_states, attention_mask, query_length
)
cu_seqlens_q, cu_seqlens_k = cu_seq_lens
max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
attn_output_unpad = flash_attn_varlen_func(
query_states,
key_states,
value_states,
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=max_seqlen_in_batch_q,
max_seqlen_k=max_seqlen_in_batch_k,
dropout_p=dropout,
softmax_scale=softmax_scale,
causal=causal,
)
attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
else:
attn_output = flash_attn_func(
query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal
)
return attn_output
def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
key_layer = index_first_axis(
key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
)
value_layer = index_first_axis(
value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
)
if query_length == kv_seq_len:
query_layer = index_first_axis(
query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim), indices_k
)
cu_seqlens_q = cu_seqlens_k
max_seqlen_in_batch_q = max_seqlen_in_batch_k
indices_q = indices_k
elif query_length == 1:
max_seqlen_in_batch_q = 1
cu_seqlens_q = torch.arange(
batch_size + 1, dtype=torch.int32, device=query_layer.device
) # There is a memcpy here, that is very bad.
indices_q = cu_seqlens_q[:-1]
query_layer = query_layer.squeeze(1)
else:
# The -q_len: slice assumes left padding.
attention_mask = attention_mask[:, -query_length:]
query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
return (
query_layer,
key_layer,
value_layer,
indices_q,
(cu_seqlens_q, cu_seqlens_k),
(max_seqlen_in_batch_q, max_seqlen_in_batch_k),
)
class LlamaSdpaAttention(LlamaAttention):
"""
......@@ -723,6 +617,7 @@ class LlamaDecoderLayer(nn.Module):
output_attentions=output_attentions,
use_cache=use_cache,
cache_position=cache_position,
**kwargs,
)
hidden_states = residual + hidden_states
......
src/transformers/models/llava_next_video/diff_llava_next_video.py
View file @ e3143952
......@@ -389,8 +389,13 @@ class LlavaNextVideoForConditionalGeneration(LlavaNextForConditionalGeneration):
image_newline=self.image_newline,
)
inputs_embeds = inputs_embeds.to(image_features.dtype)
inputs_embeds, attention_mask, position_ids, labels, input_ids = (
self._merge_input_ids_with_image_features(
(
inputs_embeds,
attention_mask,
position_ids,
labels,
input_ids,
) = self._merge_input_ids_with_image_features(
image_features,
feature_lens,
inputs_embeds,
......@@ -400,7 +405,6 @@ class LlavaNextVideoForConditionalGeneration(LlavaNextForConditionalGeneration):
labels=labels,
image_token_index=self.config.image_token_index,
)
)
# Then merge video tokens if there are any
if pixel_values_videos is not None and pixel_values_videos.size(0) > 0:
video_features = self._get_video_features(pixel_values_videos)
......@@ -408,8 +412,13 @@ class LlavaNextVideoForConditionalGeneration(LlavaNextForConditionalGeneration):
feature_lens = [feature.size(0) for feature in video_features]
video_features = torch.cat(video_features, dim=0)
feature_lens = torch.tensor(feature_lens, dtype=torch.long, device=video_features.device)
inputs_embeds, attention_mask, position_ids, labels, input_ids = (
self._merge_input_ids_with_image_features(
(
inputs_embeds,
attention_mask,
position_ids,
labels,
input_ids,
) = self._merge_input_ids_with_image_features(
video_features,
feature_lens,
inputs_embeds,
......@@ -419,7 +428,6 @@ class LlavaNextVideoForConditionalGeneration(LlavaNextForConditionalGeneration):
labels=labels,
image_token_index=self.config.video_token_index,
)
)
# pixel_values is not None but is empty ---> text only cases
elif (pixel_values is not None and pixel_values.size(0) == 0) or (
......
src/transformers/models/llava_next_video/modeling_llava_next_video.py
View file @ e3143952
......@@ -862,8 +862,13 @@ class LlavaNextVideoForConditionalGeneration(LlavaNextVideoPreTrainedModel):
image_newline=self.image_newline,
)
inputs_embeds = inputs_embeds.to(image_features.dtype)
inputs_embeds, attention_mask, position_ids, labels, input_ids = (
self._merge_input_ids_with_image_features(
(
inputs_embeds,
attention_mask,
position_ids,
labels,
input_ids,
) = self._merge_input_ids_with_image_features(
image_features,
feature_lens,
inputs_embeds,
......@@ -873,7 +878,6 @@ class LlavaNextVideoForConditionalGeneration(LlavaNextVideoPreTrainedModel):
labels=labels,
image_token_index=self.config.image_token_index,
)
)
# Then merge video tokens if there are any
if pixel_values_videos is not None and pixel_values_videos.size(0) > 0:
video_features = self._get_video_features(pixel_values_videos)
......@@ -881,8 +885,13 @@ class LlavaNextVideoForConditionalGeneration(LlavaNextVideoPreTrainedModel):
feature_lens = [feature.size(0) for feature in video_features]
video_features = torch.cat(video_features, dim=0)
feature_lens = torch.tensor(feature_lens, dtype=torch.long, device=video_features.device)
inputs_embeds, attention_mask, position_ids, labels, input_ids = (
self._merge_input_ids_with_image_features(
(
inputs_embeds,
attention_mask,
position_ids,
labels,
input_ids,
) = self._merge_input_ids_with_image_features(
video_features,
feature_lens,
inputs_embeds,
......@@ -892,7 +901,6 @@ class LlavaNextVideoForConditionalGeneration(LlavaNextVideoPreTrainedModel):
labels=labels,
image_token_index=self.config.video_token_index,
)
)
# pixel_values is not None but is empty ---> text only cases
elif (pixel_values is not None and pixel_values.size(0) == 0) or (
......
src/transformers/models/m2m_100/modeling_m2m_100.py
View file @ e3143952
......@@ -18,7 +18,6 @@ import math
from typing import List, Optional, Tuple, Union
import torch
import torch.nn.functional as F
from torch import nn
from torch.nn import CrossEntropyLoss
......@@ -46,8 +45,7 @@ from .configuration_m2m_100 import M2M100Config
if is_flash_attn_2_available():
from flash_attn import flash_attn_func, flash_attn_varlen_func
from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input
from ...modeling_flash_attention_utils import _flash_attention_forward
logger = logging.get_logger(__name__)
......@@ -335,31 +333,14 @@ class M2M100Attention(nn.Module):
return attn_output, attn_weights_reshaped, past_key_value
# Copied from transformers.models.llama.modeling_llama._get_unpad_data
def _get_unpad_data(attention_mask):
seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
max_seqlen_in_batch = seqlens_in_batch.max().item()
cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
return (
indices,
cu_seqlens,
max_seqlen_in_batch,
)
class M2M100FlashAttention2(M2M100Attention):
def __init__(
self,
embed_dim: int,
num_heads: int,
dropout: float = 0.0,
is_decoder: bool = False,
bias: bool = True,
is_causal: bool = False,
config: Optional[M2M100Config] = None,
):
super().__init__(embed_dim, num_heads, dropout, is_decoder, bias, is_causal, config)
# Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2.__init__
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
# flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
# Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
def _reshape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
......@@ -425,8 +406,16 @@ class M2M100FlashAttention2(M2M100Attention):
if past_key_value is not None:
kv_seq_len += past_key_value[0].shape[-2]
attn_output = self._flash_attention_forward(
query_states, key_states, value_states, attention_mask, q_len, dropout=self.dropout, softmax_scale=None
attn_output = _flash_attention_forward(
query_states,
key_states,
value_states,
attention_mask,
q_len,
dropout=self.dropout,
softmax_scale=None,
use_top_left_mask=self._flash_attn_uses_top_left_mask,
is_causal=self.is_causal,
)
# Use the `embed_dim` from the config (stored in the class) rather than `hidden_state` because `attn_output` can be
......@@ -437,105 +426,6 @@ class M2M100FlashAttention2(M2M100Attention):
return attn_output, None, past_key_value
# Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._flash_attention_forward
def _flash_attention_forward(
self, query_states, key_states, value_states, attention_mask, query_length, dropout=0.0, softmax_scale=None
):
"""
Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
first unpad the input, then computes the attention scores and pad the final attention scores.
Args:
query_states (`torch.Tensor`):
Input query states to be passed to Flash Attention API
key_states (`torch.Tensor`):
Input key states to be passed to Flash Attention API
value_states (`torch.Tensor`):
Input value states to be passed to Flash Attention API
attention_mask (`torch.Tensor`):
The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
position of padding tokens and 1 for the position of non-padding tokens.
dropout (`float`):
Attention dropout
softmax_scale (`float`, *optional*):
The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
"""
if not self._flash_attn_uses_top_left_mask:
causal = self.is_causal
else:
# TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
causal = self.is_causal and query_length != 1
# Contains at least one padding token in the sequence
if attention_mask is not None:
batch_size = query_states.shape[0]
query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
query_states, key_states, value_states, attention_mask, query_length
)
cu_seqlens_q, cu_seqlens_k = cu_seq_lens
max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
attn_output_unpad = flash_attn_varlen_func(
query_states,
key_states,
value_states,
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=max_seqlen_in_batch_q,
max_seqlen_k=max_seqlen_in_batch_k,
dropout_p=dropout,
softmax_scale=softmax_scale,
causal=causal,
)
attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
else:
attn_output = flash_attn_func(
query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal
)
return attn_output
# Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._upad_input
def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
key_layer = index_first_axis(
key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
)
value_layer = index_first_axis(
value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
)
if query_length == kv_seq_len:
query_layer = index_first_axis(
query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim), indices_k
)
cu_seqlens_q = cu_seqlens_k
max_seqlen_in_batch_q = max_seqlen_in_batch_k
indices_q = indices_k
elif query_length == 1:
max_seqlen_in_batch_q = 1
cu_seqlens_q = torch.arange(
batch_size + 1, dtype=torch.int32, device=query_layer.device
) # There is a memcpy here, that is very bad.
indices_q = cu_seqlens_q[:-1]
query_layer = query_layer.squeeze(1)
else:
# The -q_len: slice assumes left padding.
attention_mask = attention_mask[:, -query_length:]
query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
return (
query_layer,
key_layer,
value_layer,
indices_q,
(cu_seqlens_q, cu_seqlens_k),
(max_seqlen_in_batch_q, max_seqlen_in_batch_k),
)
# Copied from transformers.models.mbart.modeling_mbart.MBartEncoderLayer with MBart->M2M100, MBART->M2M100
class M2M100EncoderLayer(nn.Module):
......
src/transformers/models/mbart/modeling_mbart.py
View file @ e3143952
......@@ -19,7 +19,6 @@ import math
from typing import List, Optional, Tuple, Union
import torch
import torch.nn.functional as F
import torch.utils.checkpoint
from torch import nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
......@@ -50,8 +49,7 @@ from .configuration_mbart import MBartConfig
if is_flash_attn_2_available():
from flash_attn import flash_attn_func, flash_attn_varlen_func
from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input # noqa
from ...modeling_flash_attention_utils import _flash_attention_forward
logger = logging.get_logger(__name__)
......@@ -63,19 +61,6 @@ _CONFIG_FOR_DOC = "MBartConfig"
_EXPECTED_OUTPUT_SHAPE = [1, 8, 1024]
# Copied from transformers.models.llama.modeling_llama._get_unpad_data
def _get_unpad_data(attention_mask):
seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
max_seqlen_in_batch = seqlens_in_batch.max().item()
cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
return (
indices,
cu_seqlens,
max_seqlen_in_batch,
)
def shift_tokens_right(input_ids: torch.Tensor, pad_token_id: int):
"""
Shift input ids one token to the right, and wrap the last non pad token (the <LID> token) Note that MBart does not
......@@ -400,8 +385,15 @@ class MBartFlashAttention2(MBartAttention):
key_states = key_states.to(target_dtype)
value_states = value_states.to(target_dtype)
attn_output = self._flash_attention_forward(
query_states, key_states, value_states, attention_mask, q_len, dropout=self.dropout
attn_output = _flash_attention_forward(
query_states,
key_states,
value_states,
attention_mask,
q_len,
dropout=self.dropout,
is_causal=self.is_causal,
use_top_left_mask=self._flash_attn_uses_top_left_mask,
)
attn_output = attn_output.reshape(bsz, q_len, -1)
......@@ -412,105 +404,6 @@ class MBartFlashAttention2(MBartAttention):
return attn_output, attn_weights, past_key_value
# Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._flash_attention_forward
def _flash_attention_forward(
self, query_states, key_states, value_states, attention_mask, query_length, dropout=0.0, softmax_scale=None
):
"""
Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
first unpad the input, then computes the attention scores and pad the final attention scores.
Args:
query_states (`torch.Tensor`):
Input query states to be passed to Flash Attention API
key_states (`torch.Tensor`):
Input key states to be passed to Flash Attention API
value_states (`torch.Tensor`):
Input value states to be passed to Flash Attention API
attention_mask (`torch.Tensor`):
The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
position of padding tokens and 1 for the position of non-padding tokens.
dropout (`float`):
Attention dropout
softmax_scale (`float`, *optional*):
The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
"""
if not self._flash_attn_uses_top_left_mask:
causal = self.is_causal
else:
# TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
causal = self.is_causal and query_length != 1
# Contains at least one padding token in the sequence
if attention_mask is not None:
batch_size = query_states.shape[0]
query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
query_states, key_states, value_states, attention_mask, query_length
)
cu_seqlens_q, cu_seqlens_k = cu_seq_lens
max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
attn_output_unpad = flash_attn_varlen_func(
query_states,
key_states,
value_states,
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=max_seqlen_in_batch_q,
max_seqlen_k=max_seqlen_in_batch_k,
dropout_p=dropout,
softmax_scale=softmax_scale,
causal=causal,
)
attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
else:
attn_output = flash_attn_func(
query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal
)
return attn_output
# Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._upad_input
def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
key_layer = index_first_axis(
key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
)
value_layer = index_first_axis(
value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
)
if query_length == kv_seq_len:
query_layer = index_first_axis(
query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim), indices_k
)
cu_seqlens_q = cu_seqlens_k
max_seqlen_in_batch_q = max_seqlen_in_batch_k
indices_q = indices_k
elif query_length == 1:
max_seqlen_in_batch_q = 1
cu_seqlens_q = torch.arange(
batch_size + 1, dtype=torch.int32, device=query_layer.device
) # There is a memcpy here, that is very bad.
indices_q = cu_seqlens_q[:-1]
query_layer = query_layer.squeeze(1)
else:
# The -q_len: slice assumes left padding.
attention_mask = attention_mask[:, -query_length:]
query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
return (
query_layer,
key_layer,
value_layer,
indices_q,
(cu_seqlens_q, cu_seqlens_k),
(max_seqlen_in_batch_q, max_seqlen_in_batch_k),
)
MBART_ATTENTION_CLASSES = {
"eager": MBartAttention,
......
src/transformers/models/mistral/modeling_mistral.py
View file @ e3143952
......@@ -19,12 +19,10 @@
# limitations under the License.
"""PyTorch Mistral model."""
import inspect
import math
from typing import List, Optional, Tuple, Union
import torch
import torch.nn.functional as F
import torch.utils.checkpoint
from torch import nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
......@@ -51,30 +49,13 @@ from .configuration_mistral import MistralConfig
if is_flash_attn_2_available():
from flash_attn import flash_attn_func, flash_attn_varlen_func
from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input # noqa
_flash_supports_window_size = "window_size" in list(inspect.signature(flash_attn_func).parameters)
from ...modeling_flash_attention_utils import _flash_attention_forward
logger = logging.get_logger(__name__)
_CONFIG_FOR_DOC = "MistralConfig"
# Copied from transformers.models.llama.modeling_llama._get_unpad_data
def _get_unpad_data(attention_mask):
seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
max_seqlen_in_batch = seqlens_in_batch.max().item()
cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
return (
indices,
cu_seqlens,
max_seqlen_in_batch,
)
# Copied from transformers.models.llama.modeling_llama.LlamaRMSNorm with Llama->Mistral
class MistralRMSNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-6):
......@@ -337,18 +318,6 @@ class MistralFlashAttention2(MistralAttention):
cos, sin = self.rotary_emb(value_states, position_ids)
query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
use_sliding_windows = (
_flash_supports_window_size
and getattr(self.config, "sliding_window", None) is not None
and kv_seq_len > self.config.sliding_window
)
if not _flash_supports_window_size:
logger.warning_once(
"The current flash attention version does not support sliding window attention, for a more memory efficient implementation"
" make sure to upgrade flash-attn library."
)
if past_key_value is not None:
# Activate slicing cache only if the config has a value `sliding_windows` attribute
cache_has_contents = past_key_value.get_seq_length(self.layer_idx) > 0
......@@ -411,14 +380,16 @@ class MistralFlashAttention2(MistralAttention):
key_states = key_states.transpose(1, 2)
value_states = value_states.transpose(1, 2)
attn_output = self._flash_attention_forward(
attn_output = _flash_attention_forward(
query_states,
key_states,
value_states,
attention_mask,
q_len,
dropout=dropout_rate,
use_sliding_windows=use_sliding_windows,
sliding_window=getattr(self.config, "sliding_window", None),
use_top_left_mask=self._flash_attn_uses_top_left_mask,
is_causal=self.is_causal,
)
attn_output = attn_output.reshape(bsz, q_len, self.hidden_size).contiguous()
......@@ -429,148 +400,6 @@ class MistralFlashAttention2(MistralAttention):
return attn_output, attn_weights, past_key_value
def _flash_attention_forward(
self,
query_states,
key_states,
value_states,
attention_mask,
query_length,
dropout=0.0,
softmax_scale=None,
use_sliding_windows=False,
):
"""
Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
first unpad the input, then computes the attention scores and pad the final attention scores.
Args:
query_states (`torch.Tensor`):
Input query states to be passed to Flash Attention API
key_states (`torch.Tensor`):
Input key states to be passed to Flash Attention API
value_states (`torch.Tensor`):
Input value states to be passed to Flash Attention API
attention_mask (`torch.Tensor`):
The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
position of padding tokens and 1 for the position of non-padding tokens.
dropout (`float`):
Attention dropout
softmax_scale (`float`, *optional*):
The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
use_sliding_windows (`bool`, *optional*):
Whether to activate sliding window attention.
"""
if not self._flash_attn_uses_top_left_mask:
causal = self.is_causal
else:
# TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
causal = self.is_causal and query_length != 1
# Contains at least one padding token in the sequence
if attention_mask is not None:
batch_size = query_states.shape[0]
query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
query_states, key_states, value_states, attention_mask, query_length
)
cu_seqlens_q, cu_seqlens_k = cu_seq_lens
max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
if not use_sliding_windows:
attn_output_unpad = flash_attn_varlen_func(
query_states,
key_states,
value_states,
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=max_seqlen_in_batch_q,
max_seqlen_k=max_seqlen_in_batch_k,
dropout_p=dropout,
softmax_scale=softmax_scale,
causal=causal,
)
else:
attn_output_unpad = flash_attn_varlen_func(
query_states,
key_states,
value_states,
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=max_seqlen_in_batch_q,
max_seqlen_k=max_seqlen_in_batch_k,
dropout_p=dropout,
softmax_scale=softmax_scale,
causal=causal,
window_size=(self.config.sliding_window, self.config.sliding_window),
)
attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
else:
if not use_sliding_windows:
attn_output = flash_attn_func(
query_states,
key_states,
value_states,
dropout,
softmax_scale=softmax_scale,
causal=causal,
)
else:
attn_output = flash_attn_func(
query_states,
key_states,
value_states,
dropout,
softmax_scale=softmax_scale,
causal=causal,
window_size=(self.config.sliding_window, self.config.sliding_window),
)
return attn_output
def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
batch_size, kv_seq_len, num_heads, head_dim = key_layer.shape
# On the first iteration we need to properly re-create the padding mask
# by slicing it on the proper place
if kv_seq_len != attention_mask.shape[-1]:
attention_mask_num_tokens = attention_mask.shape[-1]
attention_mask = attention_mask[:, attention_mask_num_tokens - kv_seq_len :]
indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
key_layer = index_first_axis(key_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k)
value_layer = index_first_axis(value_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k)
if query_length == kv_seq_len:
query_layer = index_first_axis(
query_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k
)
cu_seqlens_q = cu_seqlens_k
max_seqlen_in_batch_q = max_seqlen_in_batch_k
indices_q = indices_k
elif query_length == 1:
max_seqlen_in_batch_q = 1
cu_seqlens_q = torch.arange(
batch_size + 1, dtype=torch.int32, device=query_layer.device
) # There is a memcpy here, that is very bad.
indices_q = cu_seqlens_q[:-1]
query_layer = query_layer.squeeze(1)
else:
# The -q_len: slice assumes left padding.
attention_mask = attention_mask[:, -query_length:]
query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
return (
query_layer,
key_layer,
value_layer,
indices_q,
(cu_seqlens_q, cu_seqlens_k),
(max_seqlen_in_batch_q, max_seqlen_in_batch_k),
)
# Copied from transformers.models.llama.modeling_llama.LlamaSdpaAttention with Llama->Mistral
class MistralSdpaAttention(MistralAttention):
......@@ -723,6 +552,7 @@ class MistralDecoderLayer(nn.Module):
output_attentions=output_attentions,
use_cache=use_cache,
cache_position=cache_position,
**kwargs,
)
hidden_states = residual + hidden_states
......
src/transformers/models/mixtral/modeling_mixtral.py
View file @ e3143952
......@@ -19,7 +19,6 @@
# limitations under the License.
"""PyTorch Mixtral model."""
import inspect
import math
from typing import List, Optional, Tuple, Union
......@@ -47,7 +46,6 @@ from ...utils import (
add_start_docstrings,
add_start_docstrings_to_model_forward,
is_flash_attn_2_available,
is_flash_attn_greater_or_equal_2_10,
logging,
replace_return_docstrings,
)
......@@ -56,10 +54,7 @@ from .configuration_mixtral import MixtralConfig
if is_flash_attn_2_available():
from flash_attn import flash_attn_func, flash_attn_varlen_func
from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input # noqa
_flash_supports_window_size = "window_size" in list(inspect.signature(flash_attn_func).parameters)
from ...modeling_flash_attention_utils import _flash_attention_forward
# This makes `_prepare_4d_causal_attention_mask` a leaf function in the FX graph.
# It means that the function will not be traced through and simply appear as a node in the graph.
......@@ -151,19 +146,6 @@ def load_balancing_loss_func(
return overall_loss * num_experts
# Copied from transformers.models.llama.modeling_llama._get_unpad_data
def _get_unpad_data(attention_mask):
seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
max_seqlen_in_batch = seqlens_in_batch.max().item()
cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
return (
indices,
cu_seqlens,
max_seqlen_in_batch,
)
# Copied from transformers.models.llama.modeling_llama.LlamaRMSNorm with Llama->Mixtral
class MixtralRMSNorm(nn.Module):
def __init__(self, hidden_size, eps=1e-6):
......@@ -402,15 +384,6 @@ class MixtralFlashAttention2(MixtralAttention):
flash attention and deal with padding tokens in case the input contains any of them.
"""
# Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2.__init__
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
# flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
# Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
def forward(
self,
hidden_states: torch.Tensor,
......@@ -447,18 +420,6 @@ class MixtralFlashAttention2(MixtralAttention):
query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
use_sliding_windows = (
_flash_supports_window_size
and getattr(self.config, "sliding_window", None) is not None
and kv_seq_len > self.config.sliding_window
)
if not _flash_supports_window_size:
logger.warning_once(
"The current flash attention version does not support sliding window attention, for a more memory efficient implementation"
" make sure to upgrade flash-attn library."
)
if past_key_value is not None:
# Activate slicing cache only if the config has a value `sliding_windows` attribute
cache_has_contents = past_key_value.get_seq_length(self.layer_idx) > 0
......@@ -521,14 +482,15 @@ class MixtralFlashAttention2(MixtralAttention):
key_states = key_states.transpose(1, 2)
value_states = value_states.transpose(1, 2)
attn_output = self._flash_attention_forward(
attn_output = _flash_attention_forward(
query_states,
key_states,
value_states,
attention_mask,
q_len,
dropout=dropout_rate,
use_sliding_windows=use_sliding_windows,
sliding_window=getattr(self.config, "sliding_window", None),
is_causal=self.is_causal,
)
attn_output = attn_output.reshape(bsz, q_len, self.hidden_size).contiguous()
......@@ -539,148 +501,6 @@ class MixtralFlashAttention2(MixtralAttention):
return attn_output, attn_weights, past_key_value
def _flash_attention_forward(
self,
query_states,
key_states,
value_states,
attention_mask,
query_length,
dropout=0.0,
softmax_scale=None,
use_sliding_windows=False,
):
"""
Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
first unpad the input, then computes the attention scores and pad the final attention scores.
Args:
query_states (`torch.Tensor`):
Input query states to be passed to Flash Attention API
key_states (`torch.Tensor`):
Input key states to be passed to Flash Attention API
value_states (`torch.Tensor`):
Input value states to be passed to Flash Attention API
attention_mask (`torch.Tensor`):
The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
position of padding tokens and 1 for the position of non-padding tokens.
dropout (`float`):
Attention dropout
softmax_scale (`float`, *optional*):
The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
use_sliding_windows (`bool`, *optional*):
Whether to activate sliding window attention.
"""
if not self._flash_attn_uses_top_left_mask:
causal = self.is_causal
else:
# TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
causal = self.is_causal and query_length != 1
# Contains at least one padding token in the sequence
if attention_mask is not None:
batch_size = query_states.shape[0]
query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
query_states, key_states, value_states, attention_mask, query_length
)
cu_seqlens_q, cu_seqlens_k = cu_seq_lens
max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
if not use_sliding_windows:
attn_output_unpad = flash_attn_varlen_func(
query_states,
key_states,
value_states,
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=max_seqlen_in_batch_q,
max_seqlen_k=max_seqlen_in_batch_k,
dropout_p=dropout,
softmax_scale=softmax_scale,
causal=causal,
)
else:
attn_output_unpad = flash_attn_varlen_func(
query_states,
key_states,
value_states,
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=max_seqlen_in_batch_q,
max_seqlen_k=max_seqlen_in_batch_k,
dropout_p=dropout,
softmax_scale=softmax_scale,
causal=causal,
window_size=(self.config.sliding_window, self.config.sliding_window),
)
attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
else:
if not use_sliding_windows:
attn_output = flash_attn_func(
query_states,
key_states,
value_states,
dropout,
softmax_scale=softmax_scale,
causal=causal,
)
else:
attn_output = flash_attn_func(
query_states,
key_states,
value_states,
dropout,
softmax_scale=softmax_scale,
causal=causal,
window_size=(self.config.sliding_window, self.config.sliding_window),
)
return attn_output
def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
batch_size, kv_seq_len, num_heads, head_dim = key_layer.shape
# On the first iteration we need to properly re-create the padding mask
# by slicing it on the proper place
if kv_seq_len != attention_mask.shape[-1]:
attention_mask_num_tokens = attention_mask.shape[-1]
attention_mask = attention_mask[:, attention_mask_num_tokens - kv_seq_len :]
indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
key_layer = index_first_axis(key_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k)
value_layer = index_first_axis(value_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k)
if query_length == kv_seq_len:
query_layer = index_first_axis(
query_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k
)
cu_seqlens_q = cu_seqlens_k
max_seqlen_in_batch_q = max_seqlen_in_batch_k
indices_q = indices_k
elif query_length == 1:
max_seqlen_in_batch_q = 1
cu_seqlens_q = torch.arange(
batch_size + 1, dtype=torch.int32, device=query_layer.device
) # There is a memcpy here, that is very bad.
indices_q = cu_seqlens_q[:-1]
query_layer = query_layer.squeeze(1)
else:
# The -q_len: slice assumes left padding.
attention_mask = attention_mask[:, -query_length:]
query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
return (
query_layer,
key_layer,
value_layer,
indices_q,
(cu_seqlens_q, cu_seqlens_k),
(max_seqlen_in_batch_q, max_seqlen_in_batch_k),
)
# copied from transformers.models.mistral.modeling_mistral.MistralSdpaAttention with Mistral->Mixtral
# TODO @longjie no longer copied from Mistral after static cache
......
src/transformers/models/musicgen/modeling_musicgen.py
View file @ e3143952
......@@ -23,7 +23,6 @@ from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Union
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.nn import CrossEntropyLoss
from ...activations import ACT2FN
......@@ -58,8 +57,7 @@ from .configuration_musicgen import MusicgenConfig, MusicgenDecoderConfig
if is_flash_attn_2_available():
from flash_attn import flash_attn_func, flash_attn_varlen_func
from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input # noqa
from ...modeling_flash_attention_utils import _flash_attention_forward
if TYPE_CHECKING:
from ...generation.streamers import BaseStreamer
......@@ -70,19 +68,6 @@ _CONFIG_FOR_DOC = "MusicgenConfig"
_CHECKPOINT_FOR_DOC = "facebook/musicgen-small"
# Copied from transformers.models.llama.modeling_llama._get_unpad_data
def _get_unpad_data(attention_mask):
seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
max_seqlen_in_batch = seqlens_in_batch.max().item()
cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
return (
indices,
cu_seqlens,
max_seqlen_in_batch,
)
@dataclass
class MusicgenUnconditionalInput(ModelOutput):
"""
......@@ -434,8 +419,15 @@ class MusicgenFlashAttention2(MusicgenAttention):
key_states = key_states.to(target_dtype)
value_states = value_states.to(target_dtype)
attn_output = self._flash_attention_forward(
query_states, key_states, value_states, attention_mask, q_len, dropout=self.dropout
attn_output = _flash_attention_forward(
query_states,
key_states,
value_states,
attention_mask,
q_len,
dropout=self.dropout,
is_causal=self.is_causal,
use_top_left_mask=self._flash_attn_uses_top_left_mask,
)
attn_output = attn_output.reshape(bsz, q_len, -1)
......@@ -446,104 +438,6 @@ class MusicgenFlashAttention2(MusicgenAttention):
return attn_output, attn_weights, past_key_value
# Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._flash_attention_forward
def _flash_attention_forward(
self, query_states, key_states, value_states, attention_mask, query_length, dropout=0.0, softmax_scale=None
):
"""
Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
first unpad the input, then computes the attention scores and pad the final attention scores.
Args:
query_states (`torch.Tensor`):
Input query states to be passed to Flash Attention API
key_states (`torch.Tensor`):
Input key states to be passed to Flash Attention API
value_states (`torch.Tensor`):
Input value states to be passed to Flash Attention API
attention_mask (`torch.Tensor`):
The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
position of padding tokens and 1 for the position of non-padding tokens.
dropout (`float`):
Attention dropout
softmax_scale (`float`, *optional*):
The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
"""
if not self._flash_attn_uses_top_left_mask:
causal = self.is_causal
else:
# TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
causal = self.is_causal and query_length != 1
# Contains at least one padding token in the sequence
if attention_mask is not None:
batch_size = query_states.shape[0]
query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
query_states, key_states, value_states, attention_mask, query_length
)
cu_seqlens_q, cu_seqlens_k = cu_seq_lens
max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
attn_output_unpad = flash_attn_varlen_func(
query_states,
key_states,
value_states,
cu_seqlens_q=cu_seqlens_q,
cu_seqlens_k=cu_seqlens_k,
max_seqlen_q=max_seqlen_in_batch_q,
max_seqlen_k=max_seqlen_in_batch_k,
dropout_p=dropout,
softmax_scale=softmax_scale,
causal=causal,
)
attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
else:
attn_output = flash_attn_func(
query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal
)
return attn_output
# Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._upad_input
def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
key_layer = index_first_axis(
key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
)
value_layer = index_first_axis(
value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
)
if query_length == kv_seq_len:
query_layer = index_first_axis(
query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim), indices_k
)
cu_seqlens_q = cu_seqlens_k
max_seqlen_in_batch_q = max_seqlen_in_batch_k
indices_q = indices_k
elif query_length == 1:
max_seqlen_in_batch_q = 1
cu_seqlens_q = torch.arange(
batch_size + 1, dtype=torch.int32, device=query_layer.device
) # There is a memcpy here, that is very bad.
indices_q = cu_seqlens_q[:-1]
query_layer = query_layer.squeeze(1)
else:
# The -q_len: slice assumes left padding.
attention_mask = attention_mask[:, -query_length:]
query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
return (
query_layer,
key_layer,
value_layer,
indices_q,
(cu_seqlens_q, cu_seqlens_k),
(max_seqlen_in_batch_q, max_seqlen_in_batch_k),
)
class MusicgenSdpaAttention(MusicgenAttention):
def forward(
......
src/transformers/models/opt/modeling_opt.py
View file @ e3143952
This diff is collapsed.
src/transformers/models/phi/modeling_phi.py
View file @ e3143952
This diff is collapsed.
src/transformers/models/qwen2/configuration_qwen2.py
View file @ e3143952
......@@ -119,7 +119,7 @@ class Qwen2Config(PretrainedConfig):
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.use_sliding_window = use_sliding_window
self.sliding_window = sliding_window
self.sliding_window = sliding_window if use_sliding_window else None
self.max_window_layers = max_window_layers
# for backward compatibility
......
src/transformers/models/qwen2_moe/configuration_qwen2_moe.py
View file @ e3143952
......@@ -149,7 +149,7 @@ class Qwen2MoeConfig(PretrainedConfig):
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.use_sliding_window = use_sliding_window
self.sliding_window = sliding_window
self.sliding_window = sliding_window if use_sliding_window else None
self.max_window_layers = max_window_layers
self.num_key_value_heads = num_key_value_heads
......
src/transformers/models/sew/modeling_sew.py
View file @ e3143952
This diff is collapsed.
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment