Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix mask slicing for models with HybridCache #35681

Merged
merged 16 commits into from
Jan 28, 2025
Merged

Fix mask slicing for models with HybridCache #35681

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
16 commits
Select commit Hold shift + click to select a range
e1247d1
correctly slice
Cyrilvallez Jan 13, 2025
9c59b35
check mask
Cyrilvallez Jan 14, 2025
9f47f44
Update modular_gemma2.py
Cyrilvallez Jan 14, 2025
ed2e3d7
fix
Cyrilvallez Jan 14, 2025
9414572
add tests
Cyrilvallez Jan 14, 2025
7c31343
fix typo
Cyrilvallez Jan 14, 2025
8004a70
finally fix mask slicing
Cyrilvallez Jan 14, 2025
de59ddc
Finally correctly slice in all cases!!
Cyrilvallez Jan 15, 2025
5667025
add test for all attention functions
Cyrilvallez Jan 15, 2025
1db64c6
small fix in tests
Cyrilvallez Jan 15, 2025
f1d4868
trick around dynamo tracing issue
Cyrilvallez Jan 15, 2025
ba84059
last update
Cyrilvallez Jan 15, 2025
0f0958f
more robust
Cyrilvallez Jan 15, 2025
370326a
kwargs propagation
Cyrilvallez Jan 15, 2025
213da5a
make it explicit for checkpointing
Cyrilvallez Jan 15, 2025
ee426a8
apply modular
Cyrilvallez Jan 15, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 7 additions & 7 deletions src/transformers/models/cohere2/modeling_cohere2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -258,6 +258,11 @@ def forward(
}
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)

# Here we need to slice as we use a static cache by default, but FA2 does not support it
if self.config._attn_implementation == "flash_attention_2":
seq_len = attention_mask.shape[-1]
key_states, value_states = key_states[:, :, :seq_len, :], value_states[:, :, :seq_len, :]

attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
Expand Down Expand Up @@ -344,18 +349,13 @@ def forward(
"""

if self.is_sliding and attention_mask is not None: # efficient SDPA and no padding
# Flash-attn is a 2D tensor
if self.config._attn_implementation == "flash_attention_2":
if past_key_value is not None: # when decoding
attention_mask = attention_mask[:, -self.sliding_window :]
else:
# For FA2, the mask is 2D and is of shape [bs, seq_len] (not [bs, fixed_cache_len])
if self.config._attn_implementation != "flash_attention_2":
min_dtype = torch.finfo(hidden_states.dtype).min
sliding_window_mask = torch.tril(
torch.ones_like(attention_mask, dtype=torch.bool), diagonal=-self.sliding_window
)
attention_mask = torch.where(sliding_window_mask, min_dtype, attention_mask)
if attention_mask.shape[-1] <= 1: # when decoding
attention_mask = attention_mask[:, :, :, -self.sliding_window :]

residual = hidden_states

Expand Down
14 changes: 7 additions & 7 deletions src/transformers/models/cohere2/modular_cohere2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -296,6 +296,11 @@ def forward(
}
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)

# Here we need to slice as we use a static cache by default, but FA2 does not support it
if self.config._attn_implementation == "flash_attention_2":
seq_len = attention_mask.shape[-1]
key_states, value_states = key_states[:, :, :seq_len, :], value_states[:, :, :seq_len, :]

attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
Expand Down Expand Up @@ -363,18 +368,13 @@ def forward(
"""

if self.is_sliding and attention_mask is not None: # efficient SDPA and no padding
# Flash-attn is a 2D tensor
if self.config._attn_implementation == "flash_attention_2":
if past_key_value is not None: # when decoding
attention_mask = attention_mask[:, -self.sliding_window :]
else:
# For FA2, the mask is 2D and is of shape [bs, seq_len] (not [bs, fixed_cache_len])
if self.config._attn_implementation != "flash_attention_2":
min_dtype = torch.finfo(hidden_states.dtype).min
sliding_window_mask = torch.tril(
torch.ones_like(attention_mask, dtype=torch.bool), diagonal=-self.sliding_window
)
attention_mask = torch.where(sliding_window_mask, min_dtype, attention_mask)
if attention_mask.shape[-1] <= 1: # when decoding
attention_mask = attention_mask[:, :, :, -self.sliding_window :]

residual = hidden_states

Expand Down
14 changes: 7 additions & 7 deletions src/transformers/models/gemma2/modeling_gemma2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -223,6 +223,11 @@ def forward(
cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)

# Here we need to slice as we use a static cache by default, but FA2 does not support it
if self.config._attn_implementation == "flash_attention_2":
seq_len = attention_mask.shape[-1]
key_states, value_states = key_states[:, :, :seq_len, :], value_states[:, :, :seq_len, :]

attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
Expand Down Expand Up @@ -278,18 +283,13 @@ def forward(
cache_position: Optional[torch.LongTensor] = None,
) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
if self.is_sliding and attention_mask is not None: # efficient SDPA and no padding
# Flash-attn is a 2D tensor
if self.config._attn_implementation == "flash_attention_2":
if past_key_value is not None: # when decoding
attention_mask = attention_mask[:, -self.sliding_window :]
else:
# For FA2, the mask is 2D and is of shape [bs, seq_len] (not [bs, fixed_cache_len])
if self.config._attn_implementation != "flash_attention_2":
min_dtype = torch.finfo(hidden_states.dtype).min
sliding_window_mask = torch.tril(
torch.ones_like(attention_mask, dtype=torch.bool), diagonal=-self.sliding_window
)
attention_mask = torch.where(sliding_window_mask, min_dtype, attention_mask)
if attention_mask.shape[-1] <= 1: # when decoding
attention_mask = attention_mask[:, :, :, -self.sliding_window :]
Comment on lines -291 to -292
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

this is super important ! Why is it removed?
I know it is counter intuitive, but _flash_attention_forward takes the attention mask to pad / unpad the input itds.
Thus you need the slicing otherwise this operation fails, see the blame !

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Indeed, I was too fast on this one, the HybridCache behaves slightly differently than I remembered. There was still an issue in the slicing during prefill for FA2 though!


residual = hidden_states

Expand Down
14 changes: 7 additions & 7 deletions src/transformers/models/gemma2/modular_gemma2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -259,6 +259,11 @@ def forward(
cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)

# Here we need to slice as we use a static cache by default, but FA2 does not support it
if self.config._attn_implementation == "flash_attention_2":
seq_len = attention_mask.shape[-1]
key_states, value_states = key_states[:, :, :seq_len, :], value_states[:, :, :seq_len, :]

attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
Expand Down Expand Up @@ -314,18 +319,13 @@ def forward(
cache_position: Optional[torch.LongTensor] = None,
) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
if self.is_sliding and attention_mask is not None: # efficient SDPA and no padding
# Flash-attn is a 2D tensor
if self.config._attn_implementation == "flash_attention_2":
if past_key_value is not None: # when decoding
attention_mask = attention_mask[:, -self.sliding_window :]
else:
# For FA2, the mask is 2D and is of shape [bs, seq_len] (not [bs, fixed_cache_len])
if self.config._attn_implementation != "flash_attention_2":
min_dtype = torch.finfo(hidden_states.dtype).min
sliding_window_mask = torch.tril(
torch.ones_like(attention_mask, dtype=torch.bool), diagonal=-self.sliding_window
)
attention_mask = torch.where(sliding_window_mask, min_dtype, attention_mask)
if attention_mask.shape[-1] <= 1: # when decoding
attention_mask = attention_mask[:, :, :, -self.sliding_window :]

residual = hidden_states

Expand Down
Loading