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.

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: 'Gemma2Attention' object has no attribute '_flash_attn_uses_top_left_mask' #35285

Closed
Closed
Changes from all commits
Commits
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
3 changes: 3 additions & 0 deletions src/transformers/configuration_utils.py
Original file line number Diff line number Diff line change
@@ -36,6 +36,7 @@
copy_func,
download_url,
extract_commit_hash,
is_flash_attn_greater_or_equal_2_10,
is_remote_url,
is_timm_config_dict,
is_torch_available,
@@ -304,6 +305,8 @@ def __init__(self, **kwargs):
self._attn_implementation_internal = kwargs.pop("attn_implementation", None)
self._attn_implementation_autoset = False

self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()

# Drop the transformers version info
self.transformers_version = kwargs.pop("transformers_version", None)

15 changes: 8 additions & 7 deletions src/transformers/models/gemma2/modeling_gemma2.py
Original file line number Diff line number Diff line change
@@ -199,7 +199,7 @@ def eager_attention_forward(


def flash_attention_forward(
config: Gemma2Config,
self: 'Gemma2Attention',
query: torch.Tensor,
key: torch.Tensor,
value: torch.Tensor,
@@ -218,26 +218,27 @@ def flash_attention_forward(
key_states = key.transpose(1, 2)
value_states = value.transpose(1, 2)

dropout_rate = config.attention_dropout if config.training else 0.0
dropout_rate = self.config.attention_dropout if self.training else 0.0

input_dtype = query_states.dtype
if input_dtype == torch.float32:
query_states = query_states.to(target_dtype)
key_states = key_states.to(target_dtype)
value_states = value_states.to(target_dtype)


attn_output = _flash_attention_forward(
query_states,
key_states,
value_states,
mask,
seq_len,
dropout=dropout_rate,
softmax_scale=config.scaling,
is_causal=config.is_causal,
sliding_window=config.sliding_window,
use_top_left_mask=config._flash_attn_uses_top_left_mask,
softcap=config.attn_logit_softcapping if is_flash_attn_greater_or_equal("2.6.0") else None,
softmax_scale=self.scaling,
is_causal=self.is_causal,
sliding_window=self.config.sliding_window,
use_top_left_mask=self.config._flash_attn_uses_top_left_mask,
softcap=self.config.attn_logit_softcapping if is_flash_attn_greater_or_equal("2.6.0") else None,
)

return attn_output, None