Skip to content

Fix gemma4 has flash-attention incompatbile head-dim=512 #45202

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

Closed
Qubitium wants to merge 10 commits into from
Closed

Fix gemma4 has flash-attention incompatbile head-dim=512 #45202

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
cdf601b
fix gemma4 has flash-attention incompatbile head-dim=512
Qubitium Apr 2, 2026
128f8e4
Merge branch 'main' into gemma4-fa-fix
Qubitium Apr 2, 2026
3919a91
remove head-dim override
Qubitium Apr 2, 2026
f7f9ea0
fix gemma4 tests
Qubitium Apr 3, 2026
5bc8fdb
cleanup
Qubitium Apr 3, 2026
66d2c0c
fix ci failing
Qubitium Apr 3, 2026
9479356
Merge branch 'main' into gemma4-fa-fix
Qubitium Apr 7, 2026
2903f03
revert test changes
Qubitium Apr 12, 2026
e18fe4a
Merge remote-tracking branch 'origin/main' into gemma4-fa-fix
Qubitium Apr 12, 2026
1aa1d50
fix small conflict and add comment
Cyrilvallez Apr 23, 2026
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
2 changes: 1 addition & 1 deletion src/transformers/models/gemma4/modeling_gemma4.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1443,7 +1443,7 @@ class Gemma4PreTrainedModel(PreTrainedModel):
supports_gradient_checkpointing = True
_no_split_modules = ["Gemma4TextDecoderLayer", "Gemma4VisionEncoderLayer", "Gemma4AudioLayer"]
_skip_keys_device_placement = ["past_key_values", "shared_kv_states"]
_supports_flash_attn = True
_supports_flash_attn = False # released checkpoints use head_dim=512, which is not supported yet by FA kernels
_supports_sdpa = True
_supports_flex_attn = True

Expand Down
1 change: 1 addition & 0 deletions src/transformers/models/gemma4/modular_gemma4.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1161,6 +1161,7 @@ class Gemma4TextScaledWordEmbedding(Gemma3TextScaledWordEmbedding):
class Gemma4PreTrainedModel(Gemma3nPreTrainedModel):
_no_split_modules = ["Gemma4TextDecoderLayer", "Gemma4VisionEncoderLayer", "Gemma4AudioLayer"]
input_modalities = ("image", "text", "video", "audio")
_supports_flash_attn = False # released checkpoints use head_dim=512, which is not supported yet by FA kernels
_can_record_outputs = None # override

@torch.no_grad()
Expand Down
Loading