Skip to content

Allow customization of sdpa in executorch.py #38827

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

Merged
merged 1 commit into from
Jun 17, 2025
Merged

Allow customization of sdpa in executorch.py #38827

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
8 changes: 4 additions & 4 deletions src/transformers/integrations/executorch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -65,6 +65,10 @@ def __init__(
"Using `StaticCache` for export as `layer_types` is not specified or `sliding_window` is `null` in the config."
)
self.model = TorchExportableModuleWithStaticCache(model)
# This is the same as sdpa, but mask creation does not use `vmap` which is not exportable
ALL_MASK_ATTENTION_FUNCTIONS.register("sdpa_without_vmap", sdpa_mask_without_vmap)
ALL_ATTENTION_FUNCTIONS.register("sdpa_without_vmap", ALL_ATTENTION_FUNCTIONS["sdpa"])
self.model.model.config._attn_implementation = "sdpa_without_vmap"

def forward(
self,
Expand Down Expand Up @@ -103,10 +107,6 @@ def export(
strict(`Optional[bool]`):
Flag to instruct `torch.export` to use `torchdynamo`.
"""
# This is the same as sdpa, but mask creation does not use `vmap` which is not exportable
ALL_MASK_ATTENTION_FUNCTIONS.register("sdpa_without_vmap", sdpa_mask_without_vmap)
ALL_ATTENTION_FUNCTIONS.register("sdpa_without_vmap", ALL_ATTENTION_FUNCTIONS["sdpa"])
self.model.model.config._attn_implementation = "sdpa_without_vmap"

example_input_ids = input_ids if input_ids is not None else torch.tensor([[1]], dtype=torch.long)
example_cache_position = cache_position if cache_position is not None else torch.tensor([0], dtype=torch.long)
Expand Down