transformers
Fix graph break in torch.compile when using FA2 with attention_mask=None and batch size > 1
#37332

Merged

Fix graph break in torch.compile when using FA2 with attention_mask=None and batch size > 1 #37332

ArthurZucker merged 10 commits into huggingface:main from efsotr:fa2_compile_graph_break

Fix graph break in torch.compile when using FA2 with attention_mask=N…

422e3b3b

github-actions marked this pull request as draft 266 days ago

efsotr marked this pull request as ready for review 266 days ago

github-actions requested a review from

ArthurZucker 266 days ago

github-actions requested a review from

Rocketknight1 266 days ago

fix code format

69a2b038

ArthurZucker commented on 2025-04-08

Merge remote-tracking branch 'upstream/main' into fa2_compile_graph_b…

d9424933

add test; replace position_ids with query_states becasue position_ids…

1dff2564

Merge remote-tracking branch 'upstream/main' into fa2_compile_graph_b…

812cb5a5

ArthurZucker approved these changes on 2025-04-08

Merge branch 'main' into fa2_compile_graph_break

a381b48d

Merge branch 'main' into fa2_compile_graph_break

b1c101bf

Merge branch 'main' into fa2_compile_graph_break

7cb8d411

add assert loss is not nan

5028afc6

Merge branch 'main' into fa2_compile_graph_break

ba3ad0c3

ArthurZucker enabled auto-merge (squash) 187 days ago

ArthurZucker merged 3ee72af6 into main 187 days ago

Reviewers

ArthurZucker

Rocketknight1

Assignees

No one assigned

Labels

None yet

Milestone

No milestone