Skip to content

[PyTorch] Keep Flex Attention score_mod graphs on the cuDNN backend #3796

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

Open
bvolpato wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[PyTorch] Keep Flex Attention score_mod graphs on the cuDNN backend #3796

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
9 changes: 9 additions & 0 deletions transformer_engine/pytorch/attention/dot_product_attention/flex_attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -270,12 +270,21 @@ def _finalize_cudnn_graph(graph) -> int:

graph.validate()
graph.build_operation_graph()
# Only the cuDNN backend implements score_mod. cuDNN frontend 1.31 can offer its Python SDPA
# engines for these graphs, and those engines run plain attention without the modifier.
graph.deselect_engines(["sdpa_fwd", "sdpa_bwd"])
try:
graph.create_execution_plans([cudnn.heur_mode.A, cudnn.heur_mode.FALLBACK])
graph.check_support()
except cudnn.cudnnGraphNotSupportedError as exc:
raise RuntimeError(f"cuDNN Flex Attention SDPA graph is not supported: {exc}") from exc
graph.build_plans(cudnn.build_plan_policy.HEURISTICS_CHOICE)
selected_engine = getattr(graph, "selected_engine", None)
if selected_engine is not None:
raise RuntimeError(
f"cuDNN frontend {cudnn.__version__} selected {selected_engine} for a score_mod graph. "
"Only the cuDNN backend implements score_mod."
)
return max(graph.get_workspace_size(), 1)


Expand Down
Loading