[PyTorch][torch.compile] Make get_attention_backend traceable without graph breaks - #3189

Merged

pggPL merged 12 commits into

NVIDIA:mainfrom

pggPL:attention_get_attention_backend_traceable

Jul 23, 2026

Commits on Jul 8, 2026

Make get_attention_backend traceable by torch.compile without graph breaks
pggPL
committed
17f80bd
Return int from the fused-attn backend probe and test symbolic-int path
pggPL
committed
36bb8e8
Replace the FusedAttnBackend dict with a python-side IntEnum
pggPL
committed
f7564f9
Soften the FusedAttnBackend enum transition following the DType pattern
pggPL
committed
43b1084
Add FP8 backend-selection coverage to the torch.compile tests
pggPL
committed
c76304b
Simplify the fused-attn probe wrapper and drop obsolete logging guards
pggPL
committed
ba39c93
Consolidate backend-selection compile tests; keep probe args literal
pggPL
committed
1793226
Drop references to unreleased dynamo features from comments
pggPL
committed
719b184
Assert selection results only in the compile test
pggPL
committed
6386e2c
Merge remote-tracking branch 'upstream/main' into attention_get_attention_backend_traceable
pggPL
committed
70190e6

Commits on Jul 23, 2026