[PyTorch][torch.compile] Support for UnfusedDotProductAttention#3201

Merged

pggPL merged 15 commits into

NVIDIA:mainfrom

pggPL:unfused_dpa_torch_compile

Jul 24, 2026

Commits on Jun 25, 2026

[PyTorch] Make UnfusedDotProductAttention compatible with torch.compile + CUDA graphs
pggPL
committed

Commits on Jul 9, 2026

[PyTorch] Drop FP8 torch.compile support in UnfusedDotProductAttention; run FP8 as an eager island
pggPL
committed
[PyTorch] Remove te_fp8_emu custom ops; restore upstream FP8EmulationFunc
pggPL
committed
[PyTorch] Shorten the total_tokens_q comment
pggPL
committed
[PyTorch] Drop _scale_to_tensor; pass the softmax scale as a plain float
pggPL
committed
[PyTorch] Drop redundant num_tokens comment (rationale documented at the callsite)
pggPL
committed
[PyTorch] Review cleanups: black formatting, drop unused import/duplicate dict, silence W0613
pggPL
committed

Commits on Jul 10, 2026

[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed

Commits on Jul 23, 2026