[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
4cba8bc

Commits on Jul 8, 2026

Merge branch 'main' of https://github.com/NVIDIA/TransformerEngine into unfused_attention_torch_compile
pggPL
committed
da07306
[PyTorch] torch.compile support for the FP8-emulation path of UnfusedDotProductAttention
pggPL
committed
5190150
[PyTorch] Run UnfusedDotProductAttention as an eager island when fp8_output=True
pggPL
committed
15cbcd1
[PyTorch] Test FP8-emulation compile path also with reduce-overhead (cudagraphs)
pggPL
committed
8bd9881

Commits on Jul 9, 2026

Commits on Jul 10, 2026

[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
8a7c2e8

Commits on Jul 23, 2026