[PyTorch] DotProductAttention: declarative packed qkv/kv inputs - #18

Open

pggPL wants to merge 37 commits into

mainfrom

dpa_packed_qkv_api

Commits on Jul 7, 2026

[PyTorch] DotProductAttention: declarative packed qkv/kv inputs

pggPL
and
claude
committed
a9cd697

Commits on Jul 8, 2026

[PyTorch] MultiheadAttention: pass packed projection output to DPA declaratively

pggPL
and
claude
committed
5f70d12

Commits on Jul 9, 2026

[PyTorch] DotProductAttention: factor packed-input handling into _unpack_packed_qkv
pggPL
committed
5713048
[PyTorch] DotProductAttention: validate packed-input last-dim stride
pggPL
committed
2255bcc
[PyTorch] DotProductAttention: pass packed qkv/kv buffers as separate arguments
pggPL
committed
dc4469b
[PyTorch] Move packed qkv/kv input tests into test_attention.py
pggPL
committed
f82ba70
[PyTorch] Deprecate pointer-based detection of packed qkv layouts
pggPL
committed
5cd0351
[PyTorch] Fix implicit string concatenation lint warning
pggPL
committed
361a2a0
[PyTorch] Trim packed-input test section to equivalence and MHA tests
pggPL
committed
627dd3f

Commits on Jul 10, 2026

Commits on Jul 14, 2026

Commits on Jul 15, 2026

Commits on Jul 17, 2026

[PyTorch] Add optional caller-provided output/grad-input buffers to GroupedLinear and fused grouped MLP (#3161)

authored
68493d2

Commits on Jul 20, 2026

Fix FusedAdam empty tensor handling (#3212)

wujingyue
and
vthumbe1503
authored
868d8d9

Commits on Jul 21, 2026

Commits on Jul 22, 2026

Commits on Jul 23, 2026