onnxruntime
Fix CPU Attention softcap/attn_mask ordering (onnx#7867, #7913) + consolidate CUDA spec coverage tests from #28371
#28379
Merged
Merged
Fix CPU Attention softcap/attn_mask ordering (onnx#7867, #7913) + consolidate CUDA spec coverage tests from #28371 #28379
titaiwangms
merged
15 commits into
microsoft:main
from
titaiwangms:fix-cpu-attention-softcap-mask-ordering
test: add CPU-EP guards for ONNX #7867 softcap/mask ordering (canary)
d76e45ca
fix(cpu): apply softcap before mask/bias in ONNX Attention (onnx#7867…
cef55ca5
Address PR-1-v2 review findings: oracle-verify regen, DRY mask_was_fo…
3dfc70fe
titaiwangms
commented
on 2026-05-05
tianleiwu
changed the title
tianleiwu
commented
on 2026-05-06
tianleiwu
commented
on 2026-05-06
Add mode-1+softcap and nonpad+softcap tests; address bot review minors
87c7e897
Skip pre-#7867/#7913 ONNX node-test fixtures pending bundled-ONNX update
0cd61f9c
Skip pre-onnx#7867/#7913 attention CPU node-tests in Python runner fi…
1ff6e0bd
Move output_qk plumbing + masked fp32 softcap ordering tests from #28371
d6139665
Consolidate PR #28371 (CUDA Attention spec coverage tests + SKILL.md …
bfe33d3d
titaiwangms
changed the title
Address bot review (kNone in error message, drop internal path) + add…
37a19611
Drop internal-coordination jargon from new test docstring/comment
a255941c
Fix stale ONNX PR ref in common.py comment (#7865 -> #7867)
0b4c75a2
Add cmake plumbing for transformers/test_onnx_attention/ subpackage
1bbaeac9
Add CPU twins for 6 EP-agnostic Tier A test classes in test_onnx_atte…
a8225677
Fix IOBinding race + disable obsolete determinism check in test_onnx_…
b2062c0d
Remove dead determinism-check machinery in test_onnx_attention helpers
7dfa7731
tianleiwu
approved these changes
on 2026-05-07
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub