You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[BUG] Ascend: the flash-attention correctness test uses the same full-size shapes as the benchmark #4380
I have searched the Issue Tracker that this hasn't already been reported. (comment there if it has.)
What version of TileLang are you using?
0.1.15+cuda.git210611d
System information
Ascend 950PR, CANN 9.0.0-beta.2, Linux 6.6.0-13, Python 3.12. Source build of main @ 210611d.
Problem description
// examples/ascend/README.md:45-46
Each top-level example_<name>.py has one test_<name>.py. Correctness tests use
smaller inputs than performance runs.
The two forward tests use the performance shapes verbatim:
test_gqa_bwd.py:8 and all five top-level tests do use reduced shapes, so the convention is real and only these two break it. The consequence is practical: pytest examples/ascend spends almost all its time in these two tests, and a "correctness" run costs as much as a benchmark.
Activity
cn-sys commentedon Nov 15, 2026
/formal