Commit 9ca82f8
Fix backward_dense health-check regression under the ci profile
Summary:
test_backward_dense started failing in OSS CI on 2026-06-15, green before
that. The break is D108096684, which removed the per-test
`suppress_health_check=[HealthCheck.filter_too_much, HealthCheck.data_too_large]`
from `backward_dense_test.py` and moved the suppressions into the
"derandomize" Hypothesis profile registered in `tbe/common.py`.
The problem: OSS CI runs under the "ci" Hypothesis profile, not
"derandomize". The job log shows:
hypothesis profile 'ci' -> ... suppress_health_check=(HealthCheck.too_slow,)
so the profile-level suppressions never take effect in OSS. test_backward_dense
makes heavy use of assume() (T*B*L*D<=2048, FP16/pooling guards, etc.), which
trips HealthCheck.filter_too_much once the suppression is gone, failing the
test.
Fix: apply the suppressions explicitly in the test's settings via the
centralized `suppressed_list` from `tbe/common.py`. This keeps a single source
of truth (the list still lives in common.py, including the conditional
differing_executors entry) while making the suppression independent of which
profile is loaded.
Differential Revision: D1091649071 parent 449ade1 commit 9ca82f8
1 file changed
Lines changed: 8 additions & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
32 | 32 | | |
33 | 33 | | |
34 | 34 | | |
| 35 | + | |
35 | 36 | | |
36 | 37 | | |
37 | 38 | | |
| |||
78 | 79 | | |
79 | 80 | | |
80 | 81 | | |
| 82 | + | |
| 83 | + | |
| 84 | + | |
| 85 | + | |
| 86 | + | |
| 87 | + | |
| 88 | + | |
81 | 89 | | |
82 | 90 | | |
83 | 91 | | |
| |||
0 commit comments