A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 3, 19:00 UTC
[CI] Fix `test_async_scheduling.py` flakiness (#42455)
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Nick Hill committed
May 12, 2026 at 14:38 UTC
fe8b42e80c482ea6d93bc5f800109528f2f11e31
1 file changed
+1
-1
tests/v1/e2e/general/test_async_scheduling.py
+1
-1
@@ -429,7 +429,7 @@ def _logprobs_match(
429
and lps_a.keys() == lps_b.keys()
430
and all(
431
a.decoded_token == b.decoded_token
432
- and a.rank == b.rank
432
+ and a.rank == pytest.approx(b.rank, rel=0.005)
433
and a.logprob == pytest.approx(b.logprob, rel=rel_tol, abs=abs_tol)
434
for a, b in ((lps_a[x], lps_b[x]) for x in lps_a)
435
)