iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[CI][Test] Allow 10% ngram acceptance drop under preemption on CUDA in test_async_scheduling_accuracy (#60715)
pre-commit / pre-run-check (push) Canceled after 0s
Details
pre-commit / pre-commit (push) Canceled after 0s
Details

Browse Source 
Signed-off-by: Andrey Talman <atalman@users.noreply.github.com>
Co-authored-by: Andrey Talman <atalman@users.noreply.github.com>
This commit is contained in:
Andrey Talman
2026-10-09 16:59:17 +00:00
committed by GitHub
co-authored by Andrey Talman
parent b74233860d
commit 89ac8d971b
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/v1/e2e/general/test_async_scheduling_accuracy.py
+3 -3
View File
@@ -271,9 +271,9 @@ def run_tests(
context = f"config=[{config}], params={params}" context = f"config=[{config}], params={params}"
if spec_config.get("max_model_len") is None: if spec_config.get("max_model_len") is None:
# Keep the original acceptance-quality floor per batch. # Keep the original acceptance-quality floor per batch.
relative_drop = ( # Preemption recomputes requests and loses n-gram drafts,
0.10 if current_platform.is_rocm() and test_preemption else 0.05 # so allow a larger drop there.
) relative_drop = 0.10 if test_preemption else 0.05
assert actual >= baseline * (1 - relative_drop), ( assert actual >= baseline * (1 - relative_drop), (
f"{context}: acceptance={actual}, baseline={baseline}" f"{context}: acceptance={actual}, baseline={baseline}"
) )