xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-01-02 14:20:54 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix test failure for batch invariance

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
yewentao256 2025-10-16 21:47:56 +00:00
parent fb0571b077
commit e958c583e7
1 changed files with 17 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
tests/v1/generation/test_batch_invariance.py
View File

@ -516,8 +516,23 @@ def test_logprobs_WITHOUT_batch_invariance_should_FAIL(backend):
dtype="bfloat16",
)
# Use more realistic prompts for better token generation
prompts = [_random_prompt(10, 50) for i in range(32)]
# build ragged prompts to change shapes significantly across BS=1 vs BS=N
long_min = int(os.getenv("VLLM_MIN_PROMPT", "768"))
long_max = int(os.getenv("VLLM_MAX_PROMPT", "2048"))
prompts: list[str] = []
for i in range(32):
if i % 4 == 0:
# very long
prompts.append(_random_prompt(max(long_min, 1536), max(long_max, 3072)))
elif i % 4 == 1:
# long
prompts.append(_random_prompt(max(1024, long_min), max(2048, long_max)))
elif i % 4 == 2:
# mid
prompts.append(_random_prompt(256, 512))
else:
# short
prompts.append(_random_prompt(10, 20))
sp = SamplingParams(
temperature=0.6,