xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-30 00:07:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix inductor tf32 setting error

Browse Source 
Signed-off-by: zhuhaoran <zhuhaoran.zhr@alibaba-inc.com>
This commit is contained in:
zhuhaoran 2025-12-13 18:09:36 +08:00
parent c5df2565ab
commit b4d755ac05
1 changed files with 2 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tests/v1/e2e/test_async_scheduling.py
View File

@ -4,6 +4,7 @@ from itertools import repeat
from typing import Any from typing import Any
import pytest import pytest
import torch
import torch._dynamo.config as dynamo_config import torch._dynamo.config as dynamo_config
from vllm import SamplingParams from vllm import SamplingParams
@ -158,6 +159,7 @@ def run_tests(
m.setenv("VLLM_ATTENTION_BACKEND", "FLEX_ATTENTION") m.setenv("VLLM_ATTENTION_BACKEND", "FLEX_ATTENTION")
# lock matmul precision to full FP32 (IEEE) # lock matmul precision to full FP32 (IEEE)
m.setenv("VLLM_FLOAT32_MATMUL_PRECISION", "ieee") m.setenv("VLLM_FLOAT32_MATMUL_PRECISION", "ieee")
torch.backends.cuda.matmul.allow_tf32 = False
# m.setenv("VLLM_BATCH_INVARIANT", "1") # m.setenv("VLLM_BATCH_INVARIANT", "1")
outputs: list[tuple[str, list, list]] = [] outputs: list[tuple[str, list, list]] = []
for n, ( for n, (