[Chore] Fix torch precision warning (#30428)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-12-10 23:05:56 -05:00
parent 7e24e5d4d6
commit d6464f2679
3 changed files with 9 additions and 7 deletions
--- a/tests/v1/e2e/test_async_scheduling.py
+++ b/tests/v1/e2e/test_async_scheduling.py
@@ -152,8 +152,8 @@ def run_tests(
                m.setenv("VLLM_ATTENTION_BACKEND", "ROCM_AITER_FA")
        else:
            m.setenv("VLLM_ATTENTION_BACKEND", "FLEX_ATTENTION")
-        # lock matmul precision to full FP32
-        m.setenv("VLLM_FLOAT32_MATMUL_PRECISION", "highest")
+        # lock matmul precision to full FP32 (IEEE)
+        m.setenv("VLLM_FLOAT32_MATMUL_PRECISION", "ieee")
        # m.setenv("VLLM_BATCH_INVARIANT", "1")
        outputs: list[tuple[str, list, list]] = []
        for n, (