diff --git a/tests/quantization/test_fp8.py b/tests/quantization/test_fp8.py index 0ae51652c6..4a2db91af9 100644 --- a/tests/quantization/test_fp8.py +++ b/tests/quantization/test_fp8.py @@ -93,9 +93,6 @@ def test_online_quantization( use_rocm_aiter: bool, monkeypatch, ) -> None: - if kv_cache_dtype == "fp8" and current_platform.is_device_capability_family(90): - pytest.skip("FA3 currently rejects FP8 KV cache output dtype on SM90") - if use_rocm_aiter: monkeypatch.setenv("VLLM_ROCM_USE_AITER", "1") @@ -105,9 +102,15 @@ def test_online_quantization( if force_marlin: monkeypatch.setenv("VLLM_TEST_FORCE_FP8_MARLIN", "1") + model_dtype = "auto" + if kv_cache_dtype == "fp8" and current_platform.is_device_capability_family(90): + # FA3 requires BF16 output when the query input is FP8. + model_dtype = "bfloat16" + with vllm_runner( "facebook/opt-125m", quantization="fp8", + dtype=model_dtype, enforce_eager=True, kv_cache_dtype=kv_cache_dtype, ) as llm: