Skip to content

vllm.entrypoints.pooling.embed.conftest

Pytest configuration for vLLM pooling embed tests.

pytest_collection_modifyitems 

pytest_collection_modifyitems(config, items)

Configure ROCm-specific settings based on collected tests.

Source code in vllm/entrypoints/pooling/embed/conftest.py 
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
def pytest_collection_modifyitems(config, items):
    """Configure ROCm-specific settings based on collected tests."""
    if not current_platform.is_rocm():
        return

    # Disable Flash/MemEfficient SDP on ROCm to avoid HF Transformers
    # accuracy issues: https://github.com/vllm-project/vllm/issues/30167
    # TODO: Remove once ROCm SDP accuracy issues are resolved on HuggingFace
    torch.backends.cuda.enable_flash_sdp(False)
    torch.backends.cuda.enable_mem_efficient_sdp(False)
    torch.backends.cuda.enable_math_sdp(True)
    warnings.warn(
        "ROCm: Disabled flash_sdp and mem_efficient_sdp, enabled math_sdp "
        "to avoid HuggingFace Transformers accuracy issues",
        UserWarning,
        stacklevel=1,
    )