[CI] Fix MXFP8 MOE backend selection tests on gfx942 (#50222)

Signed-off-by: Felix Marty <Felix.Marty@amd.com>
This commit is contained in:
fxmarty-amd
2026-07-29 17:17:11 +08:00
committed by GitHub
parent dad7a6383b
commit 5b14019576
@@ -27,8 +27,8 @@ from vllm.model_executor.layers.fused_moe.experts.aiter_mxfp8_moe import ( # no
_AITER_SWIGLU_BETA,
AiterMxfp8Experts,
)
from vllm.model_executor.layers.fused_moe.experts.mxfp8_native_moe import ( # noqa: E402
Mxfp8NativeTritonExperts,
from vllm.model_executor.layers.fused_moe.experts.mxfp8_emulation_moe import ( # noqa: E402
Mxfp8EmulationTritonExperts,
)
from vllm.model_executor.layers.fused_moe.modular_kernel import ( # noqa: E402
FusedMoEActivationFormat,
@@ -150,16 +150,20 @@ def test_explicit_moe_backend_aiter():
def test_gfx950_picks_aiter():
"""Auto-select on real ROCm hardware with flydsl usable -> FlyDSL wins."""
with _flydsl_installed(True):
# NOTE: Fp8MoeBackend.AITER_MXFP8 does not require VLLM_ROCM_USE_AITER=1
with (
patch(f"{_AITER_MOD}.current_platform.supports_mx", return_value=True),
_flydsl_installed(True),
):
backend, experts_cls = select_mxfp8_moe_backend(_config())
assert backend is Fp8MoeBackend.AITER_MXFP8
assert experts_cls is AiterMxfp8Experts
def test_gfx942_picks_triton():
def test_gfx942_picks_emulation():
"""flydsl unusable (e.g. gfx942, no FlyDSL support) -> native Triton
dot_scaled backend wins instead."""
with _flydsl_installed(False):
with patch(f"{_AITER_MOD}.current_platform.supports_mx", return_value=False):
backend, experts_cls = select_mxfp8_moe_backend(_config())
assert backend is Fp8MoeBackend.TRITON_MXFP8
assert experts_cls is Mxfp8NativeTritonExperts
assert backend is Fp8MoeBackend.EMULATION
assert experts_cls is Mxfp8EmulationTritonExperts