[Kernel][Performance] Add FlashInfer cutedsl NVFP4 GEMM backend (#42235)

Signed-off-by: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com>
This commit is contained in:
Mohammad Miadh Angkad
2026-06-22 16:17:18 -04:00
committed by GitHub
parent 2b4a7491ec
commit d1a38c2762
6 changed files with 93 additions and 1 deletions
+7
View File
@@ -90,6 +90,7 @@ def test_models(example_prompts, model_name) -> None:
EAGER = [True, False]
SM_100_NVFP4_BACKENDS = [
"flashinfer_cutedsl",
"flashinfer_cudnn",
"flashinfer_trtllm",
"flashinfer_cutlass",
@@ -102,12 +103,18 @@ SM_100_NVFP4_BACKENDS = [
"backend",
[
"emulation",
"flashinfer_cutedsl",
"flashinfer_cudnn",
"flashinfer_trtllm", # the small seq_len ensures trtllm_8x4_layout backend is used
"flashinfer_cutlass",
],
)
def test_nvfp4(vllm_runner, model, eager, backend):
if backend == "flashinfer_cutedsl" and not (
current_platform.is_device_capability_family(100)
):
pytest.skip("The flashinfer_cutedsl backend is only supported on SM10x")
if (
not current_platform.has_device_capability(100)
and backend in SM_100_NVFP4_BACKENDS