[MoE Refactor][13/N] Convert FI to Use PFNoEP (#31533)

Signed-off-by: Robert Shaw <robshaw@redhat.com>
Signed-off-by: Robert Shaw <robertgshaw2@gmail.com>
Signed-off-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
Co-authored-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
This commit is contained in:
Robert Shaw
2026-01-03 12:26:36 -08:00
committed by GitHub
co-authored by Robert Shaw Wentao Ye
parent 4f9ce35afe
commit 268b1c55ad
7 changed files with 79 additions and 169 deletions
@@ -46,7 +46,6 @@ from vllm.model_executor.layers.quantization.utils.flashinfer_fp4_moe import (
from vllm.model_executor.layers.quantization.utils.flashinfer_utils import (
FlashinferMoeBackend,
apply_flashinfer_per_tensor_scale_fp8,
build_flashinfer_fp8_cutlass_moe_prepare_finalize,
flashinfer_cutlass_moe_fp8,
get_flashinfer_moe_backend,
is_flashinfer_supporting_global_sf,
@@ -751,14 +750,7 @@ class ModelOptFp8MoEMethod(FusedMoEMethodBase):
# TRT LLM not supported with all2all yet.
if self.flashinfer_moe_backend == FlashinferMoeBackend.TENSORRT_LLM:
return None
elif self.flashinfer_moe_backend == FlashinferMoeBackend.CUTLASS:
prepare_finalize = build_flashinfer_fp8_cutlass_moe_prepare_finalize(
self.moe
)
logger.debug_once("%s", prepare_finalize.__class__.__name__)
return prepare_finalize
else:
return super().maybe_make_prepare_finalize(routing_tables)
return super().maybe_make_prepare_finalize(routing_tables)
def select_gemm_impl(
self,