[MoE Refactor] Create MK for TRTLLM Kernels (#32564)

Signed-off-by: Robert Shaw <robshaw@redhat.com>
Signed-off-by: Robert Shaw <rshaw@neuralmagic.com>
Signed-off-by: Robert Shaw <robertgshaw2@gmail.com>
Co-authored-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Robert Shaw <rshaw@neuralmagic.com>
This commit is contained in:
Robert Shaw
2026-03-03 10:39:50 -08:00
committed by GitHub
co-authored by Robert Shaw Robert Shaw
parent 881a6b011b
commit 97995f6376
77 changed files with 2575 additions and 2087 deletions
@@ -11,13 +11,13 @@ from vllm.model_executor.layers.fused_moe.config import FusedMoEParallelConfig
from vllm.model_executor.layers.quantization.utils.quant_utils import QuantKey
class FallbackExperts(mk.FusedMoEPermuteExpertsUnpermute, ABC):
class FallbackExperts(mk.FusedMoEExpertsModular, ABC):
"""Base class for runtime dispatching of expert implementations."""
def __init__(
self,
experts: mk.FusedMoEPermuteExpertsUnpermute,
fallback_experts: mk.FusedMoEPermuteExpertsUnpermute,
experts: mk.FusedMoEExpertsModular,
fallback_experts: mk.FusedMoEExpertsModular,
):
super().__init__(
moe_config=experts.moe_config, quant_config=experts.quant_config
@@ -27,8 +27,8 @@ class FallbackExperts(mk.FusedMoEPermuteExpertsUnpermute, ABC):
@staticmethod
def get_clses() -> tuple[
type[mk.FusedMoEPermuteExpertsUnpermute],
type[mk.FusedMoEPermuteExpertsUnpermute],
type[mk.FusedMoEExpertsModular],
type[mk.FusedMoEExpertsModular],
]:
"""
Get the cls for the experts and fallback experts.
@@ -149,7 +149,7 @@ class FallbackExperts(mk.FusedMoEPermuteExpertsUnpermute, ABC):
hidden_states: torch.Tensor,
w1: torch.Tensor,
w2: torch.Tensor,
) -> mk.FusedMoEPermuteExpertsUnpermute:
) -> mk.FusedMoEExpertsModular:
raise NotImplementedError
def apply(