[MoE Refactor] Introduce RoutedExperts alias for FusedMoE and don't store SharedExperts in MK (#40735)

Signed-off-by: Bill Nell <bnell@redhat.com>
Signed-off-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
Signed-off-by: Robert Shaw <robertgshaw2@gmail.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
Co-authored-by: Robert Shaw <robertgshaw2@gmail.com>
This commit is contained in:
bnellnm
2026-05-12 13:37:44 -04:00
committed by GitHub
co-authored by Robert Shaw Robert Shaw
parent 6ff7405b81
commit 4d591db470
53 changed files with 369 additions and 332 deletions
@@ -7,11 +7,11 @@ import torch
from torch.nn import Module
if TYPE_CHECKING:
from vllm.model_executor.layers.fused_moe import FusedMoE
from vllm.model_executor.layers.fused_moe.config import (
FusedMoEQuantConfig,
)
from vllm.model_executor.layers.fused_moe import RoutedExperts
from vllm.model_executor.layers.fused_moe.oracle.int8 import (
make_int8_moe_kernel,
make_int8_moe_quant_config,
@@ -91,7 +91,7 @@ class Int8OnlineMoEMethod(OnlineMoEMethodBase):
replace_parameter(layer, "w13_scale", w13_scale)
replace_parameter(layer, "w2_scale", w2_scale)
def _setup_kernel(self, layer: "FusedMoE") -> None:
def _setup_kernel(self, layer: RoutedExperts) -> None:
self.moe_quant_config = self.get_fused_moe_quant_config(layer)
assert self.moe_quant_config is not None
assert self.experts_cls is not None
@@ -100,7 +100,6 @@ class Int8OnlineMoEMethod(OnlineMoEMethodBase):
moe_config=self.moe,
experts_cls=self.experts_cls,
routing_tables=layer._expert_routing_tables(),
shared_experts=layer.shared_experts,
)
def get_fused_moe_quant_config(