[MoE Refactor] FusedMoE/MoERunner inversion refactor (#41184)

Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2026-06-08 10:42:58 -04:00
committed by GitHub
parent 753e9d55e6
commit dc68bd8c41
90 changed files with 2738 additions and 2031 deletions
+2 -1
View File
@@ -37,6 +37,7 @@ from vllm.distributed import (
from vllm.model_executor.layers.activation import SiluAndMul
from vllm.model_executor.layers.fused_moe import (
FusedMoE,
MoERunner,
fused_moe_make_expert_params_mapping,
)
from vllm.model_executor.layers.layernorm import RMSNorm
@@ -355,7 +356,7 @@ class SarvamMLAMoE(nn.Module):
routed_scaling_factor=self.routed_scaling_factor,
)
def maybe_get_fused_moe(self) -> FusedMoE:
def maybe_get_fused_moe(self) -> MoERunner:
return self.experts
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor: