[MoE Refactor] Remove SharedFusedMoE class (#35782)

Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2026-04-21 18:12:12 -04:00
committed by GitHub
parent 1842447c09
commit 5e584ce9ec
33 changed files with 112 additions and 141 deletions
+3 -3
View File
@@ -40,7 +40,7 @@ from vllm.distributed import (
)
from vllm.model_executor.layers.activation import SiluAndMul
from vllm.model_executor.layers.attention import Attention
from vllm.model_executor.layers.fused_moe import SharedFusedMoE
from vllm.model_executor.layers.fused_moe import FusedMoE
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.linear import (
MergedColumnParallelLinear,
@@ -155,7 +155,7 @@ class Dots1MoE(nn.Module):
else:
self.shared_experts = None
self.experts = SharedFusedMoE(
self.experts = FusedMoE(
shared_experts=self.shared_experts,
num_experts=config.n_routed_experts,
top_k=config.num_experts_per_tok,
@@ -413,7 +413,7 @@ class Dots1Model(nn.Module):
return hidden_states
def get_expert_mapping(self) -> list[tuple[str, str, int, str]]:
return SharedFusedMoE.make_expert_params_mapping(
return FusedMoE.make_expert_params_mapping(
self,
ckpt_gate_proj_name="gate_proj",
ckpt_down_proj_name="down_proj",