[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
@@ -14,7 +14,7 @@ from vllm.distributed import (
)
from vllm.logger import init_logger
from vllm.model_executor.layers.activation import SiluAndMul
from vllm.model_executor.layers.fused_moe import SharedFusedMoE
from vllm.model_executor.layers.fused_moe import FusedMoE
from vllm.model_executor.layers.kda import KimiDeltaAttention
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.linear import (
@@ -144,7 +144,7 @@ class KimiMoE(nn.Module):
else:
self.shared_experts = None
self.experts = SharedFusedMoE(
self.experts = FusedMoE(
shared_experts=self.shared_experts,
num_experts=num_experts,
top_k=config.num_experts_per_token,
@@ -476,7 +476,7 @@ class KimiLinearModel(nn.Module):
if self.config.is_moe:
# Params for weights, fp8 weight scales, fp8 activation scales
# (param_name, weight_name, expert_id, shard_id)
expert_params_mapping = SharedFusedMoE.make_expert_params_mapping(
expert_params_mapping = FusedMoE.make_expert_params_mapping(
self,
ckpt_gate_proj_name="w1",
ckpt_down_proj_name="w2",