[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 -7
View File
@@ -37,7 +37,7 @@ from vllm.distributed.parallel_state import (
get_eplb_group,
)
from vllm.forward_context import set_forward_context
from vllm.model_executor.layers.fused_moe import FusedMoE, SharedFusedMoE, fused_experts
from vllm.model_executor.layers.fused_moe import FusedMoE, fused_experts
from vllm.model_executor.layers.fused_moe.activation import MoEActivation
from vllm.model_executor.layers.fused_moe.config import FusedMoEQuantConfig
from vllm.model_executor.layers.fused_moe.router.router_factory import (
@@ -858,11 +858,7 @@ def make_fused_moe_layer(
quant_config, qw = make_quant_config(quantization, w1, w2, global_num_experts)
kwargs = dict()
if shared_experts is None:
builder = FusedMoE
else:
builder = SharedFusedMoE
kwargs["shared_experts"] = shared_experts
kwargs["shared_experts"] = shared_experts
# Add gate and routed_input_transform if provided
if gate is not None:
@@ -872,7 +868,7 @@ def make_fused_moe_layer(
kwargs["routed_input_transform"] = routed_input_transform
kwargs["routed_output_transform"] = routed_output_transform
layer = builder(
layer = FusedMoE(
num_experts=global_num_experts,
top_k=top_k,
hidden_size=hidden_size,