[MoE Refactor] Remove SharedFusedMoE class (#35782)
Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
@@ -18,7 +18,7 @@ from vllm.distributed import (
|
||||
)
|
||||
from vllm.logger import init_logger
|
||||
from vllm.model_executor.layers.attention import Attention
|
||||
from vllm.model_executor.layers.fused_moe.shared_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 (
|
||||
ColumnParallelLinear,
|
||||
@@ -124,8 +124,8 @@ class AfmoeMoE(nn.Module):
|
||||
prefix=f"{prefix}.shared_experts",
|
||||
)
|
||||
|
||||
# Routed experts using SharedFusedMoE
|
||||
self.experts = SharedFusedMoE(
|
||||
# Routed experts using FusedMoE
|
||||
self.experts = FusedMoE(
|
||||
shared_experts=self.shared_experts,
|
||||
num_experts=config.num_experts,
|
||||
top_k=config.num_experts_per_tok,
|
||||
@@ -479,7 +479,7 @@ class AfmoeModel(nn.Module, EagleModelMixin):
|
||||
def get_expert_mapping(self) -> list[tuple[str, str, int, str]]:
|
||||
# Params for weights, fp8 weight scales, fp8 activation scales
|
||||
# (param_name, weight_name, expert_id, shard_id)
|
||||
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",
|
||||
@@ -637,7 +637,7 @@ class AfmoeForCausalLM(nn.Module, SupportsPP, SupportsEagle3, SupportsLoRA):
|
||||
self.num_moe_layers = config.num_hidden_layers - config.num_dense_layers
|
||||
self.num_expert_groups = config.n_group
|
||||
|
||||
self.moe_layers: list[SharedFusedMoE] = []
|
||||
self.moe_layers: list[FusedMoE] = []
|
||||
example_moe = None
|
||||
for layer in self.model.layers:
|
||||
if isinstance(layer, PPMissingLayer):
|
||||
|
||||
Reference in New Issue
Block a user