biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[MoE][Refactor] Make select_experts a non-static method (#29067)

Browse Source 
Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2025-11-24 13:38:04 -05:00
committed by GitHub
parent cec418b5df
commit 8f066146c3
18 changed files with 163 additions and 472 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
vllm/model_executor/layers/quantization/mxfp4.py
View File

@@ -862,7 +862,7 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
def apply(
self,
layer: torch.nn.Module,
layer: FusedMoE,
x: torch.Tensor,
router_logits: torch.Tensor,
top_k: int,
@@ -887,18 +887,9 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
raise NotImplementedError("EPLB is not supported for mxfp4")
if self.mxfp4_backend == Mxfp4Backend.MARLIN:
topk_weights, topk_ids, _ = FusedMoE.select_experts(
topk_weights, topk_ids, _ = layer.select_experts(
hidden_states=x,
router_logits=router_logits,
use_grouped_topk=use_grouped_topk,
top_k=top_k,
renormalize=renormalize,
topk_group=topk_group,
num_expert_group=num_expert_group,
custom_routing_function=custom_routing_function,
scoring_func=scoring_func,
routed_scaling_factor=routed_scaling_factor,
e_score_correction_bias=e_score_correction_bias,
)
return fused_marlin_moe(
@@ -989,17 +980,9 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
):
from vllm.utils.flashinfer import flashinfer_cutlass_fused_moe
topk_weights, topk_ids, _ = FusedMoE.select_experts(
topk_weights, topk_ids, _ = layer.select_experts(
hidden_states=x,
router_logits=router_logits,
use_grouped_topk=use_grouped_topk,
top_k=top_k,
renormalize=renormalize,
topk_group=topk_group,
num_expert_group=num_expert_group,
custom_routing_function=custom_routing_function,
scoring_func=scoring_func,
e_score_correction_bias=e_score_correction_bias,
)
# Backend-specific preparation