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

[MoE Refactor] Oracle Select FP8+NVFP4 Kernels In Priority (#32414)

Browse Source
This commit is contained in:
Robert Shaw
2026-01-21 08:22:33 -05:00
committed by GitHub
parent e14467be43
commit 42135d6898
82 changed files with 2710 additions and 1563 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/model_executor/models/qwen3_moe.py
View File

@@ -43,7 +43,6 @@ 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 FusedMoE
from vllm.model_executor.layers.fused_moe.config import RoutingMethodType
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.linear import (
MergedColumnParallelLinear,
@@ -172,7 +171,6 @@ class Qwen3MoeSparseMoeBlock(nn.Module):
enable_eplb=self.enable_eplb,
num_redundant_experts=self.n_redundant_experts,
is_sequence_parallel=self.is_sequence_parallel,
routing_method_type=RoutingMethodType.Renormalize,
)
self.gate = ReplicatedLinear(