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

7
tests/kernels/moe/test_routing.py
View File

@@ -6,7 +6,6 @@ import pytest
import torch
from vllm.distributed.eplb.eplb_state import EplbLayerState
from vllm.model_executor.layers.fused_moe.config import RoutingMethodType
from vllm.model_executor.layers.fused_moe.router.router_factory import (
create_fused_moe_router,
)
@@ -385,17 +384,11 @@ def test_grouped_topk(
global_num_experts,
)
routing_method_type = None
if scoring_func == "llama4":
routing_method_type = RoutingMethodType.Llama4
scoring_func = "sigmoid"
router = create_fused_moe_router(
use_grouped_topk=True,
num_expert_group=num_expert_group,
topk_group=topk_group,
scoring_func=scoring_func,
routing_method_type=routing_method_type,
e_score_correction_bias=e_score_correction_bias,
routed_scaling_factor=routed_scaling_factor,
top_k=top_k,