[MoE Refactor] Separate Router into OO Classes (#30623)

Signed-off-by: Bill Nell <bnell@redhat.com>
2026-01-18 11:40:49 -05:00
parent 2f03035a61
commit 327a02d8db
45 changed files with 1750 additions and 684 deletions
--- a/vllm/model_executor/layers/quantization/fp8.py
+++ b/vllm/model_executor/layers/quantization/fp8.py
@@ -23,13 +23,13 @@ from vllm.model_executor.layers.fused_moe import (
    FusedMoEMethodBase,
    FusedMoEPermuteExpertsUnpermute,
    FusedMoEPrepareAndFinalize,
+    FusedMoERouter,
    FusedMoeWeightScaleSupported,
 )
 from vllm.model_executor.layers.fused_moe.config import (
    FusedMoEQuantConfig,
    RoutingMethodType,
 )
-from vllm.model_executor.layers.fused_moe.fused_moe_router import FusedMoERouter
 from vllm.model_executor.layers.fused_moe.layer import UnquantizedFusedMoEMethod
 from vllm.model_executor.layers.fused_moe.oracle.fp8 import (
    Fp8MoeBackend,