[Misc][Refactor] Add FusedMoERouter object (#30519)

Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2026-01-08 20:52:55 +00:00
committed by GitHub
parent aa125ecf0e
commit e74698c27a
20 changed files with 165 additions and 36 deletions
@@ -15,6 +15,7 @@ from vllm.model_executor.layers.fused_moe.config import (
FusedMoEQuantConfig,
int8_w8a16_moe_quant_config,
)
from vllm.model_executor.layers.fused_moe.fused_moe_router import FusedMoERouter
from vllm.model_executor.layers.linear import LinearBase, UnquantizedLinearMethod
from vllm.model_executor.layers.quantization import QuantizationMethods
from vllm.model_executor.layers.quantization.base_config import (
@@ -137,12 +138,13 @@ class ExpertsInt8MoEMethod(FusedMoEMethodBase):
def apply(
self,
layer: FusedMoE,
router: FusedMoERouter,
x: torch.Tensor,
router_logits: torch.Tensor,
) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor]:
from vllm.model_executor.layers.fused_moe import fused_experts
topk_weights, topk_ids = layer.select_experts(
topk_weights, topk_ids = router.select_experts(
hidden_states=x,
router_logits=router_logits,
)