[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
@@ -13,6 +13,7 @@ from vllm.model_executor.layers.fused_moe import (
FusedMoE,
FusedMoEConfig,
FusedMoEMethodBase,
FusedMoERouter,
FusedMoeWeightScaleSupported,
)
from vllm.model_executor.layers.fused_moe.config import (
@@ -350,10 +351,11 @@ class QuarkW8A8Fp8MoEMethod(QuarkMoEMethod):
def apply(
self,
layer: FusedMoE,
router: FusedMoERouter,
x: torch.Tensor,
router_logits: torch.Tensor,
) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor]:
topk_weights, topk_ids = layer.select_experts(
topk_weights, topk_ids = router.select_experts(
hidden_states=x,
router_logits=router_logits,
)
@@ -542,6 +544,7 @@ class QuarkW4A8Fp8MoEMethod(QuarkMoEMethod):
def apply(
self,
layer: FusedMoE,
router: FusedMoERouter,
x: torch.Tensor,
router_logits: torch.Tensor,
) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor]:
@@ -750,10 +753,11 @@ class QuarkOCP_MX_MoEMethod(QuarkMoEMethod):
def apply(
self,
layer: FusedMoE,
router: FusedMoERouter,
x: torch.Tensor,
router_logits: torch.Tensor,
) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor]:
topk_weights, topk_ids = layer.select_experts(
topk_weights, topk_ids = router.select_experts(
hidden_states=x,
router_logits=router_logits,
)