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

Add routed_scaling_factor to MoE grouped topk (#23123)

Browse Source 
Signed-off-by: Xin Yang <xyangx@amazon.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
This commit is contained in:
Xin Yang
2025-08-29 21:36:48 -07:00
committed by GitHub
parent 5b31cb1781
commit 8fb85b7bb6
19 changed files with 77 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/model_executor/layers/quantization/mxfp4.py
View File

@@ -546,6 +546,7 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
expert_map: Optional[torch.Tensor] = None,
custom_routing_function: Optional[Callable] = None,
scoring_func: str = "softmax",
routed_scaling_factor: float = 1.0,
e_score_correction_bias: Optional[torch.Tensor] = None,
apply_router_weight_on_input: bool = False,
activation: str = "silu",
@@ -569,6 +570,7 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):
num_expert_group=num_expert_group,
custom_routing_function=custom_routing_function,
scoring_func=scoring_func,
routed_scaling_factor=routed_scaling_factor,
e_score_correction_bias=e_score_correction_bias)
return torch.ops.vllm.fused_marlin_moe(