[BugFix] Fix routed_scaling_factor double mul for dots1 and glm4 MoE models (vllm-project#24132)

Signed-off-by: Yong Hoon Shin <[email protected]>

Author: sarckk

vllm/model_executor/models/dots1.py

@@ -137,7 +137,8 @@ def __init__(
             topk_group=config.topk_group,
             prefix=f"{prefix}.experts",
             scoring_func=config.scoring_func,
-            routed_scaling_factor=self.routed_scaling_factor,
+            # we do scaling outside, set factor to 1.0 to avoid double mul
+            routed_scaling_factor=1.0,
             e_score_correction_bias=self.gate.e_score_correction_bias)
 
         if config.n_shared_experts is not None:

vllm/model_executor/models/glm4_moe.py

@@ -159,7 +159,8 @@ def __init__(
             topk_group=config.topk_group,
             prefix=f"{prefix}.experts",
             scoring_func="sigmoid",
-            routed_scaling_factor=self.routed_scaling_factor,
+            # we do scaling outside, set factor to 1.0 to avoid double mul
+            routed_scaling_factor=1.0,
             e_score_correction_bias=self.gate.e_score_correction_bias,
             enable_eplb=self.enable_eplb,
             num_redundant_experts=self.n_redundant_experts)