fix: add missing clamp_limit for CompressedTensorsWNA16MoE (#27328)

2026-06-16 19:29:57 -07:00
parent 9c53853ea3
commit 0ae4740bd1
1 changed files with 1 additions and 0 deletions
@@ -425,6 +425,7 @@ class CompressedTensorsWNA16MoE(CompressedTensorsMoEScheme):
            num_bits=self.num_bits,
            is_k_full=self.is_k_full,
            routed_scaling_factor=self.moe_runner_config.routed_scaling_factor,
+            clamp_limit=self.moe_runner_config.swiglu_limit,
            workspace=layer.workspace,
        )
        return StandardCombineInput(hidden_states=output)