diff --git a/fastdeploy/model_executor/layers/backends/xpu/moe/fused_moe.py b/fastdeploy/model_executor/layers/backends/xpu/moe/fused_moe.py
index 89af18a1b01..b8a8b6457d0 100644
--- a/fastdeploy/model_executor/layers/backends/xpu/moe/fused_moe.py
+++ b/fastdeploy/model_executor/layers/backends/xpu/moe/fused_moe.py
@@ -644,7 +644,7 @@ def apply_tp(
             layer.down_proj_weight,
             None,  # moe_ffn1_bias
             None,  # moe_ffn2_bias
-            getattr(layer, "up_gate_proj_in_scale", None),
+            (ffn1_act_scale_per_token if hasattr(layer, "up_gate_proj_in_scale") else None),
             getattr(layer, "down_proj_in_scale", None),
             getattr(layer, "up_gate_proj_weight_scale", None),
             getattr(layer, "down_proj_weight_scale", None),