Update modeling_glm4_moe.py

modeling_glm4_moe.py

@@ -356,7 +356,7 @@ class Glm4MoeDecoderLayer(GradientCheckpointingLayer):
             route_scale=config.routed_scaling_factor,
             score_before_experts=False,
             top_k=config.num_experts_per_tok,
-            use_grouped_mm=torch.cuda.get_device_capability(0)[0] >= 9
+            use_grouped_mm=torch.cuda.get_device_capability(0)[0] >= 9,
             load_balance_coeff=1e-3,
         )