diff --git a/vllm_mindspore/model_executor/models/mf_models/deepseek_v3.py b/vllm_mindspore/model_executor/models/mf_models/deepseek_v3.py
index 2a6b6dd5c7bc7bce15c64b193ad3abc00b5411f4..4b9b3fac3029883eb763648c7cf3c1bcd2da0840 100644
--- a/vllm_mindspore/model_executor/models/mf_models/deepseek_v3.py
+++ b/vllm_mindspore/model_executor/models/mf_models/deepseek_v3.py
@@ -146,7 +146,7 @@ class DeepseekV3ForCausalLM(MfModelBase):
         self.mf_model_config = DeepseekV3Config_MF(**self.mf_config.model.model_config)
         if self.mf_config.moe_config:
             self.mf_model_config.moe_config = self.mf_config.moe_config
-            # dispatch/combine in moe need max_num_seqs as global_max_bs
+            # dispatch/combine in moe need dispatch_global_max_bs as global_max_bs
             if hasattr(self.mf_model_config.moe_config, "dispatch_global_max_bs"):
                 self.mf_model_config.moe_config.dispatch_global_max_bs = self.scheduler_config.max_num_batched_tokens
         self.mf_model_config.return_hidden_states = True