diff --git a/examples/pytorch/chatglm6b/finetune.py b/examples/pytorch/chatglm6b/finetune.py
index bf3953e8..6d04924c 100644
--- a/examples/pytorch/chatglm6b/finetune.py
+++ b/examples/pytorch/chatglm6b/finetune.py
@@ -196,7 +196,7 @@ tokenizer = ChatGLMTokenizer.from_pretrained(model_dir, trust_remote_code=True)
 device_map_kwargs = {}
 device_kwargs = {}
 if args.use_lora != 0:
-    device_kwargs['device_map'] = 'auto'
+    device_map_kwargs['device_map'] = 'auto'
     # No placement for model, leave the model to `device_map`
     device_kwargs['device'] = 'cpu'
 model = Model.from_pretrained(
diff --git a/modelscope/models/nlp/chatglm/text_generation.py b/modelscope/models/nlp/chatglm/text_generation.py
index 53f8f105..64b82862 100644
--- a/modelscope/models/nlp/chatglm/text_generation.py
+++ b/modelscope/models/nlp/chatglm/text_generation.py
@@ -931,7 +931,6 @@ class ChatGLMModel(ChatGLMPreTrainedModel):
             self.num_attention_heads,
             self.hidden_size // self.num_attention_heads)
         # seq_len, b, nh, hidden_size
-        print('#########################:', past_key_values.device)
         past_key_values = self.dropout(past_key_values)
         past_key_values = past_key_values.permute([2, 1, 0, 3, 4]).split(2)
         # past_key_values = [(v[0], v[1]) for v in past_key_values]
diff --git a/modelscope/pipelines/nlp/text_generation_pipeline.py b/modelscope/pipelines/nlp/text_generation_pipeline.py
index a7806702..cb9c89db 100644
--- a/modelscope/pipelines/nlp/text_generation_pipeline.py
+++ b/modelscope/pipelines/nlp/text_generation_pipeline.py
@@ -197,8 +197,8 @@ class ChatGLM6bTextGenerationPipeline(Pipeline):
         if isinstance(model, str):
             model_dir = snapshot_download(
                 model) if not os.path.exists(model) else model
-            config = ChatGLMConfig.from_pretrained(model_dir)
-            model = ChatGLMForConditionalGeneration(config).half()
+            model = ChatGLMForConditionalGeneration.from_pretrained(
+                model_dir).half()
             if torch.cuda.is_available():
                 model = model.cuda()
         if quantization_bit is not None:
@@ -235,8 +235,7 @@ class ChatGLM6bV2TextGenerationPipeline(Pipeline):
         if isinstance(model, str):
             model_dir = snapshot_download(
                 model) if not os.path.exists(model) else model
-            config = ChatGLM2Config.from_pretrained(model_dir)
-            model = ChatGLM2ForConditionalGeneration(config)
+            model = ChatGLM2ForConditionalGeneration.from_pretrained(model_dir)
             if torch.cuda.is_available():
                 model = model.cuda()
         if quantization_bit is not None: