fix pt deploy lora (#999)

Author: Jintao-Huang

swift/llm/infer.py

@@ -194,7 +194,7 @@ def prepare_model_template(args: InferArguments,
                              f'args.max_model_len: {args.max_model_len}, model.max_model_len: {model.max_model_len}')
     # Preparing LoRA
     if is_adapter(args.sft_type) and args.ckpt_dir is not None:
-        if isinstance(args, DeployArguments):
+        if isinstance(args, DeployArguments) and args.lora_request_list is not None:
             for lora_request in args.lora_request_list:
                 model = Swift.from_pretrained(
                     model, lora_request.lora_local_path, lora_request.lora_name, inference_mode=True)

swift/llm/utils/argument.py

@@ -1133,7 +1133,7 @@ def handle_infer_backend(self):
                 or self.infer_backend == 'pt' and isinstance(self, DeployArguments) and self.sft_type == 'lora'):
             assert self.ckpt_dir is not None
             self.lora_modules.append(f'default-lora={self.ckpt_dir}')
-            self.lora_request_list = _parse_lora_modules(self.lora_modules, True)
+            self.lora_request_list = _parse_lora_modules(self.lora_modules, self.infer_backend == 'vllm')
             logger.info(f'args.lora_request_list: {self.lora_request_list}')
 
         template_info = TEMPLATE_MAPPING[self.template_type]