diff --git a/README.md b/README.md
index 8e0c6075a..0c936a270 100644
--- a/README.md
+++ b/README.md
@@ -247,7 +247,13 @@ python3 -m fastchat.serve.gradio_web_server_multi
 ```
 - The default model worker based on huggingface/transformers has great compatibility but can be slow. If you want high-throughput batched serving, you can try [vLLM integration](docs/vllm_integration.md).
 - If you want to host it on your own UI or third party UI, see [Third Party UI](docs/third_party_ui.md).
- 
+
+## Use models from modelscope
+For Chinese users, you can use models from www.modelscope.cn via specify the following environment variables.
+```bash
+export FASTCHAT_USE_MODELSCOPE=True
+```
+
 ## API
 ### OpenAI-Compatible RESTful APIs & SDK
 FastChat provides OpenAI-compatible APIs for its supported models, so you can use FastChat as a local drop-in replacement for OpenAI APIs.
diff --git a/fastchat/model/model_adapter.py b/fastchat/model/model_adapter.py
index 0d1a02dae..91881d214 100644
--- a/fastchat/model/model_adapter.py
+++ b/fastchat/model/model_adapter.py
@@ -337,7 +337,8 @@ def load_model(
         try:
             from modelscope.hub.snapshot_download import snapshot_download
 
-            model_path = snapshot_download(model_id=model_path, revision=revision)
+            if not os.path.exists(model_path):
+                model_path = snapshot_download(model_id=model_path, revision=revision)
         except ImportError as e:
             warnings.warn(
                 "Use model from www.modelscope.cn need pip install modelscope"