Invalid model uri when register model

[usegitforwhat]

System Info / 系統信息

windows 11 + docker

Running Xinference with Docker? / 是否使用 Docker 运行 Xinfernece？

• docker / docker
• pip install / 通过 pip install 安装
• installation from source / 从源码安装

Version info / 版本信息

1.4.1

The command used to start Xinference / 用以启动 xinference 的命令

docker run -e XINFERENCE_MODEL_SRC=modelscope -p 9998:9997 --gpus all xprobe/xinference:latest xinference-local -H 0.0.0.0 --log-level debug

Reproduction / 复现过程

The documentation states that model_uri is a string representing the model's source URI (e.g., 'file:///path/to/your_model'), and that Xinference will attempt to download from Hugging Face using the model ID if no URI is provided.

However, during new embedding model registration, the frontend requires model_uri to be filled. I'm unsure whether this is a bug or if I should input a specific URI to allow downloads from Hugging Face or Modelscope.

Expected behavior / 期待表现

Custom model can auto download from model hub.

[qinxuye]

If registered custom model, no download is required, the model uri is required.

[usegitforwhat]

If registered custom model, no download is required, the model uri is required.

Based on the documentation, if I leave the model URI field empty, it should automatically download from Hub. However, when I leave the "Model Path" box blank, I get the error: "Please fill in valid value for all fields." I assume the "Model Path" means "Model URI".

I then tried entering the path "/root/.xinference/cache/gte_Qwen2-1.5B-instruct", but received a server error: "400 - [address=0.0.0.0:49953, pid=55] Invalid model URI /root/.xinference/cache/gte_Qwen2-1.5B-instruct."

Interestingly, manually creating the directory "/root/.xinference/cache/gte_Qwen2-1.5B-instruct" resolves the path error, but I still can't launch or download the model. When attempting to launch, the logs show: "Unrecognized model in /root/.xinference/cache/gte_Qwen2-1.5B-instruct. Should have a model_type key in its config.json, or contain one of the following strings in its name:..."

It appears the system is attempting to load from local storage despite the documentation suggesting automatic downloads. How can I resolve this? Is manual downloading of the model files required for proper setup?

[qinxuye]

It's better to have a model weight already for custom models.

[usegitforwhat]

Got it. Thanks a lot.