🐛 bug: DeepSeek R1 model integration and adaptation issues.

[akinlong]

Description

I am trying to use DeepSeek R1 as backend LLM service of MM AI copilot, it fails when calling the backend API, the server logs shows:
message: invalid character '<' looking for beginning of value

As we know, the DeepSeek R1 response starts with <think> part, some AI client tools will display the 'think' part as a single breakdown UI.
My question is, do we have plans to fix the reason model or fully support the reason model in the future?

Steps to reproduce

Config DeepSeek R1 model as backup back AI service and post a message to the copilot.

[akinlong]

My mistake, standard DS API doesn't return think part, it was designed by qwen.

After I switched to the latest version of VLLM and re-run the DeepSeek R1 model, the copilot is now able to properly accept the LLM's output. However, the reasoning part still experiences lag in the UI and does not display the reasoning process. Will this feature be implemented in the UI in the future?