Error when using qwen-omni-turbo

[AlanZsm]

I config the host_agent and app_agent with API_TYPE: "qwen", API_MODEL: "qwen-omni-turbo". it always return below error “This model does not support non-streaming output” for any task:

[vyokky]

Thanks for the notice. UFO is now optimized for GPT/Gemini models and other model API may out of date. We will update their API in the next release.

[nice-mee]

Alibaba has added OpenAI-compatible API endpoint for Qwen models: https://dashscope.aliyuncs.com/compatible-mode/v1. And this is even listed as the recommended method in their API platform documentation (https://bailian.console.aliyun.com/?tab=doc#/doc). I can't even find usage example for dashscope (The original Python API developed by Alibaba). I suppose Alibaba has dropped support for their dashscope API.

You simply need to set API_TYPE to openai and API_BASE to https://dashscope.aliyuncs.com/compatible-mode/v1.

We will update our Qwen-related code to use OpenAI service later. For now you can use the workaround above to resolve this issue.

[nice-mee]

I made an initial implementation on my fork https://github.com/nice-mee/UFO/tree/vyokky/dev. This branch will be merged to main soon.

I have conducted some tests and it seems to be working fine. However qwen-omni-turbo seems to be pretty bad at understanding PC screenshots, I tried the newly released qwen3-235b-a22b with VISUAL_MODE set to False (Qwen3 is not multi-modal) and the results seem to be much better.

Usage:

API_TYPE: "qwen" ,
API_KEY: "sk-xxx",
API_MODEL: "qwen3-235b-a22b",