feat: support vision models from xinference

[Minamiyama]

Description

Please include a summary of the change and which issue is fixed. Please also include relevant motivation and context. List any dependencies that are required for this change.

Fixes # (issue)

Type of Change

Please delete options that are not relevant.

• New feature (non-breaking change which adds functionality)

How Has This Been Tested?

• launch a vision model on xinference like qwen-vl-chat
• add a model as a provider
• in workflow mode, select the vision model in LLM node, an eye vison mark will be shown

[Yeuoly]

LGTM

[sunhaha123]

Great job! How to triggle it? Git pull haved.

[Yeuoly]

Great job! How to triggle it? Git pull haved.

just setup qwen-chat-vl again in Dify, you will reach it :).

[sunhaha123]

@Minamiyama xorbitsai/inference#1425 Should I update xinference first? Only update dify, I still can't see the "eye" in "qwen-chat-vl". If using openai api comaptiblie, it showing that not support stream.

Updating xinference from source code seem not smoothly because of some new changes.

[Minamiyama]

@Minamiyama xorbitsai/inference#1425 Should I update xinference first? Only update dify, I still can't see the "eye" in "qwen-chat-vl". If using openai api comaptiblie, it showing that not support stream.

Updating xinference from source code seem not smoothly because of some new changes.

remove the old qwen-chat-vl from xinference provider, and add it as a totally new one, you should see the eye mark. And try to create a new workflow, just in case

[sunhaha123]

I have find out, the docker image is out of date.