takatost

Feel free to contribute! 😁

Could you please provide some detailed screenshots to help us better understand the issue?

Milvus 2.4 brought some breaking changes, please refer to #3795.

What's the architecture and version of your server?

Can you try switching the image version from 0.6.4 to the previous one `langgenius/dify-api:3aa182e26a134d11e0958141a5622df6e087124c` and see if it works fine? And the version you may think has problem: `langgenius/dify-api:d5448e07abd840580bb8ffa69d886cb559575fe6`?

This issue is caused by upgrading `gmpy2` from `2.1.5` to `2.2.0a1`. It seems that some CPU instructions are not compatible with `2.2.0a1`, leading to abnormal exits. You can try replacing...

Version 0.6.5 has been released!

There is a lint issue that did not pass, please fix it.

If we need to add unit testing, please mock the vdb client interface. If we really need integration testing, I don't think the test time and resource consumption are suitable...

https://github.com/langgenius/dify/blob/9525ccac4f429cea9e3e263cb32ae7cee137a7e0/api/core/model_runtime/model_providers/xinference/llm/llm.py#L119 https://github.com/langgenius/dify/blob/9525ccac4f429cea9e3e263cb32ae7cee137a7e0/api/core/model_runtime/model_providers/xinference/llm/llm.py#L311 The model's max tokens and context size depend on the `context_length` returned by xinference when adding the model in Dify. You can check if xinference server returns correctly.