dify - 💡(How to fix) Fix 422 error when connection embedding model served via Litellm by Openai compatible provider [1 comments, 2 participants]

3.422 error occurs Litellm return 422 error due to Openai compatible provider set encoding_format to None explicitly, which Litellm only accept float or base64 as value. Here is the complete error message:

Error Code: 422 Message: litellm.BadRequestError: OpenAIException - Error code: 422 - {'detail': [{'type': 'literal_error', 'loc': ['body', 'encoding_format'], 'msg': "Input should be 'float' or 'base64'", 'input': None, 'ctx': {'expected': "'float' or 'base64'"}}]}. Received Model Group=Qwen3-Embedding-8B-4bit-DWQ-mlx Available Model Group Fallbacks=None

TL;DR

Set the encoding_format to either 'float' or 'base64' when using the OpenAI compatible provider to connect the embedding model.

Guidance

• Verify that the OpenAI compatible provider version 0.0.47 is correctly configured to set the encoding_format to a valid value.
• Check the Litellm documentation to ensure that the encoding_format parameter is correctly formatted as either 'float' or 'base64'.
• Update the OpenAI compatible provider to set the encoding_format to a valid value, such as 'float' or 'base64', to resolve the 422 error.
• Test the embedding model connection again after updating the encoding_format to ensure that the model is added to the list successfully.

Example

No code snippet is provided as the issue does not include specific code details.

Notes

The issue is specific to the OpenAI compatible provider version 0.0.47 and Litellm, so the solution may not apply to other versions or configurations.

Recommendation

Apply workaround: Set the encoding_format to either 'float' or 'base64' when using the OpenAI compatible provider to connect the embedding model, as this is the most straightforward solution to resolve the 422 error.