Dear Hugging Face Support Team,
I hope you are doing well.
We are integrating the Dify knowledge-base tool with the Hugging Face Inference API to generate embeddings (feature-extraction pipeline). We now consistently receive the following 404 error:
404 Client Error: Not Found for url: https://api-inference.huggingface.co/pipeline/feature-extraction/intfloat/multilingual-e5-large-instruct
(Request ID: Root=1-683b4db9-1e980b4c755dcffd2fc32730; bdc8fc2e-733e-4d9b-afee-955f5c940bdb)
Until about two weeks ago this model could be registered and used without any issues, but it suddenly started returning 404 errors.
Model page: intfloat/multilingual-e5-large-instruct · Hugging Face
Steps to reproduce
- In Dify, select Hugging Face Inference API as the embedding model.
- Enter an access token with only the
read
scope. - Specify the model name intfloat/multilingual-e5-large-instruct.
- Save the settings and run a test → the 404 error appears.
What we have already checked
- Although the token has only the
read
scope, another model (e.g.,sentence-transformers/all-MiniLM-L6-v2
) works fine through the same API. - The model page is accessible via the web interface.
Questions
- Has the feature-extraction endpoint for this model been disabled, or has the endpoint URL format changed?
- If the endpoint has changed, could you share the current correct URL format?
- Are there any other possible causes or misconfigurations we should investigate?
Your guidance would be greatly appreciated.
Thank you very much for your time and support.