Hi! I’m trying to implement DeepSeek-R1 for a RAG-based chatbot, but I’m running into token permission issues. Even though my token has “write” access, I keep getting an access error.
I have a similar project in another folder using the same code, and it works fine. However, when I started a new project from scratch with mostly the same code, I began encountering token errors. The old project still works without issues, but the new one keeps failing.
I’ve already tried logging in via CLI and generating a new token, but the problem persists.
Additionally, I’d love some recommendations on lighter and more efficient models to use instead of DeepSeek-R1—both for inference and for the embedding process. Right now, I’m using HuggingFaceEmbedding, but I’m open to alternatives that might improve efficiency. Any suggestions?
1 Like
There aren’t many models that can be used with the Inference API that are as good as R1…
That said, the fact that a 403 error occurs when changing folders is strange. If they are both virtual environments, the cause could be a different version of the huggingface_hub library, but if they are both live environments, I don’t think the difference would be noticeable from the Hugging Face server…
In other words, it would be strange if they both caused an error or if they both worked.
I am quite clue less so I decided to change the model and go for llama… but since its gated ill need to wait for permission. In the mean time I’ll be using some models which don’t require permission. Could you recommend me some good ones. I am trying to make a Rag trained chatbot for a website like I mentions in the description.
1 Like
I think these model would be suitable for that purpose, for example.