Fine Tuning Successful but Q/A testing 0% correct

Trained with 50 questions … the process went thru just fine. Followed everything correctly and once done started testing and it didn’t answer one question correctly


Did you read this?

Fine-tuning isn’t the best option when it comes to adding factual information that the model can use. It’s typically best used for fortifying a style or a schema.

If you want to give the model information to work with, a RAG solution with embeddings ( or assistants might be a better approach.

hope this helps :slight_smile:

1 Like