I have fine-tuned a model with my own text data. It is giving only 67% accuracy. I want to improve the performance of that model. How?
Can we fine-tune the same model again?
I believe you have to fine tune a new one each time. Add more data to the set and submit a new one. However, fine-tuning is becoming more of a last resort in 2023, as you can have better quality and lower price with most other options.
Also when you say “67% accuracy”, it sounds like you’re using fine-tuning as a way to provide it with a data set? This is probably not a good way to use fine tuning, as it doesn’t retain “memory” of the answer. It just makes the model more likely to respond in a certain manner (more verbose, more sarcastic, giving step-by-step instructions, etc). The fine tuning makes it hallucinate more towards a certain answer, so that might artificially increase your accuracy measure. Embeddings is usually a better solution for a knowledge base.
Yes, it is possible to continue fine tuning a model, read more in our docs: OpenAI Platform.
2 Likes