Why does the fine-tuned model not return my customized data as prompted?

[victory_1128]

Why does the fine-tuned model not return my customized data as prompted (for example: ABC system admin info, name: jack, mail: xxx@abc.com, tel: 123456)?

My homework steps:

1. Prepare jsonl file 2. Upload to openai 3. Generate job 4. View the new model after fine-tuning
2. Call the fine-tuned model
   The results did not return the data I defined.

[PaulBellow]

Fine-tuning isn’t meant for data retrieval…

See:

How to fine-tune gpt3.5-turbo to give it new knowledge?

Welcome to the community, @brandojazz! I agree with @elmstedt. Fine-tuning is recommended to save on prompt tokens for a high volume of calls for specific use case(s) or to set model behavior. For knowledge augmentation and retrieval, embeddings is the go-to approach. Here’s the documentation on embeddings. Currently, GPT models have a finite context length, which limits the number of tokens (prompt + completion) they can handle. To overcome this, you can pass an outline of your repository…

Look into RAG for data retrieval…

[victory_1128]

OK.Maybe I don’t understand this sentence accurately(Ability to train on more examples than can fit in a prompt ). So fine-tuning does not support training with custom data? Is it just an optimization of the model’s existing data?

[victory_1128]

Do you mean to use embeddings?
But I don’t quite understand what is RAG,can you give me more details?

[PaulBellow]

Yes, embeddings, sorry.

RAG = Retrieval Augmented Generation… If you search the forums, you’ll find a lot of useful posts and tools.

[victory_1128]

Thank you very much. I also want to confirm that embeddings can realize the training and use of own data, which is the RAG you mentioned, right?
Yes, I will search the forum later.