I have been tasked with fine tuning an LLM that will allow people at my work to ask questions on our websites back end CMS.
I have prepared a CSV file with “prompt” and “output”.
I have generated a set of instructions on two tasks in my training data to test this out, how to access the classic back end, and how to log in to the back end with email and password.
I fine tuned GPT Neo 1.3B for a few hours on 110 prompts each with one of two outputs, each prompt has either a set of instructions on how to log into the cms, or how to access the classic cms back end.
Now nonmatter what I type into it, it only gives me the set of instructions of how to access the classic back end, it does not give the instructions of how to log in to the system.
I need to fine tune this to understand hundreds of tasks and give specific instructions based on the tasks prompted to it.
This is my first time fine tuning my own LLM.
Please give me some help and suggestions, thank you.