I have a great customized chatGPT that I created using a file with a lot of unique definitions and taxonomies. It does great in the prompt UI, but I can’t seem to access it via the API in order to call it within a script. This functionality only seems to be available for fine-tuned models. So I’ve been trying to build training data from the file in order to fine tune and get the same performance.
The results are terrible. I’ve been trying to generate training data using the original custom chatGPT, because it has access to the file. It’s still not getting great performance.
Are there any best practices around doing this? I would love just a python script that could take a file, parse it, then output the needed prompts to feed the fine tuned model to replicate the file-customized model.
Any advice would be much appreciated!