Clarify fine tuning services

[josephjclark]

I'm a bit lost on where we got to with fine tuning, but this issue still stands....

The openfn_llama service looks like it was set up to fine tune a llama model. Which is fine. But it also contains a gpt_finetune.py, so it's actually a general fine tuning service now.

Is it even a service? Is it not a one-shot script to trigger a round of fine tuning? There is a run script with a generate_code endpoint. But should not the inference service expose the model fine tuned by the openfn_llama command?

I don't know.

Here are some steps to consider:

• Add documention to the README of openfn_llama
• Rename openfn_llama to finetuning or training
• Decouple the web service from the openfn_llama service entirely
• Probably move openfn_llama out of /services
• Maybe move fine tuning out into a separate repo

I do really like the idea of moving this to another repo and enabling llama_ft to be a model type argument on the codegen services.

[josephjclark]

In the bun server refactor, I am for the moment keeping the fine tuning stuff up in the root, in a folder called fine_tuning.

Note that the old openfn_llama folder contains fine tuning code for various models - not just llama.