Fine-tuning
gonzalonunez opened this issue · comments
If the GPT-4 32k context window still leaves us wanting with a few-shot prompt, consider fine-tuning a base model with examples of well-documented Swift code.
I attempted this. Findings can be found here. Verdict is basically that this is not worth it / would take a lot more money to get somewhere close to what we're seeing with the out-of-the-box models.