Ask HN: How do you fine-tune LLMs?
Do you usually do full fine-tuning or LoRA (or its variants)? and why? I'm curious if the how ppl fine-tune LLMs depends on why ppl fine-tune LLMs. The couple of times I've tried to fine-tune a text model I've consistently gotten way worse results (I'm not good at it), so I've just stopped trying...