High alpha trick for fine-tuning: Make your system prompts in your dataset really great. It'll help the model learn to do your task much faster, with less data. If you have lots of data, you can ignore this, but at small dataset sizes, this changes everything.

→ View original post on X — @mattshumer_