Fetching the paper…

Maybe Only 0.5% Data is Needed: A Preliminary Exploration of Low Training Data Instruction Tuning · Around