1 paper · 1 filter
Chengwei Wu, Li Du, Hanyu Zhao +4
Scaling the amount of data used for supervied fine-tuning(SFT) does not guarantee the proportional gains in model performance, highlighting a critical need to understand what makes…