Question about SFT

[yantijin]

Exciting to read this work. I'm curious about why SFT for all parameters is chosen in the paper instead of existing PEFT methods like LoRA series. I would like to know if you have tried and evaluated the effectiveness of these two fine-tuning schemes.