1 paper
Qiwei Di, Xuheng Li, Kaixuan Ji +3
On-policy distillation (OPD) learns from teacher feedback on student-generated responses and has shown promise in reducing forgetting relative to supervised fine-tuning (SFT). Howe…