1 paper
Jiazheng Li, Yawei Li, Junru Lu +7
On-policy self-distillation (OPSD) trains a large language model using supervision from a teacher instantiated from the same model as the student. The teacher is conditioned on pri…