Paper

Beyond Passive Imitation: On-Policy Mid-Training Distillation for Language Models · ByteTrail