Submitted by Bingxiang He 37 Rethinking On-Policy Distillation of Large Language Models II: One Training Example Thinking Space 1