从分布视角看SFT、RL和OPD
SFT、RL、OPD 的分布形变示意图
SFT, RL, and On-Policy Distillation Through a Distributional Lens
从分布式视角看 SFT、RL 以及基于策略的蒸馏技术
I have been thinking about post-training methods in terms of distributions. A language model is a distribution over sequen