Nathan Lambert 评 Tinker StableReinforce 逆向分析
Nice thread, some RL details used by closed labs slowly leaking out.
AI 导读
Martin Marek 发帖称，Tinker 的新 StableReinforce 目标函数是 Reinforce 加 Score Centering，并分享了自己的逆向分析过程。Nathan Lambert 转发评论称，这条帖子不错，封闭实验室使用的一些 RL 细节正在慢慢流出。
查看原帖Tinker 新的 StableReinforce 目标函数是 Reinforce + Score Centering。以下是我逆向分析它的过程 🧵
查看原文
Tinker's new StableReinforce objective is Reinforce + Score Centering. Here's how I reverse-engineered it 🧵
正文
Nice thread, some RL details used by closed labs slowly leaking out.