millioniron

Results 2 comments of


                                            millioniron

为什么reinforce++和GRPO需要累积rewards?

token-level-loss，一个技巧可以看：https://zhuanlan.zhihu.com/p/1887453157099037961 或https://zhuanlan.zhihu.com/p/31770741283

Dataloader is blazingly fast for ACT training but VERY slow for SmolVLA and Diffusion Policy training

waiting