Is There an Implementation for Offline SRPO?

[ye450450]

Hi, thanks for the great work.

I would like to ask whether there is an available implementation or example for offline SRPO training. The current training scripts appear to support only online RL with rollouts in the LIBERO environment.

Is there any code, configuration, or recommended setup for running SRPO purely in an offline setting without environment interaction?

Thanks!

[SII-limingliu]

Hi @ye450450, thanks for your interest and the kind words!

You are correct that the current codebase mainly focuses on the online RL setting. The implementation for offline SRPO has not been open-sourced yet.

We plan to release the offline training code and configurations in a future update. Please stay tuned!