Plan for multi-turn RL or agent_loop rollout in latest Verl？

[JianPeng20]

Thanks a lot for your awesome work on Verl!
Just wondering—are there any plans to add multi-turn RL training?

A couple of quick ideas:

1. Try multi-turn tasks and compare with non-quantized runs
2. Make agent_loop work with importance sampling in Verl

Would love to hear your thoughts. Thanks again!

[LiyuanLucasLiu]

this is definitely on the road map, but we dont have concrete timelines yet. as usual, community contributions are highly welcomed.