Update to the new verl version

[Shiguang-Guo]

Very impressive work! I noticed that the current verl version being used is v0.1, which lacks the corresponding official documentation (the documentation at https://verl.readthedocs.io/en/latest/index.html
starts at least from v0.2). Do you have plans to update to the latest verl version (perhaps 0.5)?

[telxt]

Yes! We do plan a full update of the codebase to support newer versions of verl and vllm. We’ll update it as soon as possible and make sure the new version reliably reproduces the results.

[wy20907104]

please support rl lora training style

[telxt]

Sorry for my late reply. RL lora training and our HPT algorithm are decoupled. It can be implemented by modifying the verl code. I will also update it after upgrading the code.