How has the behavior cloning data been generated ?

[subhojitdas]

Hi ,

I am able to run the GRPO training from Train/verifier/script.py
According to the WebAgentR1 paper , the GRPO training is done on the data generated by behavor cloning.

Can you please refer me in the code , where and how the behavior cloning data has been generated ?

Thanks,
Subhojit

[huangtiansheng]

Hi, I have the same question here.

Is that possible that the authors can release the SFT checkpoint to huggingface and then we can directly apply GRPO on this model?

I check there is one model on huggingface but not sure whether it is a correct one or not

https://huggingface.co/yczhuang/qwen-3b-sft-webagent

Thanks,
Tiansheng