Zhoujian Sun

icon company Ant Group icon location Hangzhou,China icon location I want to dedicate my career to building an AI Doctor

Results 1 issues of Zhoujian Sun

Inconsistent Training and Inference Tokenization Result in Multi-Turn Rollout Training

3 comment

I am currently using verl for multi-turn interaction RL training and have identified two potential issues. - There might be a problem with the usage of _req.add_assistant_message in the script...