/rlhf-ppo

Notes and commented code for RLHF (PPO)

Primary LanguagePython

Reinforcement Learning from Human Feedback

Notes and commented code for RLHF (PPO)

The code has been commented by using the trl library from Hugging Face with version 0.7.10: https://github.com/huggingface/trl/

You will find the original code of the ppo_trainer.py file and also the commented code. You can use any diff tool to check my comments.

The video is here: https://www.youtube.com/watch?v=qGyFrqc34yc