Introduction

Reinforcement Learning (RL) has become one of the core ingredients of modern Large Language Model (LLM) post-training. While pre-training teaches a model to predict the next token from massive text corpora, …