Paper explores applying reinforcement learning during LLM pre-training

Arxiv CS.LG (machine learning)

The paper challenges the standard LLM training pipeline by applying RL during pre-training instead of only after SFT. The authors compare RL, SFT, and combined training from scratch.

llmreinforcement-learning

Arxiv CS.LG (machine learning)·Rachit BansalClara MohriTian QinDavid Alvarez-MelisSham Kakade

7 days ago