1 paper
Chaoyi Ruan, Geng Luo, Xinyi Wan +12
LLM reinforcement learning (RL) requires frequent synchronization of large model parameters between the trainer and distributed rollout actors. High-throughput RL post-training the…