Dang'ai

New RL method slashes LLM pretraining time by 66%

Researchers have developed AC-ODM, a novel method that uses reinforcement learning to optimize the composition of pretraining data for large language models (LLMs). This approach significantly improves sample efficiency…