Scale-Distribution Decoupling: Enabling Stable and Effective Training of Large Language Models

Open in new window