Scalable Data Ablation Approximations for Language Models through Modular Training and Merging Paper • 2410.15661 • Published Oct 21, 2024