data composition

large language model (2) supervised fine-tuning (2) catastrophic forgetting (2) fixed-length sequence (1) multi-bucket data (1) model capacity (1) ability learning (1) language model pretraining (1) multilingual benchmark (1) sample-efficient training (1)