[short] Language models scale reliably with over-training and on downstream tasks

Arxiv Papers - [short] Language models scale reliably with over-training and on downstream tasks