1 paper
Ellen Xiaoqing Tan, Jack Lanchantin, Zichun Yu +11
Large language models are classically trained in stages: pretraining on raw text followed by post-training for instruction following and reasoning. However, this separation creates…