Recent work suggests that switching optimizers between pretraining and subsequent training stages incurs a performance cost, but systematic investigation of …
机构:MIT
来源:arXiv 2610.22650 | AI4Papers 论文推荐平台