Pre-training large language models is known to be extremely resource intensive and often times inefficient, under-utilizing the information encapsulated in the training text sequences. In this paper, we present SpacTor, a new training procedure consisting of (1) a hybrid objective combining span corruption (SC) and token replacement detection (RTD), and (2) a two-stage curriculum that optimizes the hybrid objective over the initial τ iterations, then transitions to standard SC loss. We show empirically that the effectiveness of the hybrid objective is tied to the two-stage pre-training schedule, and provide extensive analysis on why this is the case. In our experiments with encoder-decoder architectures (T5) on a variety of NLP tasks, SpacTor-T5 yields the same downstream performance as standard SC pre-training, while enabling a 50% reduction in pre-training iterations and 40% reduction in total FLOPs. Alternatively, given the same amount of computing budget, we find that SpacTor results in significantly improved downstream benchmark performance.
@article{arxiv.2401.13160,
title = {SpacTor-T5: Pre-training T5 Models with Span Corruption and Replaced Token Detection},
author = {Ke Ye and Heinrich Jiang and Afshin Rostamizadeh and Ayan Chakrabarti and Giulia DeSalvo and Jean-François Kagy and Lazaros Karydas and Gui Citovsky and Sanjiv Kumar},
journal= {arXiv preprint arXiv:2401.13160},
year = {2024}
}