Mask More and Mask Later: Efficient Pre-training of Masked Language Models by Disentangling the [MASK] Token
This addresses efficiency for NLP researchers and practitioners by reducing computational resources and carbon footprint, though it is incremental as it builds on existing MLM methods.
The paper tackles the high computational cost of pre-training masked language models by proposing to append [MASK] tokens at later layers and increasing the masking rate, achieving pre-training of RoBERTa-base and RoBERTa-large with only 78% and 68% of the original budget without performance loss on GLUE.
The pre-training of masked language models (MLMs) consumes massive computation to achieve good results on downstream NLP tasks, resulting in a large carbon footprint. In the vanilla MLM, the virtual tokens, [MASK]s, act as placeholders and gather the contextualized information from unmasked tokens to restore the corrupted information. It raises the question of whether we can append [MASK]s at a later layer, to reduce the sequence length for earlier layers and make the pre-training more efficient. We show: (1) [MASK]s can indeed be appended at a later layer, being disentangled from the word embedding; (2) The gathering of contextualized information from unmasked tokens can be conducted with a few layers. By further increasing the masking rate from 15% to 50%, we can pre-train RoBERTa-base and RoBERTa-large from scratch with only 78% and 68% of the original computational budget without any degradation on the GLUE benchmark. When pre-training with the original budget, our method outperforms RoBERTa for 6 out of 8 GLUE tasks, on average by 0.4%.