# Revisiting Token Dropping Strategy in Efficient BERT Pretraining

arXiv: https://arxiv.org/abs/2305.15273 · HTML: https://mlsanity.com/paper/2305.15273/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- Placed in Early exit and edge inference, one of 12 directions in Efficient LLM inference and serving.
- Builds on 39 of its 62 references that are in the map.
- First posted on arXiv 24 May 2023.

## Placed in

- Efficient LLM inference and serving / [Early exit and edge inference](https://mlsanity.com/inference/early-exit-and-edge-inference/)

## Builds on (most cited first, 30 of 39)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://mlsanity.com/paper/1907.11692/) (2019)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- [GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding](https://mlsanity.com/paper/1804.07461/) (2018)
- [SQuAD: 100,000+ Questions for Machine Comprehension of Text](https://mlsanity.com/paper/1606.05250/) (2016)
- [A Broad-Coverage Challenge Corpus for Sentence Understanding through Inference](https://mlsanity.com/paper/1704.05426/) (2017)
- [BoolQ: Exploring the Surprising Difficulty of Natural Yes/No Questions](https://mlsanity.com/paper/1905.10044/) (2019)
- [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://mlsanity.com/paper/2006.03654/) (2020)
- [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://mlsanity.com/paper/1909.08053/) (2019)
- [SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems](https://mlsanity.com/paper/1905.00537/) (2019)
- [Know What You Don't Know: Unanswerable Questions for SQuAD](https://mlsanity.com/paper/1806.03822/) (2018)
- [TinyBERT: Distilling BERT for Natural Language Understanding](https://mlsanity.com/paper/1909.10351/) (2019)
- [SemEval-2017 Task 1: Semantic Textual Similarity - Multilingual and Cross-lingual Focused Evaluation](https://mlsanity.com/paper/1708.00055/) (2017)
- [Neural Network Acceptability Judgments](https://mlsanity.com/paper/1805.12471/) (2018)
- [SpanBERT: Improving Pre-training by Representing and Predicting Spans](https://mlsanity.com/paper/1907.10529/) (2019)
- [Large Batch Optimization for Deep Learning: Training BERT in 76 minutes](https://mlsanity.com/paper/1904.00962/) (2019)
- [Universal Transformers](https://mlsanity.com/paper/1807.03819/) (2018)
- [Green AI](https://mlsanity.com/paper/1907.10597/) (2019)
- [What you can cram into a single vector: Probing sentence embeddings for linguistic properties](https://mlsanity.com/paper/1805.01070/) (2018)
- [Reducing Transformer Depth on Demand with Structured Dropout](https://mlsanity.com/paper/1909.11556/) (2019)
- [Lookahead Optimizer: k steps forward, 1 step back](https://mlsanity.com/paper/1907.08610/) (2019)
- [Knowledge Distillation Meets Self-Supervision](https://mlsanity.com/paper/2006.07114/) (2020)
- [Accelerating Training of Transformer-Based Language Models with Progressive Layer Dropping](https://mlsanity.com/paper/2010.13369/) (2020)
- [Self-Distillation as Instance-Specific Label Smoothing](https://mlsanity.com/paper/2006.05065/) (2020)
- [Understanding and Improving Lexical Choice in Non-Autoregressive Translation](https://mlsanity.com/paper/2012.14583/) (2020)
- [Token Dropping for Efficient BERT Pretraining](https://mlsanity.com/paper/2203.13240/) (2022)
- [Improving Sharpness-Aware Minimization with Fisher Mask for Better Generalization on Language Models](https://mlsanity.com/paper/2210.05497/) (2022)
- [PANDA: Prompt Transfer Meets Knowledge Distillation for Efficient Model Adaptation](https://mlsanity.com/paper/2208.10160/) (2022)

## Built on by (most cited first, 11 of 11)

- [Rho-1: Not All Tokens Are What You Need](https://mlsanity.com/paper/2404.07965/) (2024)
- [Accelerating Transformers with Spectrum-Preserving Token Merging](https://mlsanity.com/paper/2405.16148/) (2024)
- [Diversifying the Mixture-of-Experts Representation for Language Models with Orthogonal Optimizer](https://mlsanity.com/paper/2310.09762/) (2023)
- [SelecTKD: Selective Token-Weighted Knowledge Distillation for LLMs](https://mlsanity.com/paper/2510.24021/) (2025)
- [MerA: Merging Pretrained Adapters For Few-Shot Learning](https://mlsanity.com/paper/2308.15982/) (2023)
- [Unfair TOS: An Automated Approach using Customized BERT](https://mlsanity.com/paper/2401.11207/) (2024)
- [From Macro to Micro: Probing Dataset Diversity in Language Model Fine-Tuning](https://mlsanity.com/paper/2505.24768/) (2025)
- [ExLM: Rethinking the Impact of [MASK] Tokens in Masked Language Models](https://mlsanity.com/paper/2501.13397/) (2025)
- [StructSAM: Structure- and Spectrum-Preserving Token Merging for Segment Anything Models](https://mlsanity.com/paper/2603.07307/) (2026)
- [Unlocking Full Efficiency of Token Filtering in Large Language Model Training](https://mlsanity.com/paper/2502.00340/) (2025)
- [Nonparametric Teaching of Attention Learners](https://mlsanity.com/paper/2602.20461/) (2026)

## Abstract (arXiv)

Token dropping is a recently-proposed strategy to speed up the pretraining of masked language models, such as BERT, by skipping the computation of a subset of the input tokens at several middle layers. It can effectively reduce the training time without degrading much performance on downstream tasks. However, we empirically find that token dropping is prone to a semantic loss problem and falls short in handling semantic-intense tasks. Motivated by this, we propose a simple yet effective semantic-consistent learning method (ScTD) to improve the token dropping. ScTD aims to encourage the model to learn how to preserve the semantic information in the representation space. Extensive experiments on 12 tasks show that, with the help of our ScTD, token dropping can achieve consistent and significant performance gains across all task types and model sizes. More encouragingly, ScTD saves up to 57% of pretraining time and brings up to +1.56% average improvement over the vanilla token dropping.