# Accelerated Large Batch Optimization of BERT Pretraining in 54 minutes

arXiv: https://arxiv.org/abs/2006.13484 · HTML: https://mlsanity.com/paper/2006.13484/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 24 Jun 2020.

## Built on by (most cited first, 7 of 7)

- [Communication-Compressed Adaptive Gradient Method for Distributed Nonconvex Optimization](https://mlsanity.com/paper/2111.00705/) (2021)
- [LLM Content Moderation and User Satisfaction: Evidence from Response Refusals in Chatbot Arena](https://mlsanity.com/paper/2501.03266/) (2025)
- [Compressed Communication for Distributed Training: Adaptive Methods and System](https://mlsanity.com/paper/2105.07829/) (2021)
- [DropCompute: simple and more robust distributed synchronous training via compute variance reduction](https://mlsanity.com/paper/2306.10598/) (2023)
- [Accelerating Large Batch Training via Gradient Signal to Noise Ratio (GSNR)](https://mlsanity.com/paper/2309.13681/) (2023)
- [DynaMaR: Dynamic Prompt with Mask Token Representation](https://mlsanity.com/paper/2206.02982/) (2022)
- [Transformer-based Named Entity Recognition in Construction Supply Chain Risk Management in Australia](https://mlsanity.com/paper/2311.13755/) (2023)

## Abstract (arXiv)

BERT has recently attracted a lot of attention in natural language understanding (NLU) and achieved state-of-the-art results in various NLU tasks. However, its success requires large deep neural networks and huge amount of data, which result in long training time and impede development progress. Using stochastic gradient methods with large mini-batch has been advocated as an efficient tool to reduce the training time. Along this line of research, LAMB is a prominent example that reduces the training time of BERT from 3 days to 76 minutes on a TPUv3 Pod. In this paper, we propose an accelerated gradient method called LANS to improve the efficiency of using large mini-batches for training. As the learning rate is theoretically upper bounded by the inverse of the Lipschitz constant of the function, one cannot always reduce the number of optimization iterations by selecting a larger learning rate. In order to use larger mini-batch size without accuracy loss, we develop a new learning rate scheduler that overcomes the difficulty of using large learning rate. Using the proposed LANS method and the learning rate scheme, we scaled up the mini-batch sizes to 96K and 33K in phases 1 and 2 of BERT pretraining, respectively. It takes 54 minutes on 192 AWS EC2 P3dn.24xlarge instances to achieve a target F1 score of 90.5 or higher on SQuAD v1.1, achieving the fastest BERT training time in the cloud.