# Train Large, Then Compress: Rethinking Model Size for Efficient Training and Inference of Transformers

arXiv: https://arxiv.org/abs/2002.11794 · HTML: https://mlsanity.com/paper/2002.11794/

## Key facts

- Cited by 67 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 26 Feb 2020.

## Built on by (most cited first, 30 of 67)

- [A Simple and Effective Pruning Approach for Large Language Models](https://mlsanity.com/paper/2306.11695/) (2023)
- [A Survey on Deep Neural Network Pruning-Taxonomy, Comparison, Analysis, and Recommendations](https://mlsanity.com/paper/2308.06767/) (2023)
- [Block Pruning For Faster Transformers](https://mlsanity.com/paper/2109.04838/) (2021)
- [Compute Trends Across Three Eras of Machine Learning](https://mlsanity.com/paper/2202.05924/) (2022)
- [Scale Efficiently: Insights from Pre-training and Fine-tuning Transformers](https://mlsanity.com/paper/2109.10686/) (2021)
- [How to Train BERT with an Academic Budget](https://mlsanity.com/paper/2104.07705/) (2021)
- [The State of Sparse Training in Deep Reinforcement Learning](https://mlsanity.com/paper/2206.10369/) (2022)
- [The Low-Resource Double Bind: An Empirical Study of Pruning for Low-Resource Machine Translation](https://mlsanity.com/paper/2110.03036/) (2021)
- [Transformers in Time-series Analysis: A Tutorial](https://mlsanity.com/paper/2205.01138/) (2022)
- [On Data Scaling in Masked Image Modeling](https://mlsanity.com/paper/2206.04664/) (2022)
- [A Survey on Model Compression and Acceleration for Pretrained Language Models](https://mlsanity.com/paper/2202.07105/) (2022)
- [A Survey on Efficient Training of Transformers](https://mlsanity.com/paper/2302.01107/) (2023)
- [Compute-Efficient Deep Learning: Algorithmic Trends and Opportunities](https://mlsanity.com/paper/2210.06640/) (2022)
- [Early Weight Averaging meets High Learning Rates for LLM Pre-training](https://mlsanity.com/paper/2306.03241/) (2023)
- [Are Larger Pretrained Language Models Uniformly Better? Comparing Performance at the Instance Level](https://mlsanity.com/paper/2105.06020/) (2021)
- [Enable Deep Learning on Mobile Devices: Methods, Systems, and Applications](https://mlsanity.com/paper/2204.11786/) (2022)
- [Attention Mechanism in Neural Networks: Where it Comes and Where it Goes](https://mlsanity.com/paper/2204.13154/) (2022)
- [MATE-KD: Masked Adversarial TExt, a Companion to Knowledge Distillation](https://mlsanity.com/paper/2105.05912/) (2021)
- [Parameter-Efficient Sparsity for Large Language Models Fine-Tuning](https://mlsanity.com/paper/2205.11005/) (2022)
- [Scaling Laws for Deep Learning](https://mlsanity.com/paper/2108.07686/) (2021)
- [On Efficient Training of Large-Scale Deep Learning Models: A Literature Review](https://mlsanity.com/paper/2304.03589/) (2023)
- [LORD: Low Rank Decomposition Of Monolingual Code LLMs For One-Shot Compression](https://mlsanity.com/paper/2309.14021/) (2023)
- [Egeria: Efficient DNN Training with Knowledge-Guided Layer Freezing](https://mlsanity.com/paper/2201.06227/) (2022)
- [An Overview on Language Models: Recent Developments and Outlook](https://mlsanity.com/paper/2303.05759/) (2023)
- [On Anytime Learning at Macroscale](https://mlsanity.com/paper/2106.09563/) (2021)
- [Train Flat, Then Compress: Sharpness-Aware Minimization Learns More Compressible Models](https://mlsanity.com/paper/2205.12694/) (2022)
- [Effective Interplay between Sparsity and Quantization: From Theory to Practice](https://mlsanity.com/paper/2405.20935/) (2024)
- [Rethinking Perturbations in Encoder-Decoders for Fast Training](https://mlsanity.com/paper/2104.01853/) (2021)
- [Inspecting the concept knowledge graph encoded by modern language models](https://mlsanity.com/paper/2105.13471/) (2021)
- [PDP: Parameter-free Differentiable Pruning is All You Need](https://mlsanity.com/paper/2305.11203/) (2023)

## Abstract (arXiv)

Since hardware resources are limited, the objective of training deep learning models is typically to maximize accuracy subject to the time and memory constraints of training and inference. We study the impact of model size in this setting, focusing on Transformer models for NLP tasks that are limited by compute: self-supervised pretraining and high-resource machine translation. We first show that even though smaller Transformer models execute faster per iteration, wider and deeper models converge in significantly fewer steps. Moreover, this acceleration in convergence typically outpaces the additional computational overhead of using larger models. Therefore, the most compute-efficient training strategy is to counterintuitively train extremely large models but stop after a small number of iterations. This leads to an apparent trade-off between the training efficiency of large Transformer models and the inference efficiency of small Transformer models. However, we show that large models are more robust to compression techniques such as quantization and pruning than small models. Consequently, one can get the best of both worlds: heavily compressed, large models achieve higher accuracy than lightly compressed, small models.