# Analyzing Redundancy in Pretrained Transformer Models

arXiv: https://arxiv.org/abs/2004.04010 · HTML: https://mlsanity.com/paper/2004.04010/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 8 Apr 2020.

## Built on by (most cited first, 6 of 6)

- [Math Neurosurgery: Isolating Language Models' Math Reasoning Abilities Using Only Forward Passes](https://mlsanity.com/paper/2410.16930/) (2024)
- [Silence is Sweeter Than Speech: Self-Supervised Model Using Silence to Store Speaker Information](https://mlsanity.com/paper/2205.03759/) (2022)
- [Interpreting the Effects of Quantization on LLMs](https://mlsanity.com/paper/2508.16785/) (2025)
- [ElastiFormer: Learned Redundancy Reduction in Transformer via Self-Distillation](https://mlsanity.com/paper/2411.15281/) (2024)
- [KDFP: A first-principles approach to knowledge distillation in large language models](https://mlsanity.com/paper/2610.10854/) (2026)
- [HAND: A Biologically-Inspired Activation Function that Improves Generalisation and Sample Efficiency in Image Classification](https://mlsanity.com/paper/2610.11534/) (2026)

## Abstract (arXiv)

Transformer-based deep NLP models are trained using hundreds of millions of parameters, limiting their applicability in computationally constrained environments. In this paper, we study the cause of these limitations by defining a notion of Redundancy, which we categorize into two classes: General Redundancy and Task-specific Redundancy. We dissect two popular pretrained models, BERT and XLNet, studying how much redundancy they exhibit at a representation-level and at a more fine-grained neuron-level. Our analysis reveals interesting insights, such as: i) 85% of the neurons across the network are redundant and ii) at least 92% of them can be removed when optimizing towards a downstream task. Based on our analysis, we present an efficient feature-based transfer learning procedure, which maintains 97% performance while using at-most 10% of the original neurons.