Forget the Data and Fine-Tuning! Just Fold the Network to Compress

TL;DR

This paper presents model folding, a data-free compression method that merges similar neurons across layers to significantly reduce model size without fine-tuning, maintaining performance on large models.

Contribution

The paper introduces a novel data-free model compression technique called model folding that merges similar neurons across layers, preserving data statistics and avoiding fine-tuning.

Findings

Achieves comparable performance to data-driven methods.

Outperforms recent data-free compression techniques.

Effective for large-scale models at high sparsity.

Abstract

We introduce model folding, a novel data-free model compression technique that merges structurally similar neurons across layers, significantly reducing the model size without the need for fine-tuning or access to training data. Unlike existing methods, model folding preserves data statistics during compression by leveraging k-means clustering, and using novel data-free techniques to prevent variance collapse or explosion. Our theoretical framework and experiments across standard benchmarks, including ResNet18 and LLaMA-7B, demonstrate that model folding achieves comparable performance to data-driven compression techniques and outperforms recently proposed data-free methods, especially at high sparsity levels. This approach is particularly effective for compressing large-scale models, making it suitable for deployment in resource-constrained environments.