DOTResize: Reducing LLM Width via Discrete Optimal Transport-based Neuron Merging
This addresses the problem of high computational cost and inaccessibility of large pre-trained models for users in AI and machine learning, representing an incremental improvement over existing compression methods.
The paper tackles computational redundancy in large language models by proposing DOTResize, a method that reduces model width via neuron merging framed as a discrete optimal transport problem, achieving measurable reductions in computational cost while outperforming state-of-the-art pruning techniques across multiple LLM families and sizes.
Model compression offers a promising path to reducing the cost and inaccessibility of large pre-trained models, without significantly compromising their impressive performance. Large Transformer models, including large language models (LLMs), often contain computational redundancy, which can serve as a target for new model compression methods. In this work, we specifically target neuron-level redundancies in model layers by combining groups of similar neurons into fewer neurons. We frame this width reduction as a Discrete Optimal Transport problem, and propose DOTResize, a novel Transformer compression method that uses optimal transport theory to transform and compress model weights. To ensure applicability within the Transformer architecture, we motivate and incorporate entropic regularization and matrix factorization into the transportation maps produced by our method. Unlike pruning-based approaches which discard neurons based on importance measures, DOTResize re-projects the entire neuron width, allowing the retention and redistribution of useful signal across the reduced layer. Empirical results show that compared to simple or state-of-the-art neuron width-pruning techniques, DOTResize can outperform these methods across multiple LLM families and sizes, while achieving measurable reductions in real-world computational cost.