TRAWL: Tensor Reduced and Approximated Weights for Large Language Models

TL;DR

TRAWL introduces tensor decomposition to prune large language models, effectively denoising weights and improving performance by up to 16% without extra training or data.

Contribution

It proposes a novel tensor decomposition method for large language model pruning that captures global structural patterns beyond previous matrix-based approaches.

Findings

Improves model performance by up to 16% on benchmarks.

Effectively denoises weights without additional data or training.

Outperforms baseline models with minimal performance impact.

Abstract

Recent research has shown that pruning large-scale language models for inference is an effective approach to improving model efficiency, significantly reducing model weights with minimal impact on performance. Interestingly, pruning can sometimes even enhance accuracy by removing noise that accumulates during training, particularly through matrix decompositions. However, recent work has primarily focused on single matrix decompositions or lower precision techniques, which may fail to fully capture structural patterns. To address these limitations, we introduce TRAWL (Tensor Reduced and Approximated Weights for Large Language Models), a technique that applies tensor decomposition across multiple weight matrices to effectively denoise LLMs by capturing global structural patterns. Our experiments show that TRAWL improves model performance by up to 16% over baseline models on benchmark datasets, without requiring additional data, training, or fine-tuning.