WINA: Weight Informed Neuron Activation for Accelerating Large Language Model Inference

TL;DR

WINA introduces a training-free sparse activation method for large language models that combines hidden state magnitudes and weight norms, achieving tighter approximation bounds and outperforming existing methods in inference efficiency.

Contribution

WINA presents a novel, simple, training-free sparse activation framework that improves approximation accuracy and inference performance for large language models by leveraging weight and activation information.

Findings

WINA outperforms state-of-the-art methods like TEAL by up to 2.94% in average performance.

WINA achieves tighter theoretical approximation error bounds than existing techniques.

Empirical results demonstrate WINA's effectiveness across diverse LLM architectures and datasets.

Abstract

The growing computational demands of large language models (LLMs) make efficient inference and activation strategies increasingly critical. While recent approaches, such as Mixture-of-Experts (MoE), leverage selective activation but require specialized training, training-free sparse activation methods offer broader applicability and superior resource efficiency through their plug-and-play design. However, many existing methods rely solely on hidden state magnitudes to determine activation, resulting in high approximation errors and suboptimal inference accuracy. To address these limitations, we propose WINA (Weight Informed Neuron Activation), a novel, simple, and training-free sparse activation framework that jointly considers hidden state magnitudes and the column-wise $\ell_2$-norms of weight matrices. We show that this leads to a sparsification strategy that obtains optimal approximation error bounds with theoretical guarantees tighter than existing techniques. Empirically, WINA also outperforms state-of-the-art methods (e.g., TEAL) by up to $2.94\%$ in average performance at the same sparsity levels, across a diverse set of LLM architectures and datasets. These results position WINA as a new performance frontier for training-free sparse activation in LLM inference, advancing training-free sparse activation methods and setting a robust baseline for efficient inference. The source code is available at https://github.com/microsoft/wina.