FreeAct: Freeing Activations for LLM Quantization

TL;DR

FreeAct introduces a dynamic activation transformation framework for LLM quantization, effectively handling diverse token distributions and outperforming static methods in various large language models.

Contribution

It relaxes the static transformation constraint in quantization, enabling token-specific activation adjustments and improving performance in diffusion and multimodal LLMs.

Findings

Up to 5.3% performance improvement over baselines

Effective handling of diverse token distributions

Significant gains in diffusion and multimodal LLMs

Abstract

Quantization is pivotal for mitigating the significant memory and computational overhead of Large Language Models (LLMs). While emerging transformation-based methods have successfully enhanced quantization by projecting feature spaces onto smoother manifolds using orthogonal matrices, they typically enforce a rigid one-to-one transformation constraint. This static approach fails to account for the dynamic patterns inherent in input activations, particularly within diffusion LLMs (dLLMs) and Multimodal LLMs (MLLMs), where varying token types exhibit distinct distributions. To advance this, we propose FreeAct, a novel quantization framework that relaxes the static one-to-one constraint to accommodate dynamic activation disparities. Theoretically, we leverage the rank-deficient nature of activations to derive a solution space that extends beyond simple inverse matrices, enabling the decoupling of activation transformations from weights. Methodologically, FreeAct identifies token-specific dynamics (i.e., vision v.s. text, or masked tokens) and allocates distinct transformation matrices to the activation side, while maintaining a unified, static transformation for the weights. Extensive experiments across dLLMs and MLLMs demonstrate that FreeAct significantly outperforms baselines, up to 5.3% performance improvement, with in-depth analyses. Our code will be publicly released.