Saved in:
Bibliographic Details
Main Authors: Ye, Zijian, Huang, Wei, Yu, Yifei, Ren, Tianhe, Wang, Zhongrui, Qi, Xiaojuan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2507.01027
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914288253272064
author Ye, Zijian
Huang, Wei
Yu, Yifei
Ren, Tianhe
Wang, Zhongrui
Qi, Xiaojuan
author_facet Ye, Zijian
Huang, Wei
Yu, Yifei
Ren, Tianhe
Wang, Zhongrui
Qi, Xiaojuan
contents Large language models (LLMs) demonstrate remarkable performance but face substantial computational and memory challenges that limit their practical deployment. Quantization has emerged as a promising solution; however, its effectiveness is often limited by quantization errors arising from weight distributions that are not quantization-friendly and the presence of activation outliers. To address these challenges, we introduce DBellQuant, an innovative post-training quantization (PTQ) framework that achieves nearly 1-bit weight compression and 6-bit activation quantization with minimal performance degradation. DBellQuant uses Learnable Transformation for Dual-Bell (LTDB) algorithm, which transforms single-bell weight distributions into dual-bell forms to reduce binarization errors and applies inverse transformations to smooth activations. DBellQuant sets a new state-of-the-art by preserving superior model performance under aggressive weight and activation quantization. For example, on the Wikitext2 dataset, DBellQuant achieves a perplexity of 14.39 on LLaMA2-13B with 6-bit activation quantization, significantly outperforming BiLLM's 21.35 without activation quantization, underscoring its potential in compressing LLMs for real-world applications.
format Preprint
id arxiv_https___arxiv_org_abs_2507_01027
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DBellQuant: Breaking the Bell with Double-Bell Transformation for LLMs Post Training Binarization
Ye, Zijian
Huang, Wei
Yu, Yifei
Ren, Tianhe
Wang, Zhongrui
Qi, Xiaojuan
Machine Learning
Large language models (LLMs) demonstrate remarkable performance but face substantial computational and memory challenges that limit their practical deployment. Quantization has emerged as a promising solution; however, its effectiveness is often limited by quantization errors arising from weight distributions that are not quantization-friendly and the presence of activation outliers. To address these challenges, we introduce DBellQuant, an innovative post-training quantization (PTQ) framework that achieves nearly 1-bit weight compression and 6-bit activation quantization with minimal performance degradation. DBellQuant uses Learnable Transformation for Dual-Bell (LTDB) algorithm, which transforms single-bell weight distributions into dual-bell forms to reduce binarization errors and applies inverse transformations to smooth activations. DBellQuant sets a new state-of-the-art by preserving superior model performance under aggressive weight and activation quantization. For example, on the Wikitext2 dataset, DBellQuant achieves a perplexity of 14.39 on LLaMA2-13B with 6-bit activation quantization, significantly outperforming BiLLM's 21.35 without activation quantization, underscoring its potential in compressing LLMs for real-world applications.
title DBellQuant: Breaking the Bell with Double-Bell Transformation for LLMs Post Training Binarization
topic Machine Learning
url https://arxiv.org/abs/2507.01027