1bit-Merging: Dynamic Quantized Merging for Large Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Liu, Shuqi, Yao, Yuxuan, He, Bowei, Liu, Zehua, Han, Xiongwei, Yuan, Mingxuan, Wu, Han, Song, Linqi
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910963783958528
author Liu, Shuqi
Yao, Yuxuan
He, Bowei
Liu, Zehua
Han, Xiongwei
Yuan, Mingxuan
Wu, Han
Song, Linqi
author_facet Liu, Shuqi
Yao, Yuxuan
He, Bowei
Liu, Zehua
Han, Xiongwei
Yuan, Mingxuan
Wu, Han
Song, Linqi
contents Recent advances in large language models have led to specialized models excelling in specific domains, creating a need for efficient model merging techniques. While traditional merging approaches combine parameters into a single static model, they often compromise task-specific performance. However, task-specific routing methods maintain accuracy but introduce substantial storage overhead. We present \texttt{1bit}-Merging, a novel framework that integrates task-specific routing with 1-bit quantized task vectors to balance performance and storage efficiency. Our approach leverages the observation that different task-specific models store knowledge in distinct layers-chat models primarily in attention layers and math/code models in MLP layers, enabling targeted compression strategies. Through extensive experiments with LLaMA2 and Mistral model families across chat, mathematical reasoning, and code generation tasks, we demonstrate that 1bit-Merging achieves comparable or superior performance to existing methods while significantly reducing storage requirements. Our framework offers a practical solution for combining specialized models while maintaining their individual strengths and addressing the storage challenges of current approaches.
format Preprint
id arxiv_https___arxiv_org_abs_2502_10743
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle 1bit-Merging: Dynamic Quantized Merging for Large Language Models
Liu, Shuqi
Yao, Yuxuan
He, Bowei
Liu, Zehua
Han, Xiongwei
Yuan, Mingxuan
Wu, Han
Song, Linqi
Computation and Language
Recent advances in large language models have led to specialized models excelling in specific domains, creating a need for efficient model merging techniques. While traditional merging approaches combine parameters into a single static model, they often compromise task-specific performance. However, task-specific routing methods maintain accuracy but introduce substantial storage overhead. We present \texttt{1bit}-Merging, a novel framework that integrates task-specific routing with 1-bit quantized task vectors to balance performance and storage efficiency. Our approach leverages the observation that different task-specific models store knowledge in distinct layers-chat models primarily in attention layers and math/code models in MLP layers, enabling targeted compression strategies. Through extensive experiments with LLaMA2 and Mistral model families across chat, mathematical reasoning, and code generation tasks, we demonstrate that 1bit-Merging achieves comparable or superior performance to existing methods while significantly reducing storage requirements. Our framework offers a practical solution for combining specialized models while maintaining their individual strengths and addressing the storage challenges of current approaches.
title 1bit-Merging: Dynamic Quantized Merging for Large Language Models
topic Computation and Language
url https://arxiv.org/abs/2502.10743