Saved in:
Bibliographic Details
Main Authors: Mirzaei, Amir Reza, Wen, Yuqiao, Cao, Yanshuai, Mou, Lili
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.26690
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915603766312960
author Mirzaei, Amir Reza
Wen, Yuqiao
Cao, Yanshuai
Mou, Lili
author_facet Mirzaei, Amir Reza
Wen, Yuqiao
Cao, Yanshuai
Mou, Lili
contents Low-Rank Adaptation (LoRA) has become a popular technique for parameter-efficient fine-tuning of large language models (LLMs). In many real-world scenarios, multiple adapters are loaded simultaneously to enable LLM customization for personalized user experiences or to support a diverse range of tasks. Although each adapter is lightweight in isolation, their aggregate cost becomes substantial at scale. To address this, we propose LoRAQuant, a mixed-precision post-training quantization method tailored to LoRA. Specifically, LoRAQuant reparameterizes each adapter by singular value decomposition (SVD) to concentrate the most important information into specific rows and columns. This makes it possible to quantize the important components to higher precision, while quantizing the rest to ultra-low bitwidth. We conduct comprehensive experiments with LLaMA 2-7B, LLaMA 2-13B, and Mistral 7B models on mathematical reasoning, coding, and summarization tasks. Results show that our LoRAQuant uses significantly lower bits than other quantization methods, but achieves comparable or even higher performance.
format Preprint
id arxiv_https___arxiv_org_abs_2510_26690
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LoRAQuant: Mixed-Precision Quantization of LoRA to Ultra-Low Bits
Mirzaei, Amir Reza
Wen, Yuqiao
Cao, Yanshuai
Mou, Lili
Machine Learning
Low-Rank Adaptation (LoRA) has become a popular technique for parameter-efficient fine-tuning of large language models (LLMs). In many real-world scenarios, multiple adapters are loaded simultaneously to enable LLM customization for personalized user experiences or to support a diverse range of tasks. Although each adapter is lightweight in isolation, their aggregate cost becomes substantial at scale. To address this, we propose LoRAQuant, a mixed-precision post-training quantization method tailored to LoRA. Specifically, LoRAQuant reparameterizes each adapter by singular value decomposition (SVD) to concentrate the most important information into specific rows and columns. This makes it possible to quantize the important components to higher precision, while quantizing the rest to ultra-low bitwidth. We conduct comprehensive experiments with LLaMA 2-7B, LLaMA 2-13B, and Mistral 7B models on mathematical reasoning, coding, and summarization tasks. Results show that our LoRAQuant uses significantly lower bits than other quantization methods, but achieves comparable or even higher performance.
title LoRAQuant: Mixed-Precision Quantization of LoRA to Ultra-Low Bits
topic Machine Learning
url https://arxiv.org/abs/2510.26690