---
id: 20260827-T0-05
title: "AQLoRA：零搜索配方，让QLoRA微调速度提升"
title_en: "AQLoRA: Zero-Search Recipe for Faster Quantized LoRA Fine-Tuning"
url: https://ai.daily.yangsir.net/daily/20260827-T0-05
issue_date: 2026-08-27
publish_date: 2026-08-26T04:00:00.000Z
category: research
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2608.23816
---

# AQLoRA：零搜索配方，让QLoRA微调速度提升

量化微调（QLoRA）虽然节省内存，但速度较慢。新研究提出AQLoRA（自适应量化LoRA），通过一次CPU遍历数据，在保持量化内存优势的同时，减少了训练时间。这个零搜索的配方让量化微调的速度更接近fp16 LoRA。

## English Version

**AQLoRA: Zero-Search Recipe for Faster Quantized LoRA Fine-Tuning**

Quantized fine-tuning (QLoRA) saves memory but is slow. New research presents AQLoRA (Adaptive-Quantization LoRA), which uses a single CPU pass to reduce training time while retaining quantized memory benefits. This zero-search recipe makes quantized fine-tuning speed closer to fp16 LoRA.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2608.23816)

**详情页**：https://ai.daily.yangsir.net/daily/20260827-T0-05

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*