LoTA-QAF: Lossless Ternary Adaptation for Quantization-Aware Fine-Tuning

Junyu Chen, Junzhuo Li, Zhen Peng, Wenjie Wang, Yuxiang Ren, Long Shi, Xuming Hu

Southwest University of Finance and Economics· Artificial Intelligence and Digital Finance Key Laboratory of Sichuan Province· Hong Kong University of Science and Technology (Guangzhou)· Sun Yat-sen University· Nanjing University

Quantization Fine-Tuning PEFT Efficient LLMs

⋅ NeurIPS ⋅ Project Page ⋅Slides ⋅Poster ⋅OpenReview

Abstract

Quantization and fine-tuning are crucial for deploying large language models (LLMs) on resource-constrained edge devices. However, fine-tuning quantized models presents significant challenges, primarily stemming from:

First, the mismatch in data types between the low-precision quantized weights (e.g., 4-bit) and the high-precision adaptation weights (e.g., 16-bit). This mismatch limits the computational efficiency advantage offered by quantized weights during inference.
Second, potential accuracy degradation when merging these high-precision adaptation weights into the low-precision quantized weights, as the adaptation weights often necessitate approximation or truncation.
Third, as far as we know, no existing methods support the lossless merging of adaptation while adjusting all quantized weights.

To address these challenges, we introduce lossless ternary adaptation for quantization-aware fine-tuning (LoTA-QAF). This is a novel fine-tuning method specifically designed for quantized LLMs, enabling the lossless merging of ternary adaptation weights into quantized weights and the adjustment of all quantized weights. LoTA-QAF operates through a combination of:

A custom-designed ternary adaptation (TA) that aligns ternary weights with the quantization grid and uses these ternary weights to adjust quantized weights.
A TA-based mechanism that enables the lossless merging of adaptation weights.
Ternary signed gradient descent (t-SignSGD) for updating the TA weights.

We apply LoTA-QAF to Llama-3.1/3.3 and Qwen-2.5 model families and validate its effectiveness on several downstream tasks. On the MMLU benchmark, our method effectively recovers performance for quantized models, surpassing 16-bit LoRA by up to 5.14%. For task-specific fine-tuning, 16-bit LoRA achieves superior results, but LoTA-QAF still outperforms other methods. Code is available in github.com/KingdalfGoodman/LoTA-QAF.