A Comparative Study between Full-Parameter and LoRA-based Fine-Tuning on Chinese Instruction Data for Instruction Following Large Language Model
Xianghui Sun, Yunjie Ji, Baochang Ma, Xiangang Li
TL;DR
The paper benchmarks full-parameter fine-tuning against LoRA-based tuning for Chinese instruction data using LLaMA bases, examining how base model size, data scale, and training cost affect performance. It shows that larger base models and more data improve LoRA results, but full-parameter fine-tuning often achieves higher scores unless starting from an instruction-tuned base. LoRA offers substantial training-time savings, and its effectiveness increases with model size and data, though it may underperform on math tasks unless complemented with targeted tuning. The study provides practical guidance on cost-efficient strategies for training instruction-following LLMs in Chinese and emphasizes reproducibility through release of data, models, and code.
Abstract
Recently, the instruction-tuning of large language models is a crucial area of research in the field of natural language processing. Due to resource and cost limitations, several researchers have employed parameter-efficient tuning techniques, such as LoRA, for instruction tuning, and have obtained encouraging results In comparison to full-parameter fine-tuning, LoRA-based tuning demonstrates salient benefits in terms of training costs. In this study, we undertook experimental comparisons between full-parameter fine-tuning and LoRA-based tuning methods, utilizing LLaMA as the base model. The experimental results show that the selection of the foundational model, training dataset scale, learnable parameter quantity, and model training cost are all important factors. We hope that the experimental conclusions of this paper can provide inspiration for training large language models, especially in the field of Chinese, and help researchers find a better trade-off strategy between training cost and model performance. To facilitate the reproduction of the paper's results, the dataset, model and code will be released.