STAR: Constraint LoRA with Dynamic Active Learning for Data-Efficient Fine-Tuning of Large Language Models
We propose a novel DEFT (Data-Efficient Fine-Tuning) method, STAR, to effectively combine PEFT (Parameter Efficient Fine-Tuning) with active learning through criterion revision and model regularization.
conda create -n star python=3.10
conda activate star
pip install -r requirements.txtOur experiments are carried out with an NVIDIA A100 80GB GPU.
cd src
bash run_star.sh ${dataset} ${al}
This work is implemented by LoftQ and adapter-al. Sincere thanks for their efforts.
@misc{zhang2024star,
title={STAR: Constraint LoRA with Dynamic Active Learning for Data-Efficient Fine-Tuning of Large Language Models},
author={Linhai Zhang and Jialong Wu and Deyu Zhou and Guoqiang Xu},
year={2024},
eprint={2403.01165},
archivePrefix={arXiv},
primaryClass={cs.CL}
}