The surge in Large Language Models (LLMs) has revolutionized natural language processing, but fine-tuning them for specific tasks often encounters challenges in balancing performance and preserving general instruction-following abilities. In this paper, we posit that the distribution gap between task datasets and the LLMs serves as the primary underlying cause. To address the problem, we introduce Self-Distillation Fine-Tuning (SDFT), a novel approach that bridges the distribution gap by guiding fine-tuning with a distilled dataset generated by the model itself to match its original distribution. Experimental results on the Llama-2-chat model across various benchmarks demonstrate that SDFT effectively mitigates catastrophic forgetting while achieving comparable or superior performance on downstream tasks compared to the vanilla fine-tuning. Moreover, SDFT demonstrates the potential to maintain the helpfulness and safety alignment of LLMs. Our code is available at https://github.com/sail-sg/sdft.
@article{arxiv.2402.13669,
title = {Self-Distillation Bridges Distribution Gap in Language Model Fine-Tuning},
author = {Zhaorui Yang and Tianyu Pang and Haozhe Feng and Han Wang and Wei Chen and Minfeng Zhu and Qian Liu},
journal= {arXiv preprint arXiv:2402.13669},
year = {2024}
}