NEFTune: Noisy Embeddings Improve Instruction Finetuning
Neel Gautam Jain, Ping-yeh Chiang, Yuxin Wen, John Kirchenbauer, Chu, Hong-Min, Gowthami Somepalli, Brian R. Bartoldson, Bhavya Kailkhura, Avi Schwarzschild, Saha, Aniruddha, Micah Goldblum, Jonas Geiping, Tom Goldstein · arXiv (Cornell University) · 2023
We show that language model finetuning can be improved, sometimes dramatically, with a simple augmentation. NEFTune adds noise to the embedding vectors during training. Standard finetuning of LLaMA-2-7B using Alpaca achieves 29.79% on AlpacaEval, which rises to 64.69% using noisy embeddings. NEFTune also improves over strong baselines on modern instruction datasets. Models trained with Evol-Instruct see a 10% improvement, with ShareGPT an 8% improvement, and with OpenPlatypus an 8% improvement. Even powerful models further refined with RLHF such as LLaMA-2-Chat benefit from additional training with NEFTune.