Understanding and Improving Noisy Embedding Techniques in Instruction Finetuning
For practitioners of instruction fine-tuning, this work provides a better noise-based method (SymNoise) that significantly improves model performance over the current state-of-the-art.
The paper analyzes why uniform noise outperforms Gaussian noise in NEFTune, finding comparable performance theoretically and empirically. It introduces SymNoise, a symmetric noise method that improves LLaMA-2-7B's AlpacaEval score from 29.79% to 69.04%, a 6.7% gain over NEFTune (64.69%), and consistently outperforms NEFTune on various models and datasets.
Recent advancements in instructional fine-tuning have injected noise into embeddings, with NEFTune (Jain et al., 2024) setting benchmarks using uniform noise. Despite NEFTune's empirical findings that uniform noise outperforms Gaussian noise, the reasons for this remain unclear. This paper aims to clarify this by offering a thorough analysis, both theoretical and empirical, indicating comparable performance among these noise types. Additionally, we introduce a new fine-tuning method for language models, utilizing symmetric noise in embeddings. This method aims to enhance the model's function by more stringently regulating its local curvature, demonstrating superior performance over the current method, NEFTune. When fine-tuning the LLaMA-2-7B model using Alpaca, standard techniques yield a 29.79% score on AlpacaEval. However, our approach, SymNoise, increases this score significantly to 69.04%, using symmetric noisy embeddings. This is a 6.7% improvement over the state-of-the-art method, NEFTune (64.69%). Furthermore, when tested on various models and stronger baseline instruction datasets, such as Evol-Instruct, ShareGPT, OpenPlatypus, SymNoise consistently outperforms NEFTune. The current literature, including NEFTune, has underscored the importance of more in-depth research into the application of noise-based strategies in the fine-tuning of language models. Our approach, SymNoise, is another significant step towards this direction, showing notable improvement over the existing state-of-the-art method.