Fine-Tuning on Noisy Instructions: Effects on Generalization and Performance

2025-10-03

AI Summary

This work investigates whether introducing artificial perturbations—such as stopword deletion and word-order shuffling—during instruction fine-tuning enhances large language models’ robustness to noisy instructions and generalization capability. We systematically conduct noisy instruction fine-tuning and evaluation on standard benchmarks including MMLU, BBH, and GSM8K, analyzing performance under both clean and perturbed inputs, while tracking training dynamics and behavioral evolution. Results demonstrate that moderate noise injection significantly improves model robustness to ill-formed user instructions and, notably, boosts performance on original (unperturbed) benchmarks in several tasks. We further reveal that input noise acts as an implicit regularizer: it mitigates overfitting to superficial instruction patterns and fosters deeper, more generalizable instruction understanding. These findings establish noisy instruction fine-tuning as a novel paradigm for enhancing robustness in instruction-following models, offering both empirical gains and conceptual insight into the role of perturbation in alignment learning.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Instruction-tuning plays a vital role in enhancing the task-solving abilities of large language models (LLMs), improving their usability in generating helpful responses on various tasks. However, previous work has demonstrated that they are sensitive to minor variations in instruction phrasing. In this paper, we explore whether introducing perturbations in instruction-tuning data can enhance LLMs' resistance against noisy instructions. We focus on how instruction-tuning with perturbations, such as removing stop words or shuffling words, affects LLMs' performance on the original and perturbed versions of widely-used benchmarks (MMLU, BBH, GSM8K). We further assess learning dynamics and potential shifts in model behavior. Surprisingly, our results suggest that instruction-tuning on perturbed instructions can, in some cases, improve downstream performance. These findings highlight the importance of including perturbed instructions in instruction-tuning, which can make LLMs more resilient to noisy user inputs.

Problem

• Enhancing LLMs' resistance to noisy instruction variations through perturbations
• Assessing performance impact on benchmarks when using perturbed training data
• Investigating how instruction-tuning with noise affects generalization capabilities

Innovation

• Fine-tuning with perturbed instructions enhances robustness
• Introducing noise in training improves resistance to variations
• Perturbed instruction-tuning boosts performance on noisy inputs