Take the Bull by the Horns: Hard Sample-Reweighted Continual Training Improves LLM Generalization

TL;DR

This paper introduces a novel sample reweighting framework called IR-DRO that enhances large language model generalization by selectively emphasizing informative training samples during continual learning.

Contribution

It formalizes a new instance reweighting approach for continual training of LLMs, improving performance by focusing on moderately high-loss, informative samples.

Findings

Significant performance improvements across multiple benchmarks.

Effective in both continual pre-training and instruction tuning.

Framework easily integrates into existing training protocols.

Abstract

In the rapidly advancing arena of large language models (LLMs), a key challenge is to enhance their capabilities amid a looming shortage of high-quality training data. Our study starts from an empirical strategy for the light continual training of LLMs using their original pre-training data sets, with a specific focus on selective retention of samples that incur moderately high losses. These samples are deemed informative and beneficial for model refinement, contrasting with the highest-loss samples, which would be discarded due to their correlation with data noise and complexity. We then formalize this strategy into a principled framework of Instance-Reweighted Distributionally Robust Optimization (IR-DRO). IR-DRO is designed to dynamically prioritize the training focus on informative samples through an instance reweighting mechanism, streamlined by a closed-form solution for straightforward integration into established training protocols. Through rigorous experimentation with various models and datasets, our findings indicate that our sample-targeted methods significantly improve LLM performance across multiple benchmarks, in both continual pre-training and instruction tuning scenarios. Our codes are available at https://github.com/VITA-Group/HardFocusTraining.