Improving Task Diversity in Label Efficient Supervised Finetuning of LLMs

TL;DR

This paper introduces a simple, confidence-based sampling method leveraging task diversity for label-efficient supervised finetuning of LLMs, achieving comparable or better performance with significantly reduced annotation costs.

Contribution

It proposes a novel inverse confidence weighting sampling strategy that improves data efficiency in supervised finetuning of LLMs by utilizing task labels and model confidence levels.

Findings

Achieves up to 4% higher accuracy than full dataset training.

Reduces annotation costs by up to 80%.

Performs consistently at or above state-of-the-art across datasets.

Abstract

Large Language Models (LLMs) have demonstrated remarkable capabilities across diverse domains, but developing high-performing models for specialized applications often requires substantial human annotation -- a process that is time-consuming, labor-intensive, and expensive. In this paper, we address the label-efficient learning problem for supervised finetuning (SFT) by leveraging task-diversity as a fundamental principle for effective data selection. This is markedly different from existing methods based on the prompt-diversity. Our approach is based on two key observations: 1) task labels for different prompts are often readily available; 2) pre-trained models have significantly varying levels of confidence across tasks. We combine these facts to devise a simple yet effective sampling strategy: we select examples across tasks using an inverse confidence weighting strategy. This produces models comparable to or better than those trained with more complex sampling procedures, while being significantly easier to implement and less computationally intensive. Notably, our experimental results demonstrate that this method can achieve better accuracy than training on the complete dataset (a 4\% increase in MMLU score). Across various annotation budgets and two instruction finetuning datasets, our algorithm consistently performs at or above the level of the best existing methods, while reducing annotation costs by up to 80\%.