TL;DR
This paper presents Low-Confidence Gold (LCG), a filtering framework that improves instruction tuning datasets by selecting high-quality, diverse samples, leading to better model performance with fewer data.
Contribution
Introduces LCG, a semi-supervised filtering method using clustering and confidence-guided selection to enhance dataset quality for instruction tuning.
Findings
Models fine-tuned on LCG subsets outperform existing methods.
LCG achieves substantial improvements on MT-bench.
Open-source assets are available at the provided GitHub link.
Abstract
The effectiveness of instruction fine-tuning for Large Language Models is fundamentally constrained by the quality and efficiency of training datasets. This work introduces Low-Confidence Gold (LCG), a novel filtering framework that employs centroid-based clustering and confidence-guided selection for identifying valuable instruction pairs. Through a semi-supervised approach using a lightweight classifier trained on representative samples, LCG curates high-quality subsets while preserving data diversity. Experimental evaluation demonstrates that models fine-tuned on LCG-filtered subsets of 6K samples achieve superior performance compared to existing methods, with substantial improvements on MT-bench and consistent gains across comprehensive evaluation metrics. The framework's efficacy while maintaining model performance establishes a promising direction for efficient instruction…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Code & Models
Videos
