Threshold Filtering Packing for Supervised Fine-Tuning: Training Related Samples within Packs

TL;DR

This paper introduces Threshold Filtering Packing (TFP), a novel data packing method for supervised fine-tuning of autoregressive models that improves performance and fairness by selecting contextually related samples within packs.

Contribution

The paper proposes TFP, a scalable packing strategy that enhances supervised fine-tuning by selecting related samples, leading to significant performance and fairness improvements.

Findings

Up to 7% improvement on GSM8K

Up to 4% improvement on HumanEval

15% boost in fairness metrics

Abstract

Packing for Supervised Fine-Tuning (SFT) in autoregressive models involves concatenating data points of varying lengths until reaching the designed maximum length to facilitate GPU processing. However, randomly concatenating data points can lead to cross-contamination of sequences due to the significant difference in their subject matter. The mainstream approaches in SFT ensure that each token in the attention calculation phase only focuses on tokens within its own short sequence, without providing additional learning signals for the preceding context. To address these challenges, we introduce Threshold Filtering Packing (TFP), a method that selects samples with related context while maintaining sufficient diversity within the same pack. Our experiments show that TFP offers a simple-to-implement and scalable approach that significantly enhances SFT performance, with observed improvements of up to 7\% on GSM8K, 4\% on HumanEval. Furthermore, results from bias benchmark datasets highlight TFP's promising performance in improving fairness while also boosting prediction accuracy by 15\%.