AdaSTaR: Adaptive Data Sampling for Training Self-Taught Reasoners

TL;DR

AdaSTaR is a novel adaptive sampling algorithm that improves the training efficiency and accuracy of self-improving reasoning language models by balancing data diversity and difficulty.

Contribution

Introduces AdaSTaR, an adaptive sampling method that enhances self-taught reasoning models by balancing data diversity and difficulty dynamically during training.

Findings

Achieves best test accuracy in all six benchmarks.

Reduces training FLOPs by an average of 58.6%.

Generalizes across different pre-trained LMs and larger models.

Abstract

Self-Taught Reasoners (STaR), synonymously known as Rejection sampling Fine-Tuning (RFT), is an integral part of the training pipeline of self-improving reasoning Language Models (LMs). The self-improving mechanism often employs random observation (data) sampling. However, this results in trained observation imbalance; inefficiently over-training on solved examples while under-training on challenging ones. In response, we introduce Adaptive STaR (AdaSTaR), a novel algorithm that rectifies this by integrating two adaptive sampling principles: (1) Adaptive Sampling for Diversity: promoting balanced training across observations, and (2) Adaptive Sampling for Curriculum: dynamically adjusting data difficulty to match the model's evolving strength. Across six benchmarks, AdaSTaR achieves best test accuracy in all instances (6/6) and reduces training FLOPs by an average of 58.6% against an extensive list of baselines. These improvements in performance and efficiency generalize to different pre-trained LMs and larger models, paving the way for more efficient and effective self-improving LMs.