Perplexity-Aware Data Scaling Law: Perplexity Landscapes Predict Performance for Continual Pre-training

TL;DR

This paper introduces a perplexity-aware data scaling law for continual pre-training, enabling better data selection by predicting test loss from perplexity landscapes, thus improving efficiency and performance in domain-specific models.

Contribution

It proposes a novel scaling law that uses perplexity landscapes to guide data selection for continual pre-training, enhancing data efficiency and model performance.

Findings

Effectively predicts test loss using perplexity landscapes.

Achieves superior performance on medical and general benchmarks.

Identifies near-optimal data subsets for training.

Abstract

Continual Pre-training (CPT) serves as a fundamental approach for adapting foundation models to domain-specific applications. Scaling laws for pre-training define a power-law relationship between dataset size and the test loss of an LLM. However, the marginal gains from simply increasing data for CPT diminish rapidly, yielding suboptimal data utilization and inefficient training. To address this challenge, we propose a novel perplexity-aware data scaling law to establish a predictive relationship between the perplexity landscape of domain-specific data and the test loss. Our approach leverages the perplexity derived from the pre-trained model on domain data as a proxy for estimating the knowledge gap, effectively quantifying the informational perplexity landscape of candidate training samples. By fitting this scaling law across diverse perplexity regimes, we enable adaptive selection of high-utility data subsets, prioritizing content that maximizes knowledge absorption while minimizing redundancy and noise. Extensive experiments demonstrate that our method consistently identifies near-optimal training subsets and achieves superior performance on both medical and general-domain benchmarks.