Data Proportion Detection for Optimized Data Management for Large Language Models

TL;DR

This paper introduces a novel method for automatically estimating the optimal data proportions for pre-training large language models by analyzing their outputs, supported by theoretical proofs, algorithms, and preliminary experiments.

Contribution

It presents the concept of data proportion detection, providing the first theoretical and practical framework for estimating data proportions in LLM pre-training.

Findings

Theoretical proofs support data proportion detection.

Practical algorithms for estimating data proportions.

Preliminary experimental results demonstrate feasibility.

Abstract

Large language models (LLMs) have demonstrated exceptional performance across a wide range of tasks and domains, with data preparation playing a critical role in achieving these results. Pre-training data typically combines information from multiple domains. To maximize performance when integrating data from various domains, determining the optimal data proportion is essential. However, state-of-the-art (SOTA) LLMs rarely disclose details about their pre-training data, making it difficult for researchers to identify ideal data proportions. In this paper, we introduce a new topic, \textit{data proportion detection}, which enables the automatic estimation of pre-training data proportions by analyzing the generated outputs of LLMs. We provide rigorous theoretical proofs, practical algorithms, and preliminary experimental results for data proportion detection. Based on these findings, we offer valuable insights into the challenges and future directions for effective data proportion detection and data management.