Pretraining Data Detection for Large Language Models: A Divergence-based Calibration Method

TL;DR

This paper introduces a divergence-based calibration method for detecting whether texts were part of an LLM's training data, improving accuracy over existing methods, and provides a new Chinese benchmark for evaluation.

Contribution

The paper proposes a novel divergence-based calibration approach for pretraining data detection and introduces the PatentMIA benchmark for Chinese texts.

Findings

The proposed method outperforms existing detection approaches.

Experimental results show significant accuracy improvements.

The method is effective on both English and Chinese datasets.

Abstract

As the scale of training corpora for large language models (LLMs) grows, model developers become increasingly reluctant to disclose details on their data. This lack of transparency poses challenges to scientific evaluation and ethical deployment. Recently, pretraining data detection approaches, which infer whether a given text was part of an LLM's training data through black-box access, have been explored. The Min-K\% Prob method, which has achieved state-of-the-art results, assumes that a non-training example tends to contain a few outlier words with low token probabilities. However, the effectiveness may be limited as it tends to misclassify non-training texts that contain many common words with high probabilities predicted by LLMs. To address this issue, we introduce a divergence-based calibration method, inspired by the divergence-from-randomness concept, to calibrate token probabilities for pretraining data detection. We compute the cross-entropy (i.e., the divergence) between the token probability distribution and the token frequency distribution to derive a detection score. We have developed a Chinese-language benchmark, PatentMIA, to assess the performance of detection approaches for LLMs on Chinese text. Experimental results on English-language benchmarks and PatentMIA demonstrate that our proposed method significantly outperforms existing methods. Our code and PatentMIA benchmark are available at https://github.com/zhang-wei-chao/DC-PDD.