Memorization, or the tendency of large language models (LLMs) to output entire sequences from their training data verbatim, is a key concern for safely deploying language models. In particular, it is vital to minimize a model's memorization of sensitive datapoints such as those containing personal identifiable information (PII). The prevalence of such undesirable memorization can pose issues for model trainers, and may even require discarding an otherwise functional model. We therefore seek to predict which sequences will be memorized before a large model's full train-time by extrapolating the memorization behavior of lower-compute trial runs. We measure memorization of the Pythia model suite, and find that intermediate checkpoints are better predictors of a model's memorization behavior than smaller fully-trained models. We additionally provide further novel discoveries on the distribution of memorization scores across models and data.

翻译：记忆化，即大型语言模型（LLM）逐字输出训练数据中完整序列的倾向，是安全部署语言模型的关键考量。特别地，最小化模型对敏感数据点（如包含个人身份信息（PII）的数据）的记忆化至关重要。此类不良记忆化的普遍存在可能给模型训练者带来问题，甚至可能迫使丢弃原本功能正常的模型。因此，我们试图通过外推低计算量试运行的记忆化行为，在大型模型完成完整训练之前预测哪些序列会被记忆化。我们测量了Pythia模型套件的记忆化程度，发现中间检查点比小型完整训练模型能更准确地预测模型的记忆化行为。此外，我们还提出了关于模型与数据间记忆化分数分布的新发现。