LMEnt：一套从预训练数据到表征分析语言模型知识的工具集

摘要

语言模型（LMs）日益驱动着需要世界知识的现实应用。然而，模型如何将数据转化为对世界的知识和信念的内部过程，目前尚不为人所充分理解。深入这些过程的研究，可能为开发具有更一致、更稳健、更完整知识表示的LMs铺平道路。为便于探讨这些问题，我们推出了LMEnt，一套用于分析LMs在预训练期间知识获取的工具集。LMEnt包含：（1）一个基于维基百科、全面标注实体提及的知识密集型预训练语料库；（2）一种基于实体的预训练数据检索方法，其性能较以往方法提升高达80.4%；以及（3）12个参数规模达10亿、包含4000个中间检查点的预训练模型，在知识基准测试中表现与主流开源模型相当。这些资源共同构建了一个受控环境，用于分析预训练中实体提及与下游性能之间的联系，以及预训练数据中因果干预的影响。通过跨检查点研究知识获取，我们展示了LMEnt的实用性，发现事实频率是关键因素，但并不能完全解释学习趋势。我们发布LMEnt，以支持对LMs中知识的研究，包括知识表示、可塑性、编辑、归因及学习动态等方面。

English

Language models (LMs) increasingly drive real-world applications that require world knowledge. However, the internal processes through which models turn data into representations of knowledge and beliefs about the world, are poorly understood. Insights into these processes could pave the way for developing LMs with knowledge representations that are more consistent, robust, and complete. To facilitate studying these questions, we present LMEnt, a suite for analyzing knowledge acquisition in LMs during pretraining. LMEnt introduces: (1) a knowledge-rich pretraining corpus, fully annotated with entity mentions, based on Wikipedia, (2) an entity-based retrieval method over pretraining data that outperforms previous approaches by as much as 80.4%, and (3) 12 pretrained models with up to 1B parameters and 4K intermediate checkpoints, with comparable performance to popular open-sourced models on knowledge benchmarks. Together, these resources provide a controlled environment for analyzing connections between entity mentions in pretraining and downstream performance, and the effects of causal interventions in pretraining data. We show the utility of LMEnt by studying knowledge acquisition across checkpoints, finding that fact frequency is key, but does not fully explain learning trends. We release LMEnt to support studies of knowledge in LMs, including knowledge representations, plasticity, editing, attribution, and learning dynamics.

LMEnt：一套从预训练数据到表征分析语言模型知识的工具集

LMEnt: A Suite for Analyzing Knowledge in Language Models from Pretraining Data to Representations

摘要

Support