HoloBench
收藏资源简介:
HoloBench是由Megagon Labs创建的一个用于评估长上下文语言模型(LCLMs)在处理大规模文本数据时进行整体推理能力的新型基准框架。该数据集包含90个问题,涵盖了从简单到复杂的多种查询类型,旨在测试模型在不同信息密度、分布和查询复杂度下的表现。HoloBench通过将数据库推理操作引入文本上下文,实现了对LCLMs在长文本中进行信息聚合、比较和推理能力的系统评估。该数据集的应用领域主要集中在需要多文档综合、比较分析和跨大量数据集进行上下文整合的任务,旨在解决传统检索增强生成模型在复杂跨文档任务中的局限性。
HoloBench is a novel benchmark framework created by Megagon Labs for evaluating the holistic reasoning capabilities of long-context language models (LCLMs) when processing large-scale text data. This dataset includes 90 questions covering diverse query types ranging from simple to complex, aiming to test model performance under varying information density, distribution and query complexity. HoloBench introduces database reasoning operations into textual contexts, enabling systematic evaluation of LCLMs' abilities to conduct information aggregation, comparison and reasoning within long texts. Its application fields mainly focus on tasks requiring multi-document synthesis, comparative analysis and context integration across large-scale datasets, with the goal of addressing the limitations of traditional retrieval-augmented generation models in complex cross-document tasks.

- 1Holistic Reasoning with Long-Context LMs: A Benchmark for Database Operations on Massive Textual DataMegagon Labs · 2024年



