FailSafeQA
收藏资源简介:
FailSafeQA数据集是由Writer, Inc创建的,用于测试大型语言模型在金融领域中面对用户界面交互变化的鲁棒性和上下文感知能力。该数据集包含220个示例,每个示例由一个上下文和五个问题组成,其中包括原始问题、三个扰动变种和一个无关问题。数据集的上下文是截断的美国公开上市公司10-K年度报告,这些问题和上下文经过人工和自动化的生成和扰动处理。该数据集旨在评估LLM在金融应用中的可靠性,并解决模型在面对非理想条件下的输入扰动时的问题。
The FailSafeQA dataset was developed by Writer, Inc. to evaluate the robustness and contextual awareness of large language models (LLMs) when faced with user interface interaction variations in the financial domain. This dataset consists of 220 instances, each comprising a context and five questions: the original query, three perturbed variants, and one irrelevant question. The contexts within the dataset are truncated 10-K annual reports of publicly traded U.S. companies, and both the questions and contexts were generated and perturbed through a combination of manual and automated processes. This dataset is designed to assess the reliability of LLMs in financial applications and address the challenges that models encounter when exposed to input perturbations under non-ideal conditions.
数据集概述
数据集名称
FailSafeQA
许可
MIT
任务类别
- 文本生成
- 问题回答
语言
- 英语 (en)
标签
- 金融 (finance)
- 合成 (synthetic)
数据配置
- 配置名称:默认
- 测试集:
FAILSAFEQA_benchmark_data-FINAL.jsonl
- 测试集:
数据集简介
该数据集是基于论文《Expect the Unexpected: FailSafeQA Long Context for Finance》引入的基准数据。
数据集数量
220
数据结构
json { "idx": int, "tokens": int, "context": string, "ocr_context": string, "answer": string, "query": string, "incomplete_query": string, "out-of-domain_query": string, "error_query": string, "out-of-scope_query": string, "citations": string, "citations_tokens": int }
任务多样性
- 根动词及其直接宾语:展示了每个标准化查询的第一句中的前20个动词及其前五个直接宾语。
- 任务类型:
- 83.0% 问题回答 (QA)
- 17.0% 涉及文本生成 (TG)
引用信息
@misc{kiran2024failsafeqa, title={Expect the Unexpected: FailSafeQA Long Context for Finance}, author={Kiran Kamble and Melisa Russak and Dmytro Mozolevskyi and Muayad Ali and Mateusz Russak and Waseem AlShikh}, year={2024}, eprint={todo}, archivePrefix={arXiv}, primaryClass={cs.CL} }




