Formally Verified Automated Programming Progress Standards (FVAPPS)
收藏资源简介:
FVAPPS是一个包含4715个样本的数据集,用于编写程序并证明其正确性,是迄今为止最大的形式验证基准。该数据集由Beneficial AI Foundation和Independent共同开发,基于APPS数据集,包含自然语言问题、难度标记、Python中的正确解决方案以及一些输入输出测试对。数据集通过Anthropic的Claude Sonnet 3.53模型生成,旨在为定理证明和程序合成提供基准。
FVAPPS is a dataset comprising 4,715 samples intended for program writing and correctness verification, and it stands as the largest formal verification benchmark to date. Co-developed by the Beneficial AI Foundation and Independent, this dataset is built upon the APPS dataset and includes natural language problems, difficulty tags, correct Python solutions, as well as select input-output test pairs. Generated using Anthropic's Claude Sonnet 3.53 model, the dataset is designed to provide a benchmark for theorem proving and program synthesis.




