CASP
收藏资源简介:
CASP数据集是一个针对C代码形式化验证的评估数据集,包含506对用ANSI/ISO C规范语言(ACSL)编写的形式化规范和相应的C代码。该数据集由AI Sweden、Scania和KTH皇家理工学院的研究人员创建,旨在填补现有数据集在规模和多样性方面的不足,并支持对大型语言模型(LLMs)在代码生成和验证方面的性能进行基准测试。数据集内容丰富,涵盖了从大型开源数据集中提取的C代码及其形式化规范,每个规范-实现对都经过形式化验证。CASP数据集为研究人员提供了一个宝贵的资源,用于训练和评估LLMs在从代码生成规范和反之亦然方面的能力,从而推动软件验证工具的发展,促进更可靠软件系统的创建。
The CASP dataset is an evaluation dataset for formal verification of C code. It contains 506 pairs of formal specifications written in the ANSI/ISO C Specification Language (ACSL) and their corresponding C code. This dataset was created by researchers from AI Sweden, Scania, and the Royal Institute of Technology (KTH), aiming to fill the gaps in scale and diversity of existing datasets and support benchmarking the performance of Large Language Models (LLMs) on code generation and verification tasks. The dataset is rich in content, covering C code and their formal specifications extracted from large-scale open-source datasets, with each specification-implementation pair having undergone formal verification. The CASP dataset provides researchers with a valuable resource for training and evaluating LLMs' capabilities in generating specifications from code and vice versa, thereby advancing the development of software verification tools and facilitating the creation of more reliable software systems.
CASP_dataset 数据集概述
数据集基本信息
- 名称: CASP_dataset
- 来源: Hugging Face 数据集平台
- 存储位置: https://huggingface.co/datasets/nicher92/CASP_dataset
数据规模
- 总样本数: 506 条
- 总数据量: 1,702,402 字节
- 下载大小: 444,955 字节
- 数据拆分: 仅包含训练集(train)
数据特征结构
数据集包含以下10个字段:
| 字段名称 | 数据类型 | 描述 |
|---|---|---|
| file_name | string | 文件名 |
| verified_c_file_content | string | 已验证的C文件内容 |
| dependency_acls | string | 依赖ACLs |
| function_implementation | string | 函数实现 |
| verified | bool | 验证状态 |
| total_goals | int64 | 总目标数 |
| verified_goals | int64 | 已验证目标数 |
| error_cause | string | 错误原因 |
| c_code_snippet | string | C代码片段 |
| acsl_snippet | string | ACSL片段 |
数据配置
- 配置名称: default
- 数据文件路径: data/train-*




