PSB2

该数据集是一个包含25个问题的程序合成基准测试套件，这些问题类似于现实世界中的小型任务。PSB2旨在提供PSB1的一个更加真实和具有挑战性的版本，包含需要不同数据结构和控制流程的问题。这些问题都有英文描述，并包括边缘情况。训练集用于验证，而测试集用于报告结果。该数据集提供了作为测试的输入/输出对。规模上，它包含了25个问题，训练集和测试集各有100万个测试用例。任务类型是程序合成。

This dataset is a program synthesis benchmark suite containing 25 problems modeled after real-world small-scale tasks. PSB2 is designed as a more realistic and challenging successor to PSB1, incorporating problems that demand varied data structures and control flow constructs. Each problem is furnished with an English description and includes edge cases. The training split is utilized for validation, whereas the test split is reserved for reporting final benchmark results. Input/output pairs for testing are provided in this dataset. In terms of scale, the suite encompasses 25 problems, with 1 million test cases for both the training and test sets. The core task of this benchmark is program synthesis.