JetBrains-Research/diff-xyz

Diff-XYZ是一个用于评估代码变更理解能力的数据集，包含从CommitPackFT数据集中筛选出的1000个真实世界代码编辑案例。每个案例包括原始代码内容(old_code)、修改后的代码内容(new_code)和多种差分表示(udiff, udiff-h, udiff-l, search-replace)。这些格式使得可以对大型语言模型在三种代码编辑任务上的能力进行评估：应用变更、撤销变更和生成差分。

Diff-XYZ is a dataset for evaluating code change understanding capabilities, containing 1,000 real-world code edit instances sampled and filtered from the CommitPackFT dataset. Each instance includes the original code content (old_code), modified code content (new_code), and multiple diff representations (udiff, udiff-h, udiff-l, search-replace). These formats enable the evaluation of large language models on three code editing tasks: applying changes, reversing changes, and generating diffs.