CommitBART

TextsIntroduced 2022-08-17

CommitBART is a benchmark for researching commit-related task such as denoising, cross-modal generation and contrastive learning. The dataset contains over 7 million commits across 7 programming languages.

Source: CommitBART: A Large Pre-trained Model for GitHub Commits

Image Source: https://arxiv.org/pdf/2208.08100v1.pdf