Data Release v1.0

📦 DI-Bench Data Release Page

*.jsonl: DI-Bench dataset

data schema:

{
  "instance_id": "string",
  "metadata": "dict",
  "language": "string",
  "act_command": "string",
  "ci_file": "string",
  "build_files": ["string"],
  "patch": "string",
  "env_specs": {
    "SDK": "string",
    "OS": "string"
  }
}

• instance_id: Repo instance ID (unique).
• metadata: Repo-related details metadata as a dictionary.
• language: Programming language of the repo.
• act_command: Command to trigger the test CI workflow.
• ci_file: Path to the CI configuration file.
• build_files: List of build or project file paths.
• patch: Gold patch in diff format.
• env_specs:
  • SDK: SDK version.
  • OS: OS version.

*.tar.gz

Repositories with masked dependency sections.

licenses.csv

The open-source license of each repository. Please respect the license specified by the repository author. For details, refer to the original repository link.

Repository instances count:

• Regular: 387
• Large: 194