license: apache-2.0
configs:
- config_name: default
data_files:
- split: test
path: data/*.json
Dataset
The following dataset contains logs of the GitHub action for a failed workflow of some commits, followed by the commit that passes the workflow successfully. A full list of the datapoints' content is given below.
Task
The intended task for this dataset is to fix the repo to pass GitHub actions workflow.
Note that the dataset does not contain repo snapshot.
During benchmark, the method clones the necessary repo on the user's local machine.
The user's model should correct the files of the repo, and benchmark pushes repo to GitHub, returning the result of the workflow run aggregated by all datapoints.
List of items of the datapoint:
id: unique id of the dp
language: the main language of the repo
repo_name: original repo name
repo owner: original repo owner
head_branch: name of the original branch that the commit was pushed at
contributor: username of the contributor that committed changes
difficulty: the difficulty of the problem (accessor-based)
sha_fail: sha of the failed commit
sha_success: sha of the successful commit
diff: the content of diff file between failed and successful commits
logs: list of dicts [{"log": log, "step_name": step_name}]:
- log: logs of the failed job, particular step
- step_name: name of the failed step of the job
workflow: content of the workflow file that has been used to run jobs
workflow_filename: workflow filename that has been used to run jobs
workflow_name: name of the workflow that was run
workflow_path: path to the workflow file that was run