README.md

metadata

pretty_name: LiveSWEBench
language:
  - en
size_categories:
  - 1K<n<10K

LiveSWEBench Tasks

This dataset contains all task instances for the LiveSWEBench benchmark. Tasks are stored in the following format:

1. repo_name (str): the name of the repository for this task
2. task_num (int): the original PR number for the task, used now as an identifier
3. gold_patch (str): the actual changes made in the PR to resolve the issue in this task
4. test_patch (str): the changes made to test files to validate the task solution
5. edit_patch (str): a subset of the gold patch containing changes to just one file
6. autocomplete_patch (str): a subset of hunks from the gold patch for the autocomplete task
7. prompt (str): the prompt for the agent task
8. edit_prompt (str): the prompt used for the edit task
9. autocopmlete_prompts (str): one line snippet for each hunk in autocomplete_patch, to be pasted to trigger the autocomplete
10. commit (str): the base commit hash for the task