github_issues / README.md
helloadhavan's picture
Upload dataset
d9ebbd6 verified
metadata
license: mit
configs:
  - config_name: default
    data_files:
      - split: split1
        path: data/split1-*
      - split: split2
        path: data/split2-*
      - split: split3
        path: data/split3-*
      - split: split4
        path: data/split4-*
dataset_info:
  features:
    - name: repo
      dtype: string
    - name: pr_number
      dtype: int64
    - name: title
      dtype: string
    - name: body
      dtype: string
    - name: buggy_commit
      dtype: string
    - name: fix_commit
      dtype: string
    - name: buggy_distance
      dtype: int64
    - name: confidence
      dtype: string
    - name: files
      list:
        - name: filename
          dtype: string
        - name: patch
          dtype: string
        - name: additions
          dtype: int64
        - name: deletions
          dtype: int64
  splits:
    - name: split1
      num_bytes: 34024342
      num_examples: 6000
    - name: split2
      num_bytes: 31467405
      num_examples: 6000
    - name: split3
      num_bytes: 34011489
      num_examples: 6000
    - name: split4
      num_bytes: 34819671
      num_examples: 6000
  download_size: 50269862
  dataset_size: 134322907
task_categories:
  - text-generation
language:
  - en
tags:
  - code
pretty_name: Github issues dataset
size_categories:
  - 10K<n<100K

GitHub Issues + Fixes Dataset

A curated, high-signal dataset of GitHub issues collected from 25 popular open-source repositories.
Each example pairs a real GitHub issue with the exact code changes (diffs) that resolved it.

The dataset is designed for:

  • Automated bug fixing
  • LLM-based code agents
  • Issue → patch generation
  • Program repair research

How the data was extracted

The data was collected using the GitHub REST API and processed into a structured format.

To maintain quality and usefulness:

  • Only closed issues were considered
  • Each issue must have a clearly associated fix
  • Fixes are stored as unified diffs extracted from the resolving commit
  • Low-signal issues (questions, duplicates, discussions) were filtered out
  • Issues without meaningful code changes were excluded

Each row represents one issue–fix pair.


Dataset structure

Each dataset entry has the following schema:

{
  "repo": "owner/repository",
  "issue_number": 12345,
  "issue_title": "Short description of the problem",
  "issue_body": "Full issue discussion and problem description",
  "commit_sha": "abcdef123456...",
  "files": [
    {
      "filename": "path/to/file.ext",
      "patch": "unified diff showing the fix",
      "additions": 10,
      "deletions": 2
    }
  ]
}
Field Description
repo GitHub repository where the issue originated
issue_number Original GitHub issue number
issue_title Title of the issue
issue_body Full issue description and context
commit_sha Commit that fixed the issue
files List of modified files
files[].filename Path of the modified file
files[].patch Unified diff representing the fix
files[].additions Number of added lines
files[].deletions Number of removed lines

Supported languages

The dataset contains fixes across multiple programming languages, including (but not limited to):

  • C / C++
  • Python
  • JavaScript / TypeScript
  • Rust
  • Go
  • Java
  • Assembly (very rare)

Language distribution varies by repository.

Intended use cases

This dataset is well-suited for:

  • Training models to generate code patches from issue descriptions
  • Evaluating LLM reasoning over real-world bug reports
  • Building autonomous debugging or refactoring agents
  • Research on program repair, code synthesis, and software maintenance

It is not intended for:

  • Issue classification
  • sentiment analysis
  • Chatbot fine-tuning without code generation

Limitations

  • The dataset reflects real-world noise from GitHub issues
  • Issue descriptions vary widely in clarity and detail
  • Some fixes involve refactoring or design changes rather than minimal patches
  • No guarantee that all fixes are optimal or best practice

Warning: This dataset currently has the issues of 10/25 repos and 14k rows but is expected to have 50k rows and 2 GB in size