This is a 10,000-record sample. The full dataset (~354,000 records) is available for purchase. Contact andrewcampi456@gmail.com for licensing details.

What is this?

A large-scale dataset of resolved GitHub issues paired with their linked pull requests, code diffs, commits, and review discussions, all from production-quality Python repositories.

Each record captures the complete lifecycle of a software fix:

Issue (problem statement)
  └── Pull Request(s) (proposed fix, discussion)
        └── Code diff (file-level patches)
        └── Review comments (inline feedback)
        └── Commits (change history)

Why it matters

Dataset	Tasks	Notes
SWE-bench Verified	500	Current gold standard for SWE agent evaluation
SWE-bench Full	2,294	Widely used research benchmark
This dataset (full)	354,257	~150x SWE-bench Full

SWE-bench Verified is the benchmark every SWE agent is evaluated against. This dataset is ~150× larger and structured for training, not just evaluation.

Sample stats (this file)

Metric	Value
Records	10,000
Unique repositories	5,040
PRs with code diffs	11,943
PRs with review comments	4,269

Full dataset stats

Metric	Value
Records	354,257
Unique repositories	6,156
Estimated Python code additions	~32M lines
Uncompressed size	6.1 GB
Compressed size	1.4 GB
Data current through	May 2026

Schema

{
  "repo": "pydantic/pydantic",
  "stars": 27668,
  "license": "MIT",
  "issue": {
    "number": 1234,
    "title": "ValidationError message unclear for nested models",
    "body": "When I have a nested model...",
    "state": "closed",
    "labels": ["bug"],
    "created_at": "2025-11-03T10:22:00Z",
    "closed_at": "2025-11-10T14:05:00Z",
    "author": "username"
  },
  "pull_requests": [
    {
      "number": 1235,
      "title": "Improve nested validation error messages",
      "body": "Fixes #1234 by...",
      "state": "closed",
      "merged": true,
      "merged_at": "2025-11-10T14:05:00Z",
      "created_at": "2025-11-07T09:11:00Z",
      "author": "contributor",
      "diff": [
        {
          "filename": "pydantic/errors.py",
          "status": "modified",
          "additions": 12,
          "deletions": 3,
          "patch": "@@ -45,7 +45,16 @@\n ..."
        }
      ],
      "review_comments": [
        {
          "author": "maintainer",
          "body": "Can we add a test for the edge case where...",
          "created_at": "2025-11-08T11:30:00Z"
        }
      ]
    }
  ],
  "commits": [
    {
      "sha": "a1b2c3d4e5f6",
      "message": "Improve nested validation error messages",
      "author": "contributor",
      "date": "2025-11-09T16:45:00Z"
    }
  ]
}

Repository coverage (sample)

Includes issues and PRs from well-known production Python projects across ML, web, data, DevOps, and infrastructure:

apache/airflow · huggingface/transformers · huggingface/diffusers · pydantic/pydantic · ray-project/ray · streamlit/streamlit · sqlfluff/sqlfluff · python-poetry/poetry · pydata/xarray · PrefectHQ/prefect · gradio-app/gradio · Lightning-AI/pytorch-lightning · getmoto/moto · conan-io/conan · zulip/zulip · and 5,000+ more.

Licenses

All source repositories are permissively licensed (MIT or Apache-2.0). The dataset is derived from public GitHub data.

Full dataset

The complete 354,257-record dataset is available for commercial licensing.

Format: JSONL + Parquet (both provided)
Size: 6.1 GB uncompressed / 1.4 GB compressed
License: Commercial license, negotiable
Built: 29-day continuous scrape completed June 2026

Contact: andrewcampi456@gmail.com (or HuggingFace direct message)

Downloads last month: 5

Total file size:

197 MB