Datasets:

tensorvalley
/

instructed_lint_python_files

license: other
license_name: multiple
license_link: LICENSE
task_categories:
  - text-generation
language:
  - code
tags:
  - python
  - lint
  - ruff
  - code-quality
  - the-stack
  - static-analysis
pretty_name: Instructed Lint Python Files
size_categories:
  - 10M<n<100M
extra_gated_prompt: >-
  ## Terms of Use

  This dataset is derived from
  [bigcode/the-stack-dedup](https://huggingface.co/datasets/bigcode/the-stack-dedup),
  which is a gated dataset with its own Terms of Use. **Before accessing this
  dataset, you must first accept the Terms of Use for The Stack** by visiting
  [bigcode/the-stack-dedup](https://huggingface.co/datasets/bigcode/the-stack-dedup)
  and clicking "Access repository".

  By requesting access to this dataset, you acknowledge that:

  1. **You have accepted The Stack's Terms of Use** at
     [bigcode/the-stack-dedup](https://huggingface.co/datasets/bigcode/the-stack-dedup).

  2. **License compliance**: This dataset contains source code from repositories
  with various
     licenses. Any use of the code must abide by the terms of the original licenses, including
     attribution clauses when relevant. License information is provided in the `max_stars_repo_licenses`
     field of each record.

  3. **Data removal**: The Stack is regularly updated to enact validated data
  removal requests.
     You agree to update your version of this dataset when notified of removals. Follow the
     [update thread](https://huggingface.co/datasets/bigcode/the-stack/discussions/7) for
     notifications, and use the [community discussions](https://huggingface.co/datasets/bigcode/the-stack/discussions/new)
     for questions about dataset versions and allowed uses.

  4. **Redistribution**: To host, share, or otherwise provide access to this
  dataset, you must
     include these Terms of Use and require users to agree to them.

  5. **Contact sharing**: Your contact information (email address and username)
  may be shared
     with the dataset maintainers.

  6. **Security warning**: This dataset contains raw source code from public
  GitHub repositories.
     Some files may contain security vulnerabilities, malicious code, or exploits. **Do not execute
     any code from this dataset without thorough security review.** The lint annotations may help
     identify some issues but are not a substitute for security auditing.

  7. **No warranty**: This dataset is provided "as is" without warranty of any
  kind. The lint
     annotations are generated automatically by ruff and may contain false positives or miss
     real issues.
extra_gated_fields:
  Email: text
  I have accepted The Stack's Terms of Use at bigcode/the-stack-dedup: checkbox
  I agree to update my copy when notified of data removals: checkbox
  I will comply with the original licenses of included source code: checkbox

Instructed Lint Python Files

Lint-annotated Python source code from bigcode/the-stack-dedup, processed with ruff (all 800 stable rules enabled).

Dataset Description

This dataset pairs 12,962,249 Python files from The Stack (deduplicated) with their complete ruff lint diagnostics. Each record contains the original source code, file metadata, license information, and structured lint results.

Motivation

Building training data for code quality models requires large-scale lint annotations. Running ruff at scale on millions of files is expensive with the standard CLI approach. We built ruff-batch, a Rust binary that calls ruff's internal lint_only() API directly via Rayon parallel iteration, achieving 26,337 files/sec with all 800 rules enabled — a 10x speedup over the ruff CLI.

Source Data

Origin: bigcode/the-stack-dedup, Python split
Files: 12,962,249
Linter: ruff (all 800 stable rules, preview disabled)
Processing time: 8.2 minutes on Apple M3 Ultra (28 cores)

Dataset Structure

Each record contains:

Field	Type	Description
`id`	string	Unique file identifier (git hexsha from The Stack)
`code`	string	Python source code
`path`	string	Original file path in the source repository
`max_stars_repo_licenses`	list[string]	SPDX license identifiers for the most-starred repo containing this file
`max_stars_repo_name`	string	GitHub repository name (owner/repo)
`size`	int	File size in bytes
`diagnostics`	list[object]	Lint diagnostics (see below)
`num_diagnostics`	int	Number of lint issues found
`valid_syntax`	bool	Whether the file has valid Python syntax
`lint_elapsed_us`	int	Microseconds spent linting this file

Diagnostic Structure

Each diagnostic in the diagnostics list contains:

Field	Type	Description
`code`	string	Ruff rule code (e.g., "F401", "E711")
`message`	string	Human-readable description of the issue
`row`	int	1-indexed start line number
`col`	int	1-indexed start column number
`end_row`	int	1-indexed end line number
`end_col`	int	1-indexed end column number

Statistics

Total files: 12,962,249
Valid syntax: 95.6%
Average diagnostics per file: ~102 (with all 800 rules)
Total diagnostics: ~1.3 billion

Usage

from datasets import load_dataset

ds = load_dataset("tensorvalley/instructed_lint_python_files", split="train", streaming=True)

for record in ds:
    print(f"File: {record['path']}")
    print(f"License: {record['max_stars_repo_licenses']}")
    print(f"Issues: {record['num_diagnostics']}")
    for diag in record['diagnostics']:
        span = f"L{diag['row']}:{diag['col']}-L{diag['end_row']}:{diag['end_col']}"
        print(f"  {diag['code']} [{span}]: {diag['message']}")
    break

Licensing and Attribution

This dataset contains source code from GitHub repositories with various open-source licenses. The license for each file is provided in the max_stars_repo_licenses field. Any use of the code must comply with the terms of the original licenses, including attribution requirements.

The lint annotations themselves are provided under Apache-2.0.

For data provenance, each record includes:

id: The git hexsha, which can be used to trace back to the original commit
max_stars_repo_name: The GitHub repository name for attribution
path: The original file path within the repository

Data Removal

This dataset inherits The Stack's data removal policy. If your code is included and you wish to have it removed, please follow the opt-out process at Am I in The Stack?

When The Stack publishes data removal updates, this dataset will be updated accordingly. Follow the update thread for notifications.

Security Warning

This dataset contains raw source code from public GitHub repositories. Some files may contain:

Security vulnerabilities (SQL injection, XSS, command injection, etc.)
Intentionally malicious code
Backdoors or exploits
Credential leaks or hardcoded secrets

Do not execute any code from this dataset without thorough security review. The ruff lint annotations identify some code quality issues but are not a security audit and should not be treated as one.

Citation

@misc{{instructed_lint_python_files,
    title={{Instructed Lint Python Files}},
    author={{Tensor Valley}},
    year={{2026}},
    howpublished={{\url{{https://huggingface.co/datasets/tensorvalley/instructed_lint_python_files}}}}
}}

If you use the underlying source code, please also cite The Stack:

@inproceedings{{kocetkov2022thestack,
    title={{The Stack: 3 TB of permissively licensed source code}},
    author={{Denis Kocetkov and Raymond Li and Loubna Ben Allal and Jia Li and Chenghao Mou
             and Carlos Mu{{\~n}}oz Ferrandis and Yacine Jernite and Margaret Mitchell
             and Sean Hughes and Thomas Wolf and Dzmitry Bahdanau and Leandro von Werra
             and Harm de Vries}},
    booktitle={{Transactions on Machine Learning Research}},
    year={{2022}}
}}

Terms of Use for The Stack

The Stack dataset is a collection of source code in over 300 programming languages. We ask that you read and acknowledge the following points before using the dataset:

The Stack is a collection of source code from repositories with various licenses. Any use of all or part of the code gathered in The Stack must abide by the terms of the original licenses, including attribution clauses when relevant. We facilitate this by providing provenance information for each data point.
The Stack is regularly updated to enact validated data removal requests. By accessing this dataset, you agree to update your own version of The Stack to the most recent usable version specified by the maintainers in the following thread. If you have questions about dataset versions and allowed uses, please also ask them in the dataset's community discussions. We will also notify users via email when the latest usable version changes.
To host, share, or otherwise provide access to The Stack dataset, you must include these Terms of Use and require users to agree to it.
By accessing this dataset, you accept that your contact information (email address and username) can be shared with the dataset maintainers as well.

Artifacts

Rule frequency stats (rule_stats.json): per-rule counts over the dataset split (how many files contain each diagnostic code + total diagnostic occurrences), used to pick a balanced ~100-rule subset for GRPO.
- https://huggingface.co/datasets/tensorvalley/instructed_lint_python_files/blob/main/artifacts/rule_stats.json
Rule stats: https://huggingface.co/datasets/tensorvalley/instructed_lint_python_files/blob/main/artifacts/rule_stats.json