High-Quality-Code

license: other
license_name: open-data-attribution-training-disclosure-license-odatl-1.0
license_link: LICENSE
language:
  - en
tags:
  - code
  - synthetic
  - real-code
  - github
  - high-quality
  - max-code
size_categories:
  - 100M<n<1B
  - 1B<n<10B
  - 10B<n<100B

High-Quality-Code: Synthetic + Real (MAXIMUM CODE)

A massive, high-quality code dataset built with maximum code philosophy – as much code as possible.

Components

Synthetic syntax-correction dataset – 5M+ examples across 33 languages (original code_syntax_dataset_1GB.csv)
Real high-quality code from GitHub – 500 top-starred repositories – BOTH zips and extracted source

Current Status: IN PROGRESS

Target: 500 repos
Currently uploaded: 12 extracted repos in github_code/ + 6 zips in github_code_zips/
Upload method: Direct commit to main, zip version (main→master fallback), no size limit, temp files deleted after upload
Process: Running in background – final-max-code-uploader

This README will be updated again when 500 is reached.

Structure (Maximum Code)

.
├── code_syntax_dataset_1GB.csv  # Synthetic: wrong_code, correct_code, explanation, language
├── github_code/                 # Real: Extracted source – MAX CODE
│   ├── owner__repo/
│   └── ...
├── github_code_zips/            # Real: Original zips – MAX PRESERVATION
│   ├── owner__repo.zip
│   └── ...
└── README.md

github_code/ – extracted source code for training (all files, no filtering)
github_code_zips/ – original zip archives for reproducibility

Both are uploaded for maximum code.

Top 500 Source

Fetched via GitHub API: stars:>5000 sort:stars desc – 500 most-starred repos.

First 50:

codecrafters-io/build-your-own-x
sindresorhus/awesome
public-apis/public-apis
freeCodeCamp/freeCodeCamp
EbookFoundation/free-programming-books
openclaw/openclaw
donnemartin/system-design-primer
nilbuild/developer-roadmap
jwasham/coding-interview-university
vinta/awesome-python
awesome-selfhosted/awesome-selfhosted
obra/superpowers
practical-tutorials/project-based-learning
996icu/996.ICU
mattpocock/skills
affaan-m/ECC
react/react
torvalds/linux
NousResearch/hermes-agent
trimstray/the-book-of-secret-knowledge
TheAlgorithms/Python
deepseek-ai/deepseek-harness
vuejs/vue
multica-ai/andrej-karpathy-skills
ossu/computer-science
anomalyco/opencode
n8n-io/n8n
tensorflow/tensorflow
DigitalPlatDev/FreeDomain
trekhleb/javascript-algorithms
ultraworkers/claw-code
microsoft/vscode
yt-dlp/yt-dlp
massgravel/Microsoft-Activation-Scripts
ohmyzsh/ohmyzsh
Significant-Gravitas/AutoGPT
jackfrued/Python-100-Days
CyC2018/CS-Notes
getify/You-Dont-Know-JS
avelino/awesome-go
microsoft/markitdown
ollama/ollama
flutter/flutter
firecrawl/firecrawl
github/gitignore
521xueweihan/HelloGitHub
anthropics/skills
twbs/bootstrap
f/prompts.chat
huggingface/transformers

... and 450 more.

How It Was Built (Maximum Code Philosophy)

# For each repo in top 500:
# 1. Try main.zip, then master.zip, plus codeload URLs
# 2. NO SIZE LIMIT – even torvalds/linux, freeCodeCamp, etc.
# 3. Upload zip to github_code_zips/owner__repo.zip (direct commit)
# 4. Extract zip, upload folder to github_code/owner__repo/ (direct commit)
# 5. Delete local temp immediately – nothing kept locally
# 6. Repeat until 500

Use Cases

Pre-training LLMs on massive real code
Fine-tuning for code correction (synthetic)
Code generation, completion, understanding
Maximum code = maximum learning

Licenses

Synthetic CSV: ODA-TL v1.0
Real GitHub code: Respective original licenses (check each repo's LICENSE)
Use responsibly.