Cuarzo-100K v2
- Type
- dataset
- Venue
- Cuarzo-AI / Hugging Face
- Year
- 2026
- Source
- huggingface
- Access
- free
- Language
- Python, English, Spanish, French, Mandarin Chinese
- Added
- 2026-07-17T20:18:35.170058+00:00
- Verified
- 2026-07-17T20:18:35.170058+00:00
Summary
A dataset of 99,683 bidirectional deterministically verified Python-to-human-language pairs across English, Spanish, French, and Mandarin Chinese, generated by Aether (Cuarzo AI's proprietary engine for generating deterministic paired data between code and human language). Each record contains original Python from StarCoderData, structured natural language representations in all four languages, and roundtrip-validated Python regenerated from each language representation with AST equality and compilation checks. The v2 release adds Mandarin Chinese as a full fourth language surface and expands the verification schema to per-language roundtrip checks across all four surfaces.
Keywords
python code multilingual translation code-generation natural-language starcoderdata roundtrip-verification
Topics
Code / Multilingual NLP
Research notes
- Curated from StarCoderData's Python stream using a multi-stage acceptance pipeline including AST parsing, compile checks, Aether translation, and per-language roundtrip validation. Only records passing every check across all four surfaces are included. Contact: hello@cuarzoai.com. The HF page returned 401 on direct fetch; info gathered from search results and HF post.