ReactiveAI/Beta-Pre-Train-Corpus

Reactive AI / Beta Pre-Train Corpus

Pre-training corpus for RxT-Beta models, created from public & open datasets. Includes high-quality english and polish web crawl data, mathematic and scientific subsets, and code in different programming languages.

2k subsets are filtered for 1024-2048 tokens, except MegaMath Web Pro and GitHub Code subsets, that were filtered for 512-2048 tokens

Subsets & original datasets

• FineWeb-Edu
  • fineweb-edu-s100 (51.3M examples) - 50% of 'sample-100BT' subset
  • fineweb-edu-s100-sh3-1k (18.3M examples)
  • fineweb-edu-s100-sh3-2k (4.69M examples) - third 25% of 'sample-100BT' subset filtered for 1024-2048 tokens examples
  • fineweb-edu-s100-sh3-4k (2.32M examples)
  • fineweb-edu-s100-sh4-1k (14.8M examples)
  • fineweb-edu-s100-sh4-2k (3.90M examples) - fourth 25% of 'sample-100BT' subset filtered for 1024-2048 tokens examples
  • fineweb-edu-s100-sh4-4k (1.86M examples)
  • fineweb-edu-2025-08-2k (3.77M examples) - CC-MAIN-2025-08 subset, filtered for 1024-2048 tokens examples
  • fineweb-edu-2025-13-2k (3.95M examples) - CC-MAIN-2025-13 subset, filtered for 1024-2048 tokens examples
  • fineweb-edu-2025-18-1k (11.3M examples)
  • fineweb-edu-2025-18-2k (4.13M examples) - CC-MAIN-2025-18 subset, filtered for 1024-2048 tokens examples
  • fineweb-edu-2025-18-4k (1.88M examples)
  • fineweb-edu-2025-21-1k (10.1M examples)
  • fineweb-edu-2025-21-2k (3.76M examples) - CC-MAIN-2025-21 subset, filtered for 1024-2048 tokens examples
  • fineweb-edu-2025-21-4k (1.73M examples)
  • fineweb-edu-2025-26 (14M examples) - CC-MAIN-2025-26 subset - latest crawl
• FineWiki
  • finewiki-en (6.6M examples) - english wikipedia subset
  • finewiki-pl (1.5M examples) - polish wikipedia subset
• FineWeb
  • fineweb-s10-2k (1.73M examples) - 'sample-10BT' subset, filtered for 1024-2048 tokens examples
• FineWeb2-HQ
  • fineweb2-hq-pl (13.3M examples) - high-quality filtered polish web crawl
• FineWeb-2
  • fineweb2-pl-2k (2.13M examples) - latest 30% of polish subset, filtered for 1024-2048 tokens examples
  • fineweb2-pl-1k (8M examples)
• DCLM-Edu
  • dclm-edu-2k (3.59M examples) - examples filtered for 3 and higher educational score and 1024-2048 tokens
  • dclm-edu-s1 (1.91M examples)
  • dclm-edu-s2 (1.93M examples)
  • dclm-edu-s3 (1.90M examples)
  • dclm-edu-s4 (1.93M examples)
  • dclm-edu-s5 (1.92M examples)
  • dclm-edu-s6 (1.93M examples)
  • dclm-edu-s7 (1.92M examples)
  • dclm-edu-s8 (1.91M examples)
  • dclm-edu-s9 (1.92M examples)
  • dclm-edu-s10 (1.93M examples)
  • dclm-edu-s11 (1.92M examples)
  • dclm-edu-s12 (1.91M examples)
  • dclm-edu-s13 (1.91M examples)
• FinePdfs-Edu
  • finepdfs-edu-en-2k (4.38M examples) - english subset filtered for 1024-2048 tokens
  • finepfds-edu-pl-2k (0.18M examples) - polish subset filtered for 1024-2048 tokens (not used, as it's duplicated in base finepdfs-pl-2k)
  • finepdfs-edu-en-sh5-0-1k (1.88M examples)
  • finepdfs-edu-en-sh5-1-1k (1.87M examples)
  • finepdfs-edu-en-sh5-2-1k (1.86M examples)
  • finepdfs-edu-en-sh5-3-1k (1.84M examples)
  • finepdfs-edu-en-sh5-4-1k (1.78M examples)
• FinePdfs
  • finepfds-pl-2k (1.65M examples) - polish subset filtered for 1024-2048 tokens
  • finepdfs-pl-1k (4.6M examples)
  • finepdfs-pl-4k (1.3M examples)
• FineMath
  • finemath-4plus (6.7M examples) - full subset with best math quality
  • infiwebmath-4plus (6.3M examples) - full subset with best math quality
• MegaMath
  • megamath-web-pro-2k (5.38M examples) - filtered for 512-2048 tokens MegaMath Web Pro subset
• MegaMath-Web-Pro-Max
  • megamath-web-pro-max-sh4-0 (6.26M examples)
  • megamath-web-pro-max-sh4-1 (5.98M examples)
  • megamath-web-pro-max-sh4-2 (5.69M examples)
  • megamath-web-pro-max-sh4-3 (5.77M examples)
• ProofPile-2
  • pp2-arxiv (4M examples) - arXiv research papers
  • pp2-algebraic-stack (3.37M examples) - mathematic code examples
• Notebooks:
  • kaggle (0.58M examples) - from HuggingFaceTB/issues-kaggle-notebooks
  • github-jupyter (0.05M examples) - from codeparrot/github-jupyter-code-to-text
  • notebooks-all (0.63M examples) - kaggle and github-jupyter combined
• Code:
  • beta-code-short (4.51M examples) - filtered and combined code from codeparrot/codeparrot-clean and codeparrot/github-code-clean
  • github-code-clean-small-2k (8.17M examples) - filtered for 512-2048 tokens code from loubnabnl/github-code-clean-small
  • Stack-Edu (~28.15M examples)
    • stack-edu-c (1.38M examples)
    • stack-edu-cpp (1.9M examples)
    • stack-edu-csharp (1.7M examples)
    • stack-edu-go (0.77M examples)
    • stack-edu-java (3.99M examples)
    • stack-edu-javascript (2.71M examples)
    • stack-edu-markdown (3.28M examples)
    • stack-edu-php (1.67M examples)
    • stack-edu-python (5.6M examples)
    • stack-edu-ruby (0.96M examples)
    • stack-edu-rust (0.51M examples)
    • stack-edu-shell (1.3M examples)
    • stack-edu-sql (0.81M examples)
    • stack-edu-swift (0.63M examples)
    • stack-edu-typescript (0.96M examples)