Sci-Base: The Largest AI-Ready Scientific Foundation Dataset

🌌 The Sciverse Data Foundation

Sciverse is a comprehensive, multi-layered scientific data foundation designed to provide the ultimate data infrastructure for the AI for Science (AI4S) community. As scientific research becomes increasingly data-driven, Sciverse supplies the essential, high-quality data resources required to build robust scientific knowledge systems and accelerate research.

Sciverse consists of three core data pillars:

• Sci-Base (Scientific Knowledge Base Data): The massive-scale, purely objective scientific knowledge base. Comprising over 25 million deeply cleaned and parsed Open Access documents, it provides the comprehensive, purely factual scientific corpus that serves as the universal foundation for all downstream scientific applications.
• Sci-Align (Scientific Multi-Alignment Data): A highly curated, structured dataset mapping direct scientific relationships and precise factual alignments. It focuses on well-defined entity interactions—such as mapping specific chemical reaction pathways (e.g., via SMILES strings), condition-to-result pairings, and standardized structural descriptions. This layer provides the structured factual alignment needed for models to accurately connect and ground foundational scientific concepts.
• Sci-Evo (Scientific Evolution Data): A multi-layered, high-density reasoning dataset designed for complex problem-solving and deep scientific evaluation. Going beyond basic facts, this layer captures deep, causal descriptions—detailing not just the 'what', but the underlying reasoning for specific experimental designs, multi-step mathematical derivations, and the complex logic of how modifying specific conditions alters outcomes. It is constructed to rigorously measure a model's advanced scientific reasoning accuracy and logical depth.

---

📊 Dataset Summary: Sci-Base

Unlike simple aggregations of traditional raw data, the core advantage of Sci-Base lies in its "deep parsing" and "logical reconstruction." Leveraging MinerU—an advanced, intelligent document parsing engine designed for the large-model era—our team has performed a "pixel-level" digital reconstruction of over 25 million Open Access scientific papers and books.

Through the deep structural processing of complex mathematical equations, chemical formulas, and high-precision charts, Sci-Base successfully transforms fragmented academic documents into over 600 billion truly AI-Ready, pure tokens. This profound level of parsing not only flawlessly preserves the invaluable logical chains and original typographical structures inherent in scientific literature, but it also achieves a fundamental leap, converting massive raw data into a high-value digital asset.

Sci-Base marks the initial completion of a world-leading, ultra-pure scientific knowledge foundation. It achieves a comprehensive integration of basic sciences and applied engineering by deeply covering the following 10 core scientific disciplines:

• Mathematics and Computational Science
• Physics
• Chemistry
• Life Sciences
• Earth and Atmospheric Sciences
• Astronomy and Space Sciences
• Medicine and Health Sciences
• Materials Science and Engineering
• Energy and Power Science
• Engineering and Manufacturing Science

✨ Key Highlights

• 📈 Unprecedented Scale: Contains over 25 million high-quality scientific documents and 600 billion+ pure tokens, making it the largest dataset of its kind currently available.
• 🎯 High-Precision Parsing (Powered by MinerU): Deep structural processing flawlessly preserves the logical chains of complex mathematical equations, original typographical layouts, and the precise positional relationships of charts and figures. Mathematical formulas and structural context are seamlessly restored.
• ⏱️ Highly Up-to-Date: Knowledge cutoff extends to March 2026, ensuring the dataset reflects the latest scientific breakthroughs. We are committed to continuously tracking and integrating new open-access research.
• 🧬 Rich Scientific Entities: With a strong focus on core domains like Life Sciences, Physical Sciences, and Earth & Atmospheric Sciences, the dataset embeds hundreds of millions of scientific entities within their correct contextual environments.

---

🛠️ Dataset Structure

A typical instance in the dataset represents a single scientific document (paper or book chapter), provided in a clean, highly structured format.

(Example JSON structure - Please adjust based on your actual schema)

{
  "doc_id": "arxiv-...",
  "title": "...",
  "authors": ["..."],
  "domain": "Physics",
  "publication_date": "2026-02-15",🚀
  "text": "The parsed, highly-structured scientific text...",
  "metadata": {
    "license": "CC-BY-4.0",
    "source": "...",
    "entity_count": 1450
  }
}

🚀 How to Use

You can easily load Sci-Base using the Hugging Face datasets library.

from datasets import load_dataset

# Load the entire dataset (Note: This is very large!)
dataset = load_dataset("opendatalab/Sci-Base")

# Load a specific domain, e.g., Life Sciences
life_sciences_data = load_dataset("opendatalab/Sci-Base", "life_sciences")

📄 License

Sci-Base is provided as a foundational data resource. The licensing of this dataset consists of two components:

1. Dataset Structure and Processed Format: The aggregated dataset structure, structural annotations, metadata, and the deeply cleaned, MinerU-parsed formatting (including Markdown tables and LaTeX formulas) are released under the CC-BY 4.0.
2. Original Document Content: The underlying text, figures, and tables extracted from the scientific papers and books retain their original Open Access (OA) licenses. These are predominantly variations of Creative Commons licenses (such as CC-BY, CC-BY-NC, etc.) or specific publisher OA agreements.

Important Notice for Users: While we have implemented strict filtering to include only legally compliant Open Access content, it is the responsibility of the user to ensure that their specific downstream applications, modifications, or commercial uses comply with the individual license terms of the original source materials.