AutoMathText 2.5 🚀 AutoMathText 2.5: A Foundational High Quality STEM Training Dataset 📊 AutoMathText 2.5 consists of over 2 trillion tokens of high quality, deduplicated text spanning web content, mathematics, code, reasoning, and bilingual data. This dataset was meticulously curated using a three tier deduplication pipeline and AI powered quality assessment to provide superior training data for large language models. Our dataset combines 50+ premium data sources with advanced processing techniques, including semantic deduplication , contamination detection , and intelligent text cleaning to deliver exceptional model performance across diverse domains. 📜 Licensing & Citation License Released under AutoMathText Data Agreement for Model Training (See LICENSE). Citation
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy