AutoMathText-2.5
π AutoMathText-2.5: A Foundational High-Quality STEM Training Dataset
Β
π 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+β¦ See the full description on the dataset page: https://huggingface.co/datasets/math-ai/AutoMathText-2.5.