Jul 25, 03:38 AM

MINT-1T Open-Source Dataset Launches: One Trillion Tokens and 3.4 Billion Images from Diverse Sources

The MINT-1T dataset, the first open-source multimodal dataset containing one trillion tokens, has been released. This dataset includes 3.4 billion images sourced from diverse materials such as HTML, PDFs, and ArXiv papers. Developed in collaboration with SFResearch, the dataset aims to enhance training for multimodal AI models and is available on platforms like Hugging Face. The release has been positively received within the AI community, with researchers expressing enthusiasm for its potential applications in advancing multimodal reasoning and generation. In the first 48 hours post-launch, downloads of the MINT-1T dataset have been robust, reflecting significant interest from the research community.

#HTML #ArXiv #SFResearch #Hugging Face

Written with ChatGPT (GPT-4o mini).