MINT-1T: A Multimodal Dataset with One Trillion Text Tokens and 3.4 Billion Images

MINT-1T is an open-source multimodal interleaved dataset containing one trillion text tokens and 3.4 billion images, representing a 10x scale-up from prior open-source collections. It was created by a team from the University of Washington to facilitate research in multimodal pretraining. The dataset was last updated on the platform in September 2024.

Use Cases

• Pretraining multimodal large language models based on interleaved text and image sequences.
• Research on scaling laws for multimodal AI based on the dataset's trillion-token scale.
• Benchmarking model performance on data from PDF and academic paper sources mentioned in the description.

Strengths

• Contains one trillion text tokens and 3.4 billion images, a 10x scale-up from existing open-source datasets.
• Includes previously untapped sources such as PDFs and ArXiv papers.
• Explicitly designed for multimodal pretraining research.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license details are unknown, which may limit suitability assessment.

Provenance

Source

University of Washington team, via the mlfoundations organization on Hugging Face.

Collection Method

Scaling open-source multimodal data by aggregating and interleaving text and images from various sources, including PDFs and ArXiv.

Time Range

Likely includes content up to 2023, based on the title.

Freshness

Last updated 2024-09-19 21:06:59; freshness should be verified.

Geography

Region tag indicates 'us', but specific spatial coverage is unknown.