MINT-1T: Open-Source Multimodal Dataset with 1 Trillion Text Tokens

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

Use Cases

• Pretraining large multimodal models based on the interleaved text and image data.
• Researching scaling laws for multimodal AI using the 10x larger scale compared to prior open datasets.
• Analyzing the impact of previously untapped sources like PDFs and academic papers on model performance.

Strengths

• Contains 1 trillion text tokens, a specific scale metric.
• Includes 3.4 billion images.
• Explicitly designed as a 10x scale-up from existing open-source multimodal datasets.
• Incorporates previously untapped sources like PDFs and ArXiv papers.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and specific file formats are unknown, which may limit suitability assessment.
• Data may reflect source bias inherent to the aggregated open data.

Provenance

Source

University of Washington research team (mlfoundations).

Collection Method

Aggregated and scaled from multiple open data sources, including PDFs and ArXiv.

Freshness

Last updated 2024-09-21 01:50:16; freshness should be verified.