MINT-1T

A multimodal dataset comprising one trillion tokens and 3.4 billion images.

PremiumNewProductOpenSourceMultimodalDataset

MINT-1T is a multimodal dataset open-sourced by Salesforce AI, containing one trillion text tokens and 3.4 billion images, making it ten times larger than existing open-source datasets. It includes not only HTML documents but also PDF documents and ArXiv papers, enriching the dataset's diversity. The construction of MINT-1T involves multiple data collection, processing, and filtering steps to ensure high quality and diversity of the data.

Large scale: The dataset consists of one trillion tokens
ten times larger than existing datasets.
Diversity: Includes various document types such as HTML
PDF
and ArXiv papers.
High quality: Ensured through rigorous data filtering and deduplication processes.
Cross-modal reasoning: Capable of training large multimodal models for reasoning across images and text.
Broad domain coverage: Documents span multiple fields including science
technology
and humanities.

\The MINT-1T dataset is designed for researchers and developers in the field of artificial intelligence
particularly for training and research in multimodal learning and deep learning models. Its large scale and high-quality data provide a rich resource for training models
enhancing their performance in image and text processing tasks.\

The XGen-MM multimodal model pre-trained on MINT-1T performs exceptionally well in image captioning and visual question answering tasks.
On the Multidisciplinary Multimodal Understanding and Reasoning benchmark (MMMU)
MINT-1T shows significantly better performance in the science and technology domains compared to other datasets.
Under the Idefics2 architecture
MINT-1T exhibits outstanding performance in image captioning and visual question answering tasks.

1. Visit the open-source page of the MINT-1T dataset to learn about its basic information and features.
2. Download the dataset
selecting the appropriate subset based on your research or development needs.
3. Use the dataset for pre-training or fine-tuning your model to adapt to specific multimodal tasks.
4. Test your model's performance on tasks such as image captioning and visual question answering.

Visit

MINT-1T Visit Over Time

Monthly Visits

8724

Bounce Rate

53.42%

Page per Visit

1.4

Visit Duration

00:02:06

MINT-1T Visit Trend

MINT-1T Visit Geography

No Geography Data

MINT-1T Traffic Sources

MINT-1T Alternatives

Infini-Megrez — Multimodal understanding model for edge applications, enabling intelligent edge solutions through hardware-software collaboration.

Productivity•Artificial Intelligence•Deep Learning

Genie 2 — A large-scale foundational world model that generates diverse 3D operable environments.

Others•Artificial Intelligence•Deep Learning

294

Amazon Nova — Amazon Nova is Amazon's next-generation foundational model, offering cutting-edge intelligence and industry-leading cost-effectiveness.

InternationalSelection•AWS•Artificial Intelligence

138

Qwen2vl-Flux — An advanced multimodal image generation model that produces high-quality images by combining text prompts and visual references.

Image•Image Generation•Multimodal

318

LLaMA-Mesh — Unified 3D Mesh Generation with Language Models

Productivity•3D Modeling•Artificial Intelligence

DataChain — A modern Python data frame library designed specifically for artificial intelligence.

Productivity•Machine Learning•Artificial Intelligence

132

NVIDIA Video Search and Summarization — Build a video search and summarization agent to extract insights from videos.

Video•Video Analysis•Artificial Intelligence

180

Ultralight-Digital-Human — Super lightweight digital human model for real-time operation on mobile devices.

Image•Digital Human•Artificial Intelligence

426

Spirit LM — Multimodal language model that integrates text and speech

Productivity•Multimodal•Language Model

CogView3-Plus-3B — A text-to-image generation model that supports high-resolution image generation.

Image•Text-to-Image•Image Generation

360

F5-TTS — A high-quality text-to-speech synthesis model based on deep learning.

Productivity•text-to-speech•deep learning

1044

2233.ai — Instantly accessible AI conversational service.

Productivity•Artificial Intelligence•Conversational Service

408

UniMuMo — Unified model for text, music, and motion generation.

Music•Artificial Intelligence•Machine Learning

144

Yanchip Intelligence — Domestic large model supporting multimodal capabilities for quick and cost-effective digital transformation.

ChineseSelection•Artificial Intelligence•Large Models

354

Llama 3.2 3b Voice — Voice synthesis tool using the Llama model

Productivity•Speech Synthesis•Natural Language Processing

666

Udacity AI Academy — Offers AI and machine learning courses

InternationalSelection•Artificial Intelligence•Machine Learning

Intel Gaudi 3 AI Accelerator — High-performance AI accelerator designed for AI workloads.

Others•Artificial Intelligence•Accelerator

OpenMusic — Leveraging AI to create music

Music•Artificial Intelligence•Music Creation

618

DeepMind — A leading artificial intelligence research company under Google

324

West Lake AI Model — A multimodal model with high emotional and intellectual intelligence

ChineseSelection•Artificial Intelligence•Multimodal

282

GenWarp — Generate new viewpoint images while preserving semantic information.

Image•Image Generation•3D Reconstruction

402

Sapiens — An advanced AI visual model specifically designed to analyze and understand human motion.

Image•Artificial Intelligence•Visual Model

LLaVA-OneVision — An efficient model for multimodal vision task transformation.

Image•Multimodal•Visual Recognition

546

MedTrinity-25M — A large-scale multimodal medical dataset

Others•Medical Imaging•Multimodal

552

Best AI Websites & Tools