Offline copies of 153 websites for GUI agent training
An environment of 153 real websites in Docker and 3,518 tasks with automated verification for training and evaluating web agents.
- Image text → text
- English
Hugging Face datasets with plain-language descriptions: what's inside and what models to train on them. 4,788 in the catalog so far, with new ones added every hour.
Answered by Gemini 4 · the chat is kept only in this tab
An environment of 153 real websites in Docker and 3,518 tasks with automated verification for training and evaluating web agents.
10k to 100k digitized multi-page National Archives documents in English.
Large DataComp-1B image-text corpus with improved captions from LLaVA-1.5-LLaMA3-8B.
A benchmark evaluating video models on real-time video question answering (CVPR 2025).
An English-language dataset based on 500 thousand arXiv research articles for training and evaluating document processing.
Instructional dataset for educational tasks and multimodal learning.
A slice of the ultra-large MINT-1T dataset of 1M to 10M records with interleaved English text and images.
Multi-domain benchmark for geospatial intelligence and computer vision.
Benchmark for computer vision research and defect inspection applications.
Question-answer pairs on pathology images for training and evaluating medical VQA systems.
Over 1.2M action steps, screenshots, and reasoning chains for training AI agents in Windows.
An ECCV 2026 test dataset for evaluating clear text restoration in video and VQA.
An English video-text instruction dataset for training models to understand long videos.
14,935 unique CAD tasks with images and three expert answers each.
Over 10 thousand videos and 78 thousand QA pairs for training robots to detect and fix errors.
Large-scale benchmark for comparative summarization of chart pairs by language models.
Complex Olympiad-level problems with images in English and Chinese for evaluating model scientific reasoning.
Metadata and scripts for downloading bird images for visual question-answering tasks.
Document image set with physical distortions to test OCR and parsing system robustness.
Multimodal dataset of 159,549 question-answer pairs for benchmarking in gastrointestinal endoscopy.
Multimodal evaluation dataset and MME benchmark from the MixLoRA paper (ACL 2024).
A large-scale open corpus of interleaved English text and images for multimodal pre-training.
Part of the giant open MINT-1T dataset featuring interleaved English text and PDF images.
Tens of millions of text image and transcript pairs for training Telugu optical character recognition systems.