Nemotron VLM dataset v2
Nearly triple the high-quality samples of v1 for training VLM on image and video QA.
- Visual question answering
- Image text → text
- Video text → text
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
Nearly triple the high-quality samples of v1 for training VLM on image and video QA.
A massive video corpus with text descriptions and tabular metadata from the decentralized OMEGA Labs network.
Large-scale English image-text dataset for unified multimodal model training.
Visual math problem benchmark for multimodal LLMs.
7,041 dermatology cases with medical descriptions and Chain-of-Thought reasoning.
A large Physical AI dataset with 3D room scans and egocentric recordings in a unified metric system.
Test suite from Moonshot AI for evaluating basic visual perception in multimodal models.
Multi-phase abdominal CT scans with lab panels, reports, organ masks, and tumor masks.
Large-scale corpus of image-text pairs where caption phrases are grounded to corresponding image regions.
A collection of generated images and question-answer pairs based on executable code from AllenAI.
A large collection of image-centric data for training and evaluating vision-language models; subsets in JSONL with descriptions.
Cartoons and captions from the New Yorker contest for evaluating humor understanding in models.
A set of 16,288 tests across 2,937 pages for evaluating the accuracy of text, table, and layout recognition.
Preference score data and text feedback for aligning models across text, images, audio, and tables.
English interlaced text and image dataset from ArXiv scientific papers for pre-training.
A 700+ million token corpus for training compact models in step-by-step task analysis and web search.
Multimodal COCO-based dataset with images and English texts for evaluating and training models on fine details.
A synthetic dataset of question-answer pairs based on generated images of documents, tables, and charts.
100K to 1M English image-text question-answer pairs for VLM alignment training.
Short text snippets from English video tutorials with timestamps for search and RAG.
A large corpus of alt-text and image pairs from HTML for training multimodal models.
The first benchmark for evaluating recognition and parsing of real digital and photographed documents across 6 languages.
Question-answer pairs for MedPix medical images for VQA training and testing.
The first major open dataset for Ukrainian handwritten text recognition spanning over a century.