Massive synthetic corpus of mathematical reasoning
About 33 billion code-generated examples of step-by-step mathematical reasoning in English.
- Text generation
- English
Hugging Face datasets with plain-language descriptions: what's inside and what models to train on them. 4,797 in the catalog so far, with new ones added every hour.
Answered by Gemini 4 · the chat is kept only in this tab
About 33 billion code-generated examples of step-by-step mathematical reasoning in English.
15M English image-text pairs for training and evaluating open multimodal models.
An English and Chinese image-text dataset for deep evaluation of Text-to-Image models.
Data for training and evaluating safe responses: preferences, question-answer pairs, and English prompts.
Large filtered and deduplicated Chinese web text dataset for language model pre-training.
2,334 parametric 3D models and mechanisms with sand3d code, OpenUSD files, and renders for 3D generation.
Revised English test split of GSM8K for precise evaluation of LLM reasoning.
Commercial-friendly version of the GeneralThought-430K dataset without restricted prompts.
96k paragraph-based questions to test arithmetic, counting, and sorting.
14,000 English question-answer pairs about Wikidata entities with Wikipedia page view metrics.
22,637 four-choice questions across 44 languages to evaluate LLM knowledge and logic.
A multimodal benchmark of 18,000 samples to assist in CAD program generation and understanding.
Multilingual dataset of live dialogues of up to 4 people with separate audio channels for ASR.
A large corpus of text product reviews with ratings for text classification tasks.
1 to 10 million entries with spatial annotations for video and 3D generation tasks.
Filtered text dataset for improving LLM logical reasoning capabilities.
Audio questions across 26 locales to evaluate ASR, audio search, and audio embeddings.
A set of 500 malicious instructions in English to test language model robustness against adversarial attacks.
Code in many programming languages with permissive licenses after near-deduplication, about 1.5 TB.
29,980 synthetic examples for training precise adherence to instructions with constraints from IFEval.
A math-filtered version of OpenThoughts-114k with metadata and verifiable answers.
Over 9 million co-registered satellite imagery and geospatial data patches by IBM and ESA.
4.8M multimodal GPT-4-Vision video descriptions for LVLM and text-to-video models.
1.5 million synthetic trajectories working with 2,000+ tools from 495 real MCPs.