Programming tasks with synthetic test cases
Coding task dataset with automatically generated and filtered test cases for training models via RL.
- 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
Coding task dataset with automatically generated and filtered test cases for training models via RL.
Over 153k high-resolution aligned photos and tabular measurements of 8 characteristics across 4k wheat plots.
3.2M human-ChatGPT dialogues filtered for non-toxic user inputs and responses.
A collection of 8,500 full Arabic book texts totaling 1.1 billion tokens for language models.
From 100k to 1M generated language model responses to Infinity-Chat100 prompts in multiple languages.
2,771 English questions with options and logical forms to test language model reasoning.
Benchmark for evaluating spatio-temporal reasoning in LLMs and LRMs.
Giant biological dataset of 233 million photos across 933 thousand taxa for training BioCLIP 2.5-class models.
Compact English and Chinese text tasks from ByteDance to evaluate agents' ability to gather scattered facts.
Massive cleaned English text corpus (3 billion documents) in parquet format for LLM pretraining.
Recent text dataset of American Invitational Mathematics Examination 2026 problems for model logic testing.
A collection of 14M videos with text descriptions for training text-to-video diffusion models.
Large English audiobook speech and text corpus (10m to 100m records) optimized for streaming.
English chart-question pairs to evaluate whether VLMs actually analyze images instead of guessing.
A 10K–100K example test suite for checking whether an AI agent follows rules under user pressure.
A multimodal dataset of UMBRA satellite radar imagery, land cover data, biomes, and georeferencing.
English premise-hypothesis pairs from science school exams and web texts for NLI.
A multimodal English and Chinese dataset for scientific image generation and editing.
A massive cleaned text and tabular corpus (100M-1B records) featuring web text, science, math, and code.
Complete logs and traces of Qwen3.8-Flash-Next agent runs on CLI tasks.
An Amazon benchmark to test the robustness of anomaly detection models in real-world conditions.
An evaluation dataset for panoramic depth from eight open sources.
23.7k hours of spontaneous and read speech in 22 Indian languages, with 11.2k hours transcribed.
A text corpus from automotive subreddits for model training and analysis.