Fixed web document corpus for Deep-Research evaluation
About 100,000 selected web documents and complex queries for fair comparison of AI search agents.
- Question answering
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
About 100,000 selected web documents and complex queries for fair comparison of AI search agents.
Image and Chinese text dataset for evaluating MLLMs in multi-image attribute extraction.
Minecraft gameplay episode records with precomputed MineCLIP embeddings for fine-tuning STEVE-1-style agents.
350k verified reasoning chains in math, code, and science for training step-by-step reasoning.
Large English text corpus for training and fine-tuning language models.
570k human-written English sentence pairs with entailment, contradiction, and neutral labels.
Repackaged OpenVid-1M dataset into .tar shards for fast streaming during video model training.
Mirror of the well-known ExDark low-light photography dataset, reformatted for YOLO.
Parallel texts in 100 languages with English always serving as one side of the pair.
English programming instruction dataset for fine-tuning code generation with feedback.
English USMLE exam questions with four answer options for training and evaluating medical LLMs.
49K English and Chinese questions about cats and dogs with images, audio, video, and text.
A dataset for testing vision-language models (VLM) on evaluating plan and route quality.
A corpus of ML solutions, agent runs, and analytical artifacts for evaluating algorithm performance without execution.
Around 893,000 chapters from 96,000 Russian audiobooks with audio files and descriptions.
Canonical Level 2 trading records for Chinese A-shares in parquet format covering 2026.
ASR tests with audio and text, sorted by length and saved in parquet format.
Long-prompt collection (up to 100K tokens) with varying entropy for measuring language model inference speed and quality.
A set of 30 math competition problems from AIME I and AIME II 2025 for evaluating model reasoning.
Over 28k passages and nearly 100k multiple-choice questions from English exams in China.
Read Khmer speech based on Tipitaka texts for ASR and TTS tasks with manual access moderation.
Image and text sets for testing models' text recognition capabilities.
Over 1 trillion tokens of parallel text in English and 500+ languages for machine translation.
A challenging CVPR 2026 test set for evaluating text-to-image generation from multiple references.