22M synthetic reasoning traces
Synthetic question-answer pairs with reasoning chains on general topics, beyond just math and code.
- Text generation
- 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
Synthetic question-answer pairs with reasoning chains on general topics, beyond just math and code.
2.7 million US wire service articles from 1878-1977 with geolocation, topics, and entities.
17.6K published math problems of varying difficulty for testing reasoning models.
A test suite from OpenCompass with 1 to 10 thousand text tasks to evaluate LLM retrieval and reasoning over large documents.
A collection of verifiable tasks for training computer agents using reinforcement learning.
Multiple-choice questions in Traditional Chinese across 66 school and professional subjects for evaluating LLMs.
Executable benchmark for testing complex multi-step agent reasoning when working with APIs and documents.
A 700+ million token corpus for training compact models in step-by-step task analysis and web search.
105k synthetic pairs for translating natural language questions into SQL queries.
A corpus of human and ChatGPT answers for comparison and text detection.
Preprocessed English dataset for language model alignment using DPO.
Large English corpus edited by the UltraX framework for language model pre-training.
50,000 ShareGPT-format dialogues for training language models in literary writing.
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.
2556 queries with evidence across 2 to 4 documents to test RAG retrieval and reasoning.
10k multi-turn traces on agentic programming and reasoning from the Fable 5.1 model.
Large English web corpus from Common Crawl, filtered L1 layer for training language models.
Demographic and psychological data from a representative sample of people for character simulation using LLMs.
About 2 million titles, abstracts, and metadata of scientific papers from arXiv in English.
English question-sentence pair corpus for open-domain question answering research.
The Python portion of CodeSearchNet with open-source function code and brief explanations of their operation.
Massive text corpus of easy and extremely difficult sudoku puzzles for training and testing model logic.
A set of questions and URLs from OpenAI to test the ability of LLMs to find relevant information among noise in a long context.