Large cybersecurity instructional corpus
Over 1.8 million cleaned English dialogues in ShareGPT format for training LLMs on information security tasks.
- Text generation
- Question answering
- 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
Over 1.8 million cleaned English dialogues in ShareGPT format for training LLMs on information security tasks.
Synthetic text streams with rules and links for LLM pretraining.
Testing whether LLMs know what they do not know and can correct their mistakes.
A large English text dataset of 10M to 100M examples for grammatical error correction tasks.
1,973 English roleplay dialogues with chains of thought (CoT) written in character.
A corpus of 30.9 million files featuring developer discussions in GitHub issues and pull requests.
1,103 context-continuation pairs to evaluate LLMs' ability to continue dramatic scenarios.
Around 100 thousand complex multi-aspect English questions requiring deep search and analysis.
A combined and artifact-cleaned set of English prompts and answers for training instruct models.
Preference pairs (chosen/rejected) for training models to answer strictly based on the text and refuse to answer if data is missing.
A cleaned corpus of official public Indian documents in parquet format for search and question-answering systems.
CodeActInstruct data: multi-step tasks where the agent acts via executable Python code.
English prompt-response pairs for training an assistant to work with Home Assistant.
Millions of texts from the Italian official gazette Gazzetta Ufficiale for training language models.
GPT-4 characters and dialogues in Russian and English for training roleplay language models.
669 GB of cleaned and deduplicated Portuguese texts (over 139 billion tokens) for LLM training.
A subset of DISC-Med-SFT with Chinese text medical question-answer pairs for training and evaluating QA models.
Around 5,500 Turkish dialogue examples with long outputs for SFT.
English evaluation set for testing how language models respond to users' emotional requests.
Around 8k real meme dialogues from Douyin, Xiaohongshu, and Bilibili for conversational model training.
Long novels and hierarchical outlines for training models in coherent storytelling.
Benchmark for evaluating autonomous scientific paper generation systems using CVPR and ICLR materials.
52k English function calling examples with SYSTEM, USER, ASSISTANT, and JSON function calls.
1 million to 10 million examples for instruction fine-tuning of Turkish language models across various topics.