TxT360-v2: web and QA texts
Corpus of web and question-answering texts for model pre-training from the K2 Horizon release.
- Text generation
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
Corpus of web and question-answering texts for model pre-training from the K2 Horizon release.
A collection of competitive programming problems from various platforms for training and evaluating models.
Large corpus of English synthetic coding tasks with verifiable solutions and tests for SFT and RL.
A massive cleaned corpus on STEM, mathematics, code, and reasoning in English and Chinese.
Text tasks for SFT and RL with prediction tokens and model rewards.
A set of OSS-Instruct instructions generated by gpt-3.5-turbo-1106 for code generation tasks.
A quality-filtered subset of FineWeb2 across 20 languages for faster language model pretraining.
English and Chinese documents and tables for testing slide generation.
Collection of open scientific texts and metadata from arXiv, CVF, NeurIPS, and other venues.
A massive 9.3-billion-row tabular corpus with complete financial statements of fictional firms over 31 years.
A massive corpus of scientific paper summaries with OpenAlex metadata and full texts of 5.3 million articles.
Over 1.3M PII-free posts from 110 news, IT, and official Telegram channels for language model training.
162 English tasks for evaluating LLM legal reasoning across classification and question answering.
GitHub code with L0–L3 cleaning levels for training language models.
Cleaned bilingual (EN/ZH) corpus exceeding 2 trillion tokens on mathematics, code, and reasoning for LLM training.
Over 33 thousand open legal articles and publications in Traditional Chinese for fine-tuning language models.
Over 720,000 PII-clean posts spanning humor, prose, news, and tech.
Processed UltraFeedback: 64k prompts with model answers and GPT-4 scores aggregated into preference pairs.
Large raw archive of successful and failed web agent actions in English and Korean for subsequent filtering.
3,763 long captioned videos with questions to evaluate long-context multimodal models.
Chronological clinical events and 480 thousand medical images from 150 thousand PubMed Central cases.
Massive English web corpus with an educational filter for language model pre-training.
A subset of The Stack: 10K random code examples for each of 30 languages, 2.6 GB.
High-quality data selection for training small programming language models.