Skip to content

Datasets

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.

Data type

For which models

Find a dataset for your goal

Answered by Gemini 4 · the chat is kept only in this tab

Datasetstanford-oval/ccnewsTop pick

Cleaned CommonCrawl world news archive

A massive multilingual corpus of 600 million news articles from 2016 to June 2024.

  • Text classification
  • Question answering
  • Text generation
  • Multilingual
  • Afrikaans
  • Amharic
  • +9
8.1K 36100M–1B rows
DatasetFiscalNote/billsum

US bill corpus with short summaries

10k to 100k texts of US Congress and California state bills with annotations for summarization tasks.

  • Summarization
  • English
8K 5710K–100K rows
Datasetccdv/arxiv-summarization

arXiv: scientific article summarization

arXiv papers and abstracts for training long-document summarization in English.

  • Summarization
  • Text generation
  • English
8K 141100K–1M rows
DatasetBAAI/IndustryCorpus

BAAI Chinese and English industry text corpus

A large cleaned text corpus of 100 million to 1 billion records for training industry LLMs.

  • Text generation
  • Chinese
  • English
7.7K 61100M–1B rows
Datasetvicgalle/alpaca-gpt4

52k Alpaca instructions with GPT-4 answers

52k English instruction-response pairs from GPT-4 based on Alpaca prompts for fine-tuning LLMs.

  • Text generation
  • Question answering
  • English
7.7K 33810K–100K rows
Datasetclue/clue

CLUE Chinese language understanding benchmark

Text and tabular tasks for training and evaluating Chinese language understanding models.

  • Text classification
  • Multiple choice
  • Chinese
7.6K 46100K–1M rows
Datasetallenai/wildjailbreak

Synthetic dataset for conversational model safety

262K synthetic prompt-response pairs: harmful and similar harmless prompts for safety training.

  • Text generation
  • English
7.5K 1751K–10K rows🔒 on request
Datasetargilla/distilabel-capybara-dpo-7k-binarized

7k-example Capybara DPO dataset

Binarized preferences for DPO based on Capybara: multi-turn dialogues for training chat models.

  • Question answering
  • Text generation
  • English
7.5K 1931K–10K rows
Datasetmhardalov/exams

EXAMS multilingual school exam questions

A benchmark of over 24,000 school exam questions across 16 languages for model evaluation.

  • Question answering
  • Arabic
  • Bulgarian
  • German
  • +9
7.4K 40100K–1M rows
DatasetHuggingFaceH4/Multilingual-Thinking

Multilingual chains of thought

1K reasoning examples in Spanish, French, Italian, and German translated from English.

  • Text generation
  • English
  • German
  • French
  • +2
7.4K 1211K–10K rows
Datasetopenbmb/UltraData-Stack-v3-L2Top pick

Cleaned 240B-token source code corpus

Filtered code corpus of ~240B tokens across 11 programming languages from The Stack v3.

  • Text generation
  • English
  • Chinese
7.4K 37100M–1B rows
DatasetAlienKevin/SWE-ZERO-12M-trajectoriesTop pick

SWE-ZERO agentic coding trajectories

112 billion tokens of agentic coding trajectories across 122k pull requests and 16 programming languages.

  • Text generation
  • English
7.4K 13010M–100M rows
Datasetnvidia/Nemotron-Personas-Korea

Nemotron Korean synthetic personas

A large set of synthetic personas based on real South Korean demographic distributions.

  • Text generation
  • Korean
7.3K 5941M–10M rows
Datasetisaacus/open-australian-legal-corpusTop pick

Open Australian legal corpus

229,122 Australian legal documents in English: legislation and judicial acts.

  • Text generation
  • Fill mask
  • Text retrieval
  • English
7.3K 109100K–1M rows
Datasetcommon-pile/caselaw_access_projectTop pick

US court decisions: 6.7M cases

A corpus of 6.7M US court cases for training and analyzing legal texts.

  • Text generation
  • English
7.3K 2291M–10M rows
Datasetwmt/wmt19

WMT19 parallel texts for machine translation

A large six-language multilingual corpus for training machine translation systems, containing 100M to 1B rows.

  • Translation
  • Czech
  • German
  • English
  • +7
7.2K 51100M–1B rows
Datasetjmhessel/newyorker_caption_contest

New Yorker humor understanding benchmark

Cartoons and captions from the New Yorker contest for evaluating humor understanding in models.

  • Image → text
  • Multiple choice
  • Text classification
  • English
7.2K 78100K–1M rows
DatasetGryphe/Opus-WritingPrompts

Claude Opus writing prompts stories

3,008 short stories generated by Claude Opus from Reddit prompts, featuring titles and genres.

  • Text generation
  • English
7.2K 871K–10K rows
Datasetfacebook/kilt_tasks

KILT benchmark on Wikipedia

11 datasets for fact-checking, QA, entity linking, and dialogue on a single Wikipedia dump.

  • Fill mask
  • Question answering
  • Text classification
  • English
7.1K 691M–10M rows
Datasetmlfoundations/MINT-1T-ArXivTop pick

MINT-1T ArXiv scientific paper multimodal corpus

English interlaced text and image dataset from ArXiv scientific papers for pre-training.

  • Image → text
  • Text generation
  • English
7K 611M–10M rows
Datasetxlangai/DS-1000

DS-1000 code generation benchmark

A simplified format of the popular benchmark with programming tasks comprising up to 10,000 examples.

  • Text generation
  • Code
7K 391K–10K rows