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

DatasetGryphe/ChatGPT-4o-Writing-Prompts

English short stories by ChatGPT-4o

3,746 cleaned stories based on Reddit Writing Prompts, generated by ChatGPT-4o, with titles and genre tags.

  • Text generation
  • English
4.6K 381K–10K rows
DatasetIlyaGusev/gazeta

Russian news with brief summaries

A corpus of articles from the Gazeta.Ru website and their annotations for training and evaluating summarization models.

  • Summarization
  • Russian
4.6K 3110K–100K rows
Datasetpixparse/docvqa-single-page-questions

Single-page document questions and answers (DocVQA)

A classic English benchmark of 50,000 questions on document images for evaluating Document AI.

  • Question answering
  • English
4.6K 1210K–100K rows
Datasetlavita/medical-qa-datasets

Medical question-answer pairs for chatbot training

A collection of English medical question-answer pairs for SFT and verifying the truthfulness of answers.

  • Question answering
  • English
4.6K 681M–10M rows
DatasetM-A-D/Mixed-Arabic-Datasets-Repo

Large Arabic Text Corpus

A large-scale collection of diverse Arabic texts from various sources for a wide range of NLP tasks.

  • Text classification
  • Question answering
  • Translation
  • Arabic
4.5K 39100M–1B rows
DatasetBAAI/CCI3-HQ

CCI 3.0 HQ Chinese high-quality web corpus

A large cleaned Chinese text corpus from trusted internet sources for pre-training and fine-tuning language models.

  • Text generation
  • Chinese
4.5K 7010M–100M rows🔒 on request
Datasethoskinson-center/proof-pile

Proof-Pile math text corpus

High-quality English math texts for language model training and fine-tuning.

  • Text generation
  • English
4.5K 70100K–1M rows
Datasetclips/mqaTop pick

Massive multilingual web QA corpus

Hundreds of millions to a billion question-answer pairs in 6 languages, including Russian, collected from Common Crawl.

  • Question answering
  • Catalan
  • English
  • German
  • +9
4.5K 58100M–1B rows
Datasetnvidia/AceReason-MathTop pick

49k complex math problems by NVIDIA for RL

Cleaned corpus of 49,000 verifiable math problems with answers for training reasoning models via RL.

  • Text generation
  • English
4.3K 5610K–100K rows
Datasetnvidia/Nemotron-SFT-Agentic-Finance-v1Top pick

Multi-turn finance dialogues with tool calling

7,878 NVIDIA examples for training financial AI agents in tool use and SEC report analysis.

  • Text generation
  • English
4.3K 381K–10K rows
Datasetzai-org/LongAlign-10k

Long instructions for LLM alignment

10k long instructions of 8–64K tokens in en/zh for training and evaluating long-context LLMs.

  • Question answering
  • English
  • Chinese
4.2K 1001K–10K rows
Datasetinterstellarninja/hermes_reasoning_tool_use

Hermes: tool-calling training data

51,004 ShareGPT dialogues on when and how LLMs should call tools.

  • Question answering
  • English
4.2K 21010K–100K rows
DatasetUCSC-VLAA/MedTrinity-25MTop pick

MedTrinity-25M: medical images

25M+ medical images across 10 modalities with multimodal annotations for 65 diseases.

  • Question answering
  • English
4.2K 21610M–100M rows🔒 on request
Datasetai4privacy/pii-masking-300k

PII personal data masking

300k texts across 6 languages with PII annotations for training data removal models.

  • Text classification
  • Token classification
  • Table question answering
  • English
  • French
  • German
  • +3
4.2K 121100K–1M rows
DatasetRoman1111111/gemini-3.1-pro-hard-high-reasoning

Gemini 3.1 Pro synthetic reasoning chains

A compact 5.6M token corpus featuring complex multi-step English reasoning.

  • Question answering
  • Text generation
  • English
4.1K 631K–10K rows
DatasetOpenDataArena/MMFineReason-Full-2.3M-Qwen3-VL-235B-ThinkingTop pick

2.3m reasoning examples from Qwen3-VL-235B

Raw corpus of 2.3 million multimodal tasks and reasoning chains (8.8 billion tokens) before quality filtering.

  • Visual question answering
  • Question answering
  • Text generation
  • English
4K 651M–10M rows
Datasetdeclare-lab/HarmfulQA

LLM safety benchmark and dataset

A set of English prompts and responses generated via ChatGPT to evaluate model vulnerabilities and alignment.

  • Text generation
  • Text classification
  • English
4K 491K–10K rows
DatasetAdaMLLab/JpnMix

Cleaned Japanese corpus for LLM pre-training

A large Japanese corpus of 100M to 1B documents with quality filtering and MinHash deduplication.

  • Text generation
  • Japanese
4K 5100M–1B rows
Datasetnvidia/Nemotron-Safety-Guard-Dataset-v3Top pick

NVIDIA multilingual response safety dataset

Over half a million examples in 12 languages for training LLM safety filters (guard models).

  • Text classification
  • Text generation
  • English
  • Arabic
  • German
  • +9
4K 38100K–1M rows
Datasetkeivalya/MedQuad-MedicalQnADataset

Medical question-answering dataset

Medical question-answer pairs; the card references a 2019 paper on question-entailment.

  • Question answering
4K 13510K–100K rows
Datasetnvidia/Nemotron-Personas-Singapore

NVIDIA synthetic personas of Singapore residents

Hundreds of thousands of synthetic human profiles based on real Singapore demographics and psychology.

  • Text generation
  • English
4K 62100K–1M rows