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

Datasetk-mktr/improved-flux-prompts-photoreal-portrait

FLUX.1 photorealistic portrait prompts

A collection of English prompts for generating realistic photo portraits using the FLUX.1 model.

  • Text classification
  • English
718 11610K–100K rows
DatasetFredZhang7/toxi-text-3M

Multilingual toxic comments dataset

Around 3 million real-user texts across 55 languages annotated for toxicity to train and evaluate moderation filters.

  • Text classification
  • Zero shot classification
  • Arabic
  • Spanish
  • Punjabi
  • +9
711 321M–10M rows
Datasetgsarti/magpie

Magpie idiom corpus

Around 37k annotated examples of potential idioms in figurative or literal contexts.

  • Text classification
  • Translation
  • English
704 810K–100K rows
Datasetairesearch/WangchanThaiInstruct

Culturally-aware Thai instruction dataset

Human-written Thai instruction dataset for training and evaluating models with local cultural awareness.

  • Text generation
  • Question answering
  • Summarization
  • Thai
686 3310K–100K rows
DatasetMaziyarPanahi/Jev-RAG-Claim-Verification-130K

Claim verification and hallucination detection in RAG

130 thousand examples of response source verification with annotation of hallucination types and errors.

  • Text classification
  • English
683 43100K–1M rows
Datasetyachay/text_coordinates_regions

Multilingual geo-tagged social media posts

A corpus of 500,000 multilingual texts linked to geographic coordinates across 123 regions.

  • Embeddings
  • Token classification
  • Text classification
  • English
  • Chinese
  • Spanish
  • +9
683 11100K–1M rows
Datasetnvidia/Nemotron-PII-Japan

Synthetic data for Japanese PII detection

NVIDIA dataset for training and evaluating Japanese personally identifiable information (PII) detection and masking models.

  • Token classification
  • Japanese
679 12100K–1M rows
Datasetlouisbrulenaudet/clinical-trials

Clinical trials database with embeddings

Records from ClinicalTrials.gov with metadata and pre-computed embeddings for biomedical tasks.

  • Question answering
  • Embeddings
  • Text classification
  • English
  • French
  • Spanish
675 42100K–1M rows
DatasetFreedomIntelligence/Huatuo26M-Lite

Chinese medical dataset Huatuo26M-Lite

Cleaned version of Huatuo26M for training Chinese medical models.

  • Text classification
  • Question answering
  • Text generation
  • Chinese
671 72100K–1M rows
DatasetOpenClaw/clawhub-security-signals

Security and trust signals for AI agent skills

Tabular-text security assessment dataset for public OpenClaw skills with scanner verdicts.

  • Text classification
  • English
666 5310K–100K rows
DatasetJr23xd23/ArabicText-LargeTop pick

Arabic corpus of 743k articles

743,288 Arabic articles, over 244 million words, a corpus for LLM training and fine-tuning.

  • Text generation
  • Fill mask
  • Text classification
  • Arabic
564 72100K–1M rows
Datasetudell-lab/NLP4LP

Linear programming tasks in natural language

A research corpus of under 1,000 optimization tasks with text descriptions, parameters, and difficulty tags.

  • Text classification
  • English
563 35up to 1K rows🔒 on request
Datasetopen-index/open-npmTop pick

Complete archive of npm registry package metadata

A massive snapshot of all Node.js packages from the npm registry with versions, dependencies, and download statistics.

  • Text classification
  • Embeddings
  • English
560 410M–100M rows
Datasetopen-index/hacker-news-rss

Catalog of RSS feeds for all Hacker News domains

623,957 discovered RSS/Atom feeds from 1.75M Hacker News websites in parquet format.

  • Text classification
  • Embeddings
  • English
536 4100K–1M rows
DatasetExploration-Lab/IL-TUR

Indian law benchmark across six languages

From 100K to 1M examples across 8 legal tasks in English and Indian languages.

  • Text classification
  • Summarization
  • Translation
  • English
  • Hindi
  • Bangla
  • +7
529 37100K–1M rows🔒 on request
DatasetAiresPucrs/stanford-encyclopedia-philosophy

Stanford encyclopedia of philosophy texts

An English text corpus with a volume from 100 thousand to 1 million entries for generation and classification tasks.

  • Text classification
  • Text generation
  • English
528 56100K–1M rows
Datasethyperbots-inc/synthetic-pii-redaction

Synthetic PII redaction documents dataset

Corporate document pages with sensitive data markup for tokenization and detection.

  • Token classification
  • Object detection
  • English
523 46up to 1K rows
Datasetturkish-nlp-suite/turkish-wikiNER

Turkish named entity recognition (NER) markup

20,000 sentences from Turkish Wikipedia with re-annotated named entities for training NER models.

  • Token classification
  • Turkish
512 1510K–100K rows
DatasetFredZhang7/all-scam-spam

Multilingual spam and scam email corpus

Over 42 thousand emails and messages in 43 languages with binary spam labels for training anti-spam filters.

  • Text classification
  • Zero shot classification
  • Norwegian
  • Spanish
  • Somali
  • +9
511 1610K–100K rows
DatasetAlicanKiraz0/Turkish-SFT-Dataset-v1.0

Turkish SFT dataset with long responses

Around 5,500 Turkish dialogue examples with long outputs for SFT.

  • Text classification
  • Question answering
  • Text generation
  • Turkish
466 531K–10K rows
DatasetCogensec/jailbreak-corpus

Jailbreak prompt testing corpus

Labeled corpus of real jailbreak attempts for building AI detectors and security guardrails.

  • Text classification
  • English
  • German
  • Spanish
  • +4
445 810K–100K rows🔒 on request
DatasetTimKoornstra/financial-tweets-sentiment

Financial tweets sentiment

Curated English posts about finance and the stock market for investor sentiment analysis.

  • Text classification
  • English
443 2510K–100K rows
Datasetanson-huang/mirage-news

AI fake news detection (text + image)

15,000 headline-image pairs for training and evaluating neural-generated news detectors.

  • Image classification
  • Text classification
  • English
442 1310K–100K rows