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

Datasetdbarbedillo/SMS_Spam_Multilingual_Collection_Dataset

Multilingual SMS spam collection dataset

A set of real and machine-translated SMS messages in multiple languages labeled as spam or ham.

  • Text classification
  • English
  • Chinese
  • Spanish
  • +9
383 151K–10K rows
DatasetAdaptLLM/medicine-tasks

Medical benchmark for evaluating fine-tuned LLMs

Collection of medical classification and question-answering tasks from an ICLR 2024 paper to test model knowledge.

  • Text classification
  • Question answering
  • Zero shot classification
  • English
381 331K–10K rows
DatasetFBK-MT/Speech-MASSIVE-test

Test split of the multilingual SLU dataset

Test split of a multilingual spoken corpus for speech recognition and intent understanding.

  • Audio classification
  • Text classification
  • Zero shot classification
  • Arabic
  • German
  • Spanish
  • +9
377 1110K–100K rows🔒 on request
Datasetgoogle/reveal

Google benchmark for chain-of-thought verification

English dataset for evaluating verifier models that check step-by-step LLM reasoning.

  • Text classification
  • Question answering
  • English
376 401K–10K rows🔒 on request
DatasetWhitzardAgent/CyberSecurity-1M

Cybersecurity knowledge base with 1.19M records

Information security dataset in English and Chinese for non-profit research.

  • Text generation
  • Summarization
  • Question answering
  • English
  • Chinese
367 431K–10K rows🔒 on request
Datasetjoelniklaus/brazilian_court_decisions

Annotated Brazilian court decisions

4,043 judicial decision summaries from the Brazilian state of Alagoas for verdict prediction.

  • Text classification
  • Portuguese
367 251K–10K rows
Datasetlamini/earnings-calls-qa

Financial earnings call transcripts and Q&A pairs

Transcripts of quarterly corporate earnings conference calls and English Q&A pairs based on them.

  • Text classification
  • Question answering
  • Text generation
  • English
348 55100K–1M rows
DatasetAdaptLLM/law-tasks

Legal tasks benchmark for evaluating LLMs

A collection of English legal test tasks for evaluating model domain adaptation.

  • Text classification
  • Question answering
  • Zero shot classification
  • English
337 371K–10K rows
Datasetalibayram/turkish_mmlu

Turkish MMLU for evaluating LLMs

Large Turkish-language dataset in MMLU format for testing and training language models.

  • Text generation
  • Text classification
  • Table question answering
  • Turkish
306 77100K–1M rows🔒 on request
DatasetLocutusque/UltraTextbooks-2.0

Large corpus of synthetic and human textbooks

A million-scale English text dataset with educational materials and code for training language models.

  • Text generation
  • Text classification
  • Fill mask
  • English
303 531M–10M rows
Datasetnyuuzyou/wb-feedbacksTop pick

Large corpus of Wildberries customer reviews

Between 100 million and 1 billion Russian customer reviews from the Wildberries marketplace in zstd archives with jsonl.

  • Text generation
  • Text classification
  • Russian
302 9100M–1B rows
Datasetinstruction-pretrain/ft-instruction-synthesizer-collection

Data for training context-based instruction generators

English corpus (100K–1M examples) accompanying the EMNLP 2024 paper for training an instruction and response synthesizer.

  • Text classification
  • Table question answering
  • Question answering
  • English
297 63100K–1M rows
Dataset5CD-AI/Vietnamese-THUIR-T2Ranking-gg-translated

Vietnamese passage ranking dataset

A large-scale Vietnamese corpus for information retrieval and text ranking tasks.

  • Text retrieval
  • Sentence similarity
  • Text classification
  • Vietnamese
  • Chinese
297 23100M–1B rows
Datasetparisaslanidis/eleni

Greek Parliament Proceedings 1862–1967

Digitized historical corpus of Greek parliamentary debates with links to primary sources.

  • Text classification
  • Text generation
  • Greek
295 141K–10K rows
Datasetming9999/maabenchTop pick

LLM text authorship attribution benchmark

English dataset for classifying generated texts across five LLM families with domain and version shift testing.

  • Text classification
  • English
294 7100K–1M rows🔒 on request
DatasetPaul/hatecheck-german

German hate speech detection test suite

A suite of functional tests for evaluating hate speech detection models in German.

  • Text classification
  • German
283 91K–10K rows
Datasetymoslem/Law-StackExchange

Legal questions and answers from Law StackExchange

An archive of all questions and answers from the Law section of StackExchange in English up to August 2023.

  • Question answering
  • Text classification
  • Sentence similarity
  • English
277 3210K–100K rows
DatasetSanatbek/aspect-based-sentiment-analysis-uzbek

Aspect-based sentiment analysis of reviews in Uzbek

Annotated restaurant reviews in Uzbek for aspect extraction and sentiment determination.

  • Token classification
  • Text classification
  • Uzbek
259 51K–10K rows
DatasetMindgard/evaded-prompt-injection-and-jailbreak-samples

Evasion Prompt Injection and Jailbreak Samples

An English-language collection of original and obfuscated prompt attacks for testing LLM security and filters.

  • Text classification
  • Text generation
  • English
250 3910K–100K rows🔒 on request
DatasetSAVSNET/PetEVALTop pick

Annotated Veterinary Medical Records from UK Clinics

The first benchmark based on 17,600 real veterinary records annotated with ICD-11 disease codes.

  • Token classification
  • Text classification
  • English
250 351K–10K rows🔒 on request
Datasetshivaniku/UniMoral

Multilingual moral dilemmas and reasoning dataset

1k to 10k annotated moral dilemmas in six languages, including Russian, for ethical reasoning analysis.

  • Text classification
  • Zero shot classification
  • Translation
  • Arabic
  • Chinese
  • English
  • +3
245 131K–10K rows🔒 on request