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

Datasetnvidia/OpenMathReasoningTop pick

OpenMathReasoning Mathematical Reasoning

306k tasks and millions of CoT/TIR solutions for training LLM mathematical reasoning.

  • Question answering
  • Text generation
  • English
91.6K 5051M–10M rows
DatasetMMMU/MMMU

MMMU multimodal benchmark for expert tasks

Benchmark for multimodal understanding and reasoning featuring questions with images across multiple disciplines.

  • Question answering
  • Visual question answering
  • Multiple choice
  • English
90.7K 34710K–100K rows
DatasetFineEnvs/SmolDataEnvs

Tabular tasks for small model RL

5.5K data analysis tasks: table, question, and ground truth answer.

  • Question answering
  • Table question answering
89.6K 1061K–10K rows
DatasetHuggingFaceH4/ultrachat_200k

Filtered dialogues for SFT chat models

200k filtered English dialogues used to train Zephyr-7B-β.

  • Text generation
  • English
87K 1.2K100K–1M rows
Datasethotpotqa/hotpot_qa

Multi-hop Wikipedia questions

113k Wikipedia question-answer pairs with supporting facts for multi-document answering.

  • Question answering
  • English
85.8K 365100K–1M rows
DatasetHuggingFaceFW/finepdfsTop pick

3 Trillion Token PDF Corpus

The largest open text corpus from PDFs: 3 trillion tokens, 475 million documents, and 1,733 languages.

  • Text generation
  • aai
  • aak
  • aau
  • +9
85.5K 962100M–1B rows
Datasetxuejun72/HR-VILAGE-3K3M

HR-VILAGE-3K3M immunization gene profiles

Tabular and text dataset of human gene expression profiles during respiratory viral immunization.

  • Zero shot classification
  • Embeddings
  • Token classification
84.7K 31K–10K rows
DatasetSWE-bench/SWE-smith

SWE-smith Tasks for Code Agents

50,137 tasks from 128 GitHub repositories for training code repair models.

  • Text generation
  • English
84.6K 7010K–100K rows
Datasettau/commonsense_qa

Commonsense Multiple-Choice QA

12,102 English questions with four answer choices for evaluating commonsense reasoning.

  • Question answering
  • English
83.6K 17010K–100K rows
Datasetallenai/dolma3_mix-6T-1025-7B

Training data mix for the Olmo 3 7B model

A multi-billion English text corpus from AllenAI designed to replicate Olmo 3 7B training.

  • Text generation
  • English
83.6K 581B–10B rows
Datasetnvidia/Nemotron-CC-v2Top pick

Nemotron-CC-v2 for Pre-training

Very large multilingual NVIDIA corpus for LLM pre-training featuring math, code, and Q&A.

  • Text generation
83.2K 1551B–10B rows🔒 on request
Datasetai4bharat/sangrahaTop pick

Sangraha: corpus of Indian languages

Largest cleaned corpus for pre-training across 22 Indian languages, 251 billion tokens.

  • Text generation
  • Assamese
  • Bangla
  • Gujarati
  • +9
81.5K 90100M–1B rows
Datasetabisee/cnn_dailymail

English news for summarization

Over 300k English CNN and Daily Mail news articles for extractive and abstractive summarization.

  • Summarization
  • English
80.7K 368100K–1M rows
Datasetm-a-p/MatrixTop pick

Bilingual corpus of 4.69 trillion tokens

English-Chinese open corpus for language model pretraining, listed as 4.69 trillion tokens in the card.

  • Text generation
  • English
  • Chinese
78.9K 1891B–10B rows
Datasetjordangong/the-stack-v2-smollm3

Materialized Source Code from The Stack v2

1 to 10 billion source code snippets across 16 programming languages for training code models.

  • Text generation
  • Code
78.8K 81B–10B rows
DatasetIFM/TxT360-v2

TxT360-v2: web and QA texts

Corpus of web and question-answering texts for model pre-training from the K2 Horizon release.

  • Text generation
76.8K 1091B–10B rows
Datasethails/mmlu_no_train

MMLU benchmark without training split

A lightweight copy of the English MMLU text benchmark for rapid model evaluation.

  • Question answering
  • English
65.8K 3910K–100K rows
Datasetmeaecosystem/pustakaTop pick

Large text corpus focused on the Indonesian language

Filtered multilingual web corpus ranging from 1 to 10 billion records for training language models.

  • Text generation
  • Indonesian
  • English
  • Japanese
  • +2
65K 131B–10B rows
Datasetedinburgh-dawg/mmlu-redux-2.0

Relabeled MMLU benchmark (5,700 questions)

A manually cleaned set of 5,700 questions across 57 English disciplines for precise LLM evaluation.

  • Question answering
  • English
63.9K 381K–10K rows
Datasetceval/ceval-exam

C-Eval Chinese examination benchmark

13,948 multiple-choice questions across 52 disciplines for evaluating Chinese foundation models.

  • Text classification
  • Multiple choice
  • Question answering
  • Chinese
63.5K 32410K–100K rows
Datasetdeepmind/code_contests

Competitive programming tasks

A collection of competitive programming problems from various platforms for training and evaluating models.

  • Translation
  • English
62.4K 2471K–10K rows
Datasetdatabricks/databricks-dolly-15k

Databricks Dolly 15k instructions

15k English instruction-following records across InstructGPT categories for model fine-tuning and evaluation.

  • Question answering
  • Summarization
  • English
62K 1.4K10K–100K rows