Skip to content

Datasets

Hugging Face datasets with plain-language descriptions: what's inside and what models to train on them. 4,797 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

DatasetProCreations/Beam-Reasoning-TracesTop pick

Beam-1.0-501B mathematics and science reasoning traces

Verified reasoning chains from the large Beam-1.0 model with varying thinking depth levels for distillation.

  • Text generation
  • Question answering
  • English
757 5100K–1M rows
Datasetechel0nn1881/kimi-cyber-reasoning

997 cybersecurity CoT traces

997 reasoning traces across 13 cybersecurity disciplines and 4 systems engineering domains.

  • Text generation
  • English
752 124up to 1K rows
Datasetamphora/ResearchMath-14k

Research-level mathematical problems in English

14,056 complex mathematical problems from research papers and open lists for model training and evaluation.

  • Text generation
  • Question answering
  • English
741 5710K–100K rows
Datasetjoelniklaus/LEXam-hard

LEXam hard legal exam questions

A curated set of 518 challenging open-ended law exam questions from the University of Zurich for testing legal AI agents.

  • Text generation
  • Question answering
  • English
  • German
739 32up to 1K rows
DatasetKeremTT/FABLE-5.2-MAX-DATABASE

Balanced English and Turkish chat dataset

240,000 dialogue examples across 12 domains in English and Turkish for LLM training.

  • Text generation
  • Question answering
  • English
  • Turkish
735 17100K–1M rows
Datasetmicrosoft/OpenMementos

228k reasoning traces with memento

228,557 reasoning traces broken into blocks with compressed summaries for context management.

  • Text generation
  • English
732 67100K–1M rows
Datasetmlabonne/orpo-dpo-mix-40k

Preference mixture for ORPO and DPO

40k examples from several DPO datasets for preference alignment training.

  • Text generation
  • English
729 31810K–100K rows
Datasetmissvector/linux-commands

Linux commands from natural language descriptions

Synthetic description-to-shell command pairs in 10 languages for training LLMs on Linux administration.

  • Text generation
  • Russian
  • Arabic
  • English
  • +7
724 9710K–100K rows
Datasetmicrosoft/NextCoderDataset

Synthetic instructions for code editing

Around 381 thousand English examples for training code-editing models across eight programming languages.

  • Text generation
  • English
721 56100K–1M rows
Datasetm-a-p/MusicPileTop pick

First corpus for pretraining music LLMs

English corpus of 5.17 million texts (4.16 billion tokens) about music, featuring ABC notation sheet music and books.

  • Text generation
  • English
715 601M–10M rows
Datasetlmg-anon/vntl-leaderboard

Japanese visual novel translation LLM leaderboard

Evaluation results of Japanese-to-English visual novel translation quality for various language models.

  • Translation
  • English
  • Japanese
715 44up to 1K rows
DatasetRUCKBReasoning/TableLLM-SFT

Table-based SFT training data

English training dataset for fine-tuning language models on table question answering.

  • Table question answering
  • Text generation
  • English
715 30
Datasetakoksal/LongForm

LongForm instructions for long text generation

English instruction dataset reconstructed from real texts in Wikipedia, C4, and Stack Exchange.

  • Table question answering
  • Summarization
  • Text generation
  • English
713 6110K–100K rows
DatasetXYZAILab/XYZ-Aquila-SFT

EN/ZH search agent trajectories

7k multi-turn agent trajectories with search tools in English and Chinese.

  • Text generation
  • Question answering
  • English
  • Chinese
711 3791K–10K rows
DatasetMineAITechnology/mine-20B-TokensTop pick

20B-token corpus for MineAI pre-training

Cleaned multilingual corpus (English, Urdu, Arabic, Hindi, code, and math) in parquet format totaling 19.34 billion tokens.

  • Text generation
  • English
  • Urdu
  • Arabic
  • +1
711 710M–100M rows
DatasetEverMind-AI/EvoAgentBench

AI agent self-evolution benchmark on past experience

A set of tasks for evaluating the ability of AI agents to evolve and reuse accumulated experience.

  • Text generation
  • Question answering
  • English
707 20up to 1K 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
DatasetGryphe/Sonnet3.5-Charcard-Roleplay

Character card roleplay dialogues

9,736 simulated English roleplay dialogues in ShareGPT format, heavily NSFW.

  • Text generation
  • English
688 1011K–10K 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
Datasetbeyoru/kimi-k3-distillation

Kimi K3 distillation traces for SFT and tool use

A collection of responses and tool calls from the kimi-code/k3 model for SFT fine-tuning.

  • Text generation
  • English
  • Chinese
685 20100K–1M rows
Datasetbitext/Bitext-retail-ecommerce-llm-chatbot-training-dataset

Synthetic dataset for retail chatbots

10,000 to 100,000 labeled English examples for fine-tuning e-commerce virtual assistants.

  • Question answering
  • Table question answering
  • English
683 2210K–100K rows
Datasetlatam-gpt/LatamGPT-Corpus-1.0-research

Research text corpus for LatamGPT

Restricted corpus of 10 to 100 million records in Portuguese, Spanish, and English.

  • Text generation
  • Portuguese
  • Spanish
  • English
682 810M–100M rows🔒 on request
Datasetatalaydenknalbant/video-games-dataset

Video games database from the RAWG portal

Structured descriptions, ratings, tags, and video game metadata from the RAWG API for analytics and recommendation systems.

  • Sentence similarity
  • Summarization
  • Embeddings
679 32100K–1M rows
DatasetOrphanage/Baidu_Tieba_SunXiaochuan

Baidu Tieba post corpus for SFT

Text data from the Chinese Baidu Tieba community for style fine-tuning of language models.

  • Text generation
  • Question answering
  • Chinese
676 1310K–100K rows