datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
HPLT2.0_cleanedNB: HPLT2.0 is now superseded by a newer release:
HPLT3.0
We recommed switching to v3.0, unless you have a compelling reason to stay on 2.0.
This is a large-scale collection of web-crawled documents in 191 world languages, produced by the HPLT project.
The source of the data is mostly Internet Archive with some additions from Common Crawl.
For a detailed description of the dataset, please refer to our website and our pre-print.
The Cleaned variant of HPLT Datasets v2.0
This is… See the full description on the dataset page: https://huggingface.co/datasets/HPLT/HPLT2.0_cleaned.fineweb-2
🥂 FineWeb2
A sparkling update with 1000s of languages
What is it?
This is the second iteration of the popular 🍷 FineWeb dataset, bringing high quality pretraining data to over 1000 🗣️ languages.
The 🥂 FineWeb2 dataset is fully reproducible, available under the permissive ODC-By 1.0 license and extensively validated through hundreds of ablation experiments.
In particular, on the set of 9 diverse languages we used to guide our processing decisions, 🥂… See the full description on the dataset page: https://huggingface.co/datasets/HuggingFaceFW/fineweb-2.AutoMathText-V2
🚀 AutoMathText-V2: A 2.46 Trillion Token AI-Curated STEM Pretraining Dataset
🎉 AutoMathText-v2 has surpassed 1.5 million downloads! We'd love to know how you're using it. Please take 1 minute to fill out our use case survey. Your feedback will directly shape the future roadmap of this dataset.👉 Share your use case here
📊 AutoMathText-V2 consists of 2.46 trillion tokens of high-quality, deduplicated text spanning web content, mathematics, code, reasoning, and… See the full description on the dataset page: https://huggingface.co/datasets/OpenSQZ/AutoMathText-V2.FineWeb2-HQ
FineWeb2-HQ
Dataset summary
FineWeb2-HQ is a high-quality, model-filtered pretraining dataset derived as a subset of FineWeb2, spanning 20 languages. It enables around 6x faster pretraining compared to the base dataset. FineWeb2-HQ was created by selecting the top 10% quality documents of FineWeb2 in each language, based on scores assigned by a deep learning classifier trained to identify structured and knowledge-rich samples using XLM-RoBERTa embeddings.
Validation… See the full description on the dataset page: https://huggingface.co/datasets/epfml/FineWeb2-HQ.TxT360-v2
TxT360-v2
Dataset Description
Pre-training sources for the K2 Horizon training data release. This repository is part of the K2 Horizon collection.
The repository is organized into multiple subsets. Every subset has a train split backed by Parquet shards.
K2 Horizon Dataset Series
Dataset repository
Focus
Subsets
IFM/TxT360-v2
Web and question-answering text
3
IFM/Code-Reasoning
Code reasoning and task synthesis
7
IFM/Math-Reasoning… See the full description on the dataset page: https://huggingface.co/datasets/IFM/TxT360-v2.FineWeb2-embedded
FineWeb2-embedded
Dataset summary
FineWeb2-embedded is an extension of the FineWeb2 dataset, annotated with document-level XLM-RoBERTa embeddings for 20 languages, making the dataset useful for a variety of tasks, including document clustering, filtering, and other multilingual research.
Since XLM-RoBERTa has a sequence length limit of 512 tokens, each document's embeddings are obtained by mean-pooling 512 token chunks of the XLM-RoBERTa output. Therefore, longer texts… See the full description on the dataset page: https://huggingface.co/datasets/epfml/FineWeb2-embedded.fineweb-edu-score-2
📚 FineWeb-Edu-score-2
1.3 trillion tokens of the finest educational data the 🌐 web has to offer
What is it?
📚 FineWeb-Edu dataset consists of 1.3T tokens (FineWeb-Edu) and 5.4T tokens of educational web pages filtered from 🍷 FineWeb dataset. This is the 5.4 trillion version.
Note: this version uses a lower educational score threshold = 2, which results in more documents, but lower quality compared to the 1.3T version. For more details check the… See the full description on the dataset page: https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu-score-2.PersonaMem-v2
PersonaMem-v2: Towards Personalized Intelligence via Learning Implicit User Personas and Agentic Memory
📅 We have now released PersonaMem-v3!
🚨 The paper is now released. View the full paper here and codebase here.
Personalization is becoming the next milestone of artificial super-intelligence. AI cannot always satisfy every user, especially on tasks with subjective goals, but personalization offers a path toward pluralistic alignment.… See the full description on the dataset page: https://huggingface.co/datasets/bowen-upenn/PersonaMem-v2.swallow-code-v2
SwallowCode-v2
Resources
📑 arXiv: Read our paper for detailed methodology and results at arXiv:2505.02881.
🤗 Sister Dataset: Discover SwallowMath-v2, our companion dataset for mathematical reasoning.
💻 What is it?
SwallowCode-v1 was a high-quality Python code dataset generated through an LLM-based rewriting pipeline.
However, it had two significant limitations:
(1) it was distributed under the Llama 3.3 Community License, and
(2) its size was limited to… See the full description on the dataset page: https://huggingface.co/datasets/tokyotech-llm/swallow-code-v2.host-index-testing-v2
Common Crawl Host Index v2
GitHub: https://github.com/commoncrawl/cc-host-index
Each crawl, we generate a Host Index, which aggregates information about each web hosted visited during the crawl. The
information is aggregated from the Common Crawl columnar index,
web graph, and raw crawler logs.
Quickstart
The dataset is Hive-partitioned on crawl (data/crawl=CC-MAIN-2025-18/*.parquet). Open the whole
dataset once, then filter with WHERE crawl = '...': because… See the full description on the dataset page: https://huggingface.co/datasets/commoncrawl/host-index-testing-v2.stackv2_edu_filtered
Stack V2 Edu
Description
We filter the Stack V2 to only include code from openly licensed repositories, based on the license detection performed by the creators of Stack V2. When multiple licenses are detected in a single repository, we ensure that all of the licenses are on the Blue Oak Council certified license list. Per-document license information is available in the license entry of the metadata field of each example. Code for collecting, processing, and preparing… See the full description on the dataset page: https://huggingface.co/datasets/common-pile/stackv2_edu_filtered.24-game
Math Twenty Four (24s Game) Dataset
A comprehensive dataset for the classic math twenty four game (also known as the 4 numbers game / 24s game / Game of 24). This dataset of mathematical reasoning challenges was collected from 4nums.com, featuring over 1,300 unique puzzles of the Game of 24, with difficulty metrics derived from over 6.4 million human solution attempts since 2012.
In each puzzle, players must use exactly four numbers and basic arithmetic operations (+, -, ×, /) to… See the full description on the dataset page: https://huggingface.co/datasets/nlile/24-game.mm2_level
Mario Maker 2 levels
Part of the Mario Maker 2 Dataset Collection
Dataset Description
The Mario Maker 2 levels dataset consists of 26.6 million levels from Nintendo's online service totaling around 100GB of data. The dataset was created using the self-hosted Mario Maker 2 api over the course of 1 month in February 2022.
How to use it
The Mario Maker 2 levels dataset is a very large dataset so for most use cases it is recommended to make use of the streaming API… See the full description on the dataset page: https://huggingface.co/datasets/TheGreatRambler/mm2_level.MatrAIx_Persona_1M
MatrAIx Persona 1M
999,847 personas, each described by 1,290 categorical attributes.
599,847 are derived from real records, 400,000 are synthetic.
10 Zstandard Parquet shards, 4.17 GB.
Read it with pyarrow, not datasets
Attributes are packed: one persona's 1,290 attributes are 645 bytes of 4-bit
codes, low nibble first. datasets cannot open these files at all. Use pyarrow
and decode against persona_codes.schema.json.
import json, pyarrow.parquet as pq
schema =… See the full description on the dataset page: https://huggingface.co/datasets/MatrAIx2026/MatrAIx_Persona_1M.Qwen3.8-27B-Distillation-40K
Qwen3.8-27B-Distillation (40K Traces)
Qwen3.8-27B-Distillation is a dataset containing 40,000 reasoning traces distilled from Qwen's latest model — Qwen3.8-27B. We generated this dataset locally by running the model on our own infrastructure. It covers 4 domains with prompts sourced from 12 diverse open-source datasets.
Dataset Overview
Metric
Value
Total Examples
40,000
Teacher Model
Qwen3.8-27B
Model Precision
FP8
Reasoning Effort
medium… See the full description on the dataset page: https://huggingface.co/datasets/faunix/Qwen3.8-27B-Distillation-40K.chinese-fineweb-edu-v2
This version is deprecated. We recommend you to use the newest version Fineweb-edu-chinese-v2.1 !
Chinese Fineweb Edu Dataset V2 [中文] [English]
[OpenCSG Community] [👾github] [wechat] [Twitter]
📖Technical Report
Chinese Fineweb Edu Dataset V2 is a comprehensive upgrade of the original Chinese Fineweb Edu, designed and optimized for natural language processing (NLP) tasks in the education sector. This high-quality Chinese pretraining dataset has… See the full description on the dataset page: https://huggingface.co/datasets/opencsg/chinese-fineweb-edu-v2.dataset-the-stack-v2-dedup-sub
The Stack v2 Subset with File Contents (Python, Java, JavaScript, C, C++)
TempestTeam/dataset-the-stack-v2-dedup-sub
Dataset Summary
This dataset is a language-filtered and self-contained subset of bigcode/the-stack-v2-dedup, part
of the BigCode Project.
It contains only files written in the following programming languages:
Python 🐍
Java ☕
JavaScript 📜
C ⚙️
C++ ⚙️
Unlike the original dataset, which only includes metadata and Software Heritage IDs, this subset includes… See the full description on the dataset page: https://huggingface.co/datasets/Cyrile/dataset-the-stack-v2-dedup-sub.afdb-24M
AFDB-24M — AlphaFold Database Structures with Cluster-Based Splits
A curated subset of ~24 million AlphaFold Database (AFDB) v4 predicted protein structures, stored as sharded Parquet files. Each row contains the raw mmCIF structure text alongside metadata, precomputed cluster IDs, and leakage-resistant train/val/test split assignments.
Dataset Summary
Property
Value
Source
AlphaFold Database v4 (DeepMind / EMBL-EBI)
Total entries
~24,009,002
Shards
12,005… See the full description on the dataset page: https://huggingface.co/datasets/timodonnell/afdb-24M.docmath-eval-failures-200
DocMath-Eval Failures 200: Agent Benchmark & Leaderboard
A curated benchmark of 200 challenging financial math questions that leading AI models
failed to answer correctly, with comprehensive evaluation results from multiple AI agents.
Leaderboard
Evaluated on 2026-02-21 using LLM-as-Judge (Qwen QwQ-32B) for soft scoring.
Rank
Agent
Model
Exact Match
Judge: Exact
Judge: Approx
Judge: Total
Wrong
Avg Duration
Avg Tool Calls
1
TRAE Agent
Opus 4.5
98/200 (49.0%)
96… See the full description on the dataset page: https://huggingface.co/datasets/Ayushnangia/docmath-eval-failures-200.gigaverbo-v2
GigaVerbo-v2: A Large Portuguese Text Corpus with Educational and Toxicity Annotations
Dataset Summary
This repository contains a large corpus of Portuguese text (~320 billion tokens), which has been filtered and annotated using classifiers for educational content and toxicity. The dataset is intended for training language models and other NLP applications in Portuguese. GigaVerbo is part of the Polyglot project, which aims to develop multilingual resources and models… See the full description on the dataset page: https://huggingface.co/datasets/Polygl0t/gigaverbo-v2.agent-llm-traces-v2
Exgentic Agent LLM Traces v2 — Agent Chat Only
OpenTelemetry-shaped execution traces for 10,057 agent runs across 6 benchmarks (AppWorld, SWE-bench, BrowseCompPlus, τ²-bench Airline/Retail/Telecom), filtered to the agent under test's chat-only LLM calls. This is the dataset for replay testing, behavioral analysis, or any task where you care about what the benchmarked model actually did — not the eval scaffolding around it.
This v2 release expands upon Exgentic/agent-llm-traces… See the full description on the dataset page: https://huggingface.co/datasets/Exgentic/agent-llm-traces-v2.the-stack-v2-dedup
The Stack v2
The dataset consists of 4 versions:
bigcode/the-stack-v2: the full "The Stack v2" dataset
bigcode/the-stack-v2-dedup: based on the bigcode/the-stack-v2 but further near-deduplicated <-- you are here
bigcode/the-stack-v2-train-full-ids: based on the bigcode/the-stack-v2-dedup dataset but further filtered with heuristics and spanning 600+ programming languages. The data is grouped into repositories.
bigcode/the-stack-v2-train-smol-ids: based on the… See the full description on the dataset page: https://huggingface.co/datasets/bigcode/the-stack-v2-dedup.Magpie-Qwen2.5-Pro-1M-v0.1
Project Web: https://magpie-align.github.io/
Arxiv Technical Report: https://arxiv.org/abs/2406.08464
Codes: https://github.com/magpie-align/magpie
Abstract
Click Here
High-quality instruction data is critical for aligning large language models (LLMs). Although some models, such as Llama-3-Instruct, have open weights, their alignment data remain private, which hinders the democratization of AI. High human labor costs and a limited, predefined scope for prompting prevent… See the full description on the dataset page: https://huggingface.co/datasets/Magpie-Align/Magpie-Qwen2.5-Pro-1M-v0.1.swe-marathon
SWE Marathon: Ultra Long-Horizon Software Engineering Tasks
20 ultra long-horizon software-engineering tasks designed to challenge frontier coding agents. Each task ships with a containerized environment, a precise instruction, comprehensive tests, and a reference oracle solution. All tasks pass NOP-baseline / Oracle-fix validation.
Homepage: https://github.com/abundant-ai/swe-marathon
License: Apache 2.0
Format: Harbor task format (task.toml + instruction.md + environment/ +… See the full description on the dataset page: https://huggingface.co/datasets/rdesai2/swe-marathon.ID_REG_KG_2511
ID_REG_KG_2511: Indonesian Legal Regulation Knowledge Graph
ID_REG_KG_2511 is a high-quality, structured dataset specifically designed to represent Indonesian laws and regulations in a Knowledge Graph (KG) format. This dataset transforms flat legal text into a network of entities and relationships, enabling advanced Retrieval-Augmented Generation (RAG) and complex legal reasoning.
Dataset Description
The dataset focuses on the structural hierarchy and inter-connectivity… See the full description on the dataset page: https://huggingface.co/datasets/Azzindani/ID_REG_KG_2511.fineweb-2-dutchTheBlueScrubs-v2
The Blue Scrubs v2: A large refined medical dataset derived from the internet
Monique Shotande1,3, Luis Felipe1,3, Carlos Garcia1,3, Talia Kip Berler1,3, Mehmet Belgin1,3, Shane Corder1,3, Jarett DeAngelis1,3, Aakash Tripathi1,3, Issam El Naqa1,3, Vivek Rudrapatna2,3, Ghulam Rasool1,3, Gilmer Valdes1,3
Machine Learning Department, Moffitt Cancer Center, Tampa, Florida
Center for Real World Evidence, University of California San Francisco, San Francisco, California… See the full description on the dataset page: https://huggingface.co/datasets/TheBlueScrubs/TheBlueScrubs-v2.stack-v3-train
🥞 The Stack v3
What is it?
What is being released
How to download and use it
Dataset statistics
Dataset structure
Dataset creation
Considerations for using the data
Additional information
What is it?
The Stack v3 is the largest, most up-to-date open dataset of source code, crawled directly from GitHub and built to pre-train code LLMs with full-repository context. It is the successor to The Stack v2 and, like its predecessor, is released to make the training… See the full description on the dataset page: https://huggingface.co/datasets/mick260449/stack-v3-train.turkishfineweb2-cleaned
TurkishFineweb2-Cleaned
A Turkish web corpus derived from the Turkish (tur_Latn) subset of
FineWeb-2, augmented with an additional quality-classification layer
and a near-duplicate removal pass.
📄 Paper: MoganBert-TR: A Turkish Encoder Foundation Model Trained from Scratch with a CLM→MLM Curriculum
Source
FineWeb-2 is a
large-scale, multilingual web corpus built from Common Crawl. This dataset
covers the Turkish (tur_Latn) portion of FineWeb-2, spanning the… See the full description on the dataset page: https://huggingface.co/datasets/moganai/turkishfineweb2-cleaned.olm-CC-MAIN-2022-49-sampling-ratio-olm-0.15114822547
Dataset Card for OLM November/December 2022 Common Crawl
Cleaned and deduplicated pretraining dataset, created with the OLM repo here from 15% of the November/December 2022 Common Crawl snapshot.
Note: last_modified_timestamp was parsed from whatever a website returned in it's Last-Modified header; there are likely a small number of outliers that are incorrect, so we recommend removing the outliers before doing statistics with last_modified_timestamp.