datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
wikireading
Dataset Card for Wikireading
This is a dataset of book chapters scraped from a Russian website called Wikireading.
Dataset Details
Dataset Description
Wikireading is a collection of non-fiction educational books in various domains: Biology, Art, History, Religion and much more. The books are highly educational and provide vast knowledge in different domains, making this dataset a good choice for pretraining.
The resulting dataset contains ~26M rows, which in… See the full description on the dataset page: https://huggingface.co/datasets/its5Q/wikireading.IT_Support_V2
Mack: IT Support & Admin Dataset
📋 Dataset Description
This dataset consists of 100,000+ conversation logs focused on IT Support and IT Administration tasks. It was generated to fine-tune the "Mack" model—an AI persona designed to act as an expert Tier 1 & Tier 2 IT Helpdesk agent.
The data covers a wide range of technical domains, including Windows troubleshooting, SQL Server administration, driver issues, network diagnostics, and hardware debugging.
Curated by: [Dev… See the full description on the dataset page: https://huggingface.co/datasets/benjaminmacklin/IT_Support_V2.proper-agents-data
ProPer Agents — data
Data for ProPer Agents: Proactivity Driven Personalized Agents for Advancing
Knowledge Gap Navigation (ACL 2026).
Paper ·
Adapters
Three domains: code, medical, pwab (product recommendation).
Layout
{domain}/
raw/train.jsonl source examples
raw/test.jsonl
raw/{domain}_rga_{train,test}.jsonl RGA SFT data (Alpaca format)
raw/{domain}_dga_{train,test}.jsonl DGA SFT data (Alpaca format)… See the full description on the dataset page: https://huggingface.co/datasets/itsgupta/proper-agents-data.bluesky-10m-posts-15-languages
Dataset Card: Bluesky 10M Multilingual
📊 Overview
Total Posts: 10,099,990
Languages: 15 (en, tr, es, pt, de, fr, ja, it, nl, pl, ru, ko, zh, ar, hi)
Collection Period: August 9-12, 2026
Source: Bluesky Jetstream API (public firehose)
Format: JSONL
Size: ~3 GB
🌍 Language Distribution
Language
Code
Posts
%
English
en
6,843,995
67.8%
Japanese
ja
1,547,179
15.3%
German
de
373,626
3.7%
Portuguese
pt
331,093
3.3%
Spanish
es
325,865… See the full description on the dataset page: https://huggingface.co/datasets/itsmebatuhan/bluesky-10m-posts-15-languages.panorama
Dataset Summary
Dataset of satirical news from "Panorama", Russian "The Onion".
Dataset Format
Dataset is in JSONLines format, where "title" is the article title, and "body" are contents of the article.
it-support-l1-ticket-classification
IT Support L1 Multilingual Dataset
Dataset Summary
IT Support L1 Multilingual Dataset is a synthetic enterprise help desk dataset for ticket classification and troubleshooting response generation. It contains realistic Level 1 IT support scenarios in English and Czech, designed for experiments in structured classification, response generation, and multilingual support workflow prototyping.
This dataset contains synthetic IT Support L1 scenarios. The records were generated… See the full description on the dataset page: https://huggingface.co/datasets/w1z4rd3k/it-support-l1-ticket-classification.LangChoiceBench
LangChoiceBench
Welcome to LangChoiceBench - the benchmark dataset for studying programming-language choice in reasoning LLMs.
Introduced in the paper LangChoiceBench: Measuring and Explaining Programming-Language Choice in LLMs, the dataset covers 28 real-world software projects across 7 domains (mobile, frontend, low-latency, systems, embedded, games, enterprise) where Python is a known poor default, plus a 4-project python_control area where Python is the right choice… See the full description on the dataset page: https://huggingface.co/datasets/itsluketwist/LangChoiceBench.MIKE-dataset
MIKE High-Signal Indian Legal Triage
This is the curated instruction-tuning corpus for MIKE, an India-focused legal
research and triage adapter. It contains 10,913 English examples designed for
source-bounded reasoning, issue triage, document and evidence planning,
structured output, and legacy/current criminal-law transition screening.
Dataset composition
10,049 balanced, completion-deduplicated base examples;
244 explicit JSON-schema instruction variants;
500… See the full description on the dataset page: https://huggingface.co/datasets/itsalloverig/MIKE-dataset.clean_ukrainian-news
Ukrainian News Dataset
This is a dataset of news articles downloaded from various Ukrainian websites and Telegram channels.
The dataset contains 22 567 099 JSON objects (news), total size ~67GB each with the following fields:
title: The title of the news article
text: The text of the news article, which may contain HTML tags(e.g., paragraphs, links, images, etc.)
url: The URL of the news article
datetime: The time of publication or when the article was parsed and added to… See the full description on the dataset page: https://huggingface.co/datasets/itsSHAS/clean_ukrainian-news.kevin-v1-dataset
Kevin V1 — NPC Conversation Dataset
Synthetic player↔NPC conversations for training game NPC dialogue models.
Generated with a 3-role pipeline (context / player / NPC) plus a judge that
verifies every NPC reply is grounded (no hallucinated facts) and
in-character.
Format
One conversation per row (JSON Lines). Each row:
{
"id": "conv_00042",
"area_id": "01_emberpeak_forge",
"npc": {"role": "blacksmith", "name": "...", "offers": [...], "knows_about": [...]}… See the full description on the dataset page: https://huggingface.co/datasets/ItsHotdogFred/kevin-v1-dataset.swe-clarify
SWE-Clarify-CFR: Game-Theoretic Clarification Dataset
SWE-Clarify-CFR is a high-fidelity synthetic dataset designed to train Large Language Models (LLMs) to detect dangerous ambiguity in software engineering tasks.
Standard LLMs suffer from "Helpfulness Bias", when presented with a vague request (e.g., "Flush the database"), they often guess the user's intent to be helpful. In high-stakes engineering, this can lead to catastrophic data loss or security breaches.
This dataset solves… See the full description on the dataset page: https://huggingface.co/datasets/itsankitkp/swe-clarify.teletype
Dataset Card for Teletype
This dataset is a scrape of all articles published on teletype, a popular platform for publishing articles, especially in Telegram. The dataset includes the original article HTML, as well as text extracted using the trafilatura library with favor_recall=True and other metadata provided by teletype.
Additionally, language identification was applied using the lingua-py library and the identification results are available in the lang column.
Curated by: its5Q
ConvES30K
ConvES10K-HQ-LLM — High-Quality Spanish Conversations (LLM-Generated)
Recommended for AI training. This version replaces the template-based builds and is fully LLM-generated for true independence.
Why this version
Previous builds (30K and 10K-template) were template-based (gen.py + 178 situations):
13,156 distinct messages from 74,526 total → 82.3% duplicate messages, max x109 on narrative blocks (Mesa seis pegada a la mesa siete...)
Coherence failures from… See the full description on the dataset page: https://huggingface.co/datasets/itsZyn/ConvES30K.OctoCodingBench
OctoCodingBench: Instruction-Following Benchmark for Coding Agents
English | 中文
🌟 Overview
OctoCodingBench benchmarks scaffold-aware instruction following in repository-grounded agentic coding.
Why OctoCodingBench?
Existing benchmarks (SWE-bench, etc.) focus on task completion — whether the agent produces correct code. However, they miss a critical dimension: does the agent follow the rules while solving the task?
In real-world agentic coding, agents must… See the full description on the dataset page: https://huggingface.co/datasets/itsPrerna202/OctoCodingBench.LibHalluBench
LibHalluBench - Library Hallucinations Benchmark
Welcome to LibHalluBench - the benchmark dataset for testing an LLMs propensity to use non-existent library names during code generation.
Using the prompts created in the paper Library Hallucinations in LLM-Generated Code: A Risk Analysis Grounded in Developer Queries, we have curated a dataset of code generation problems that have been observed to trigger a higher rate of hallucinations in LLMs.
📋 dataset | 💾 download | 🤖… See the full description on the dataset page: https://huggingface.co/datasets/itsluketwist/LibHalluBench.seeweb-llama-it-set