datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
VMemArena
VMemArena
Open-ended QA benchmark for long-video memory. 1000 questions over 561 videos
(680 h of footage). Every question is free-form: there are no options to choose from.
Files
file
contents
vmemarena.json
561 video entries, each with its questions
videos.jsonl
per-video metadata (duration, fps, resolution, codec, size)
videos/
561 .mp4 files, named by video_id
Schema
// vmemarena.json — a list of video entries
{
"video_id":… See the full description on the dataset page: https://huggingface.co/datasets/ShijianW01/VMemArena.MathCanvas-Instruct
MathCanvas-Instruct Dataset
🚀 Data Usage
from datasets import load_dataset
dataset = load_dataset("shiwk24/MathCanvas-Instruct")
print(dataset)
📖 Overview
MathCanvas-Instruct is a high-quality, fine-tuning dataset with 219K examples of interleaved visual-textual reasoning paths. It is the core component for the second phase of the [MathCanvas] framework: Strategic Visual-Aided Reasoning.… See the full description on the dataset page: https://huggingface.co/datasets/shiwk24/MathCanvas-Instruct.MathCanvas-Bench
MathCanvas-Bench
🚀 Data Usage
from datasets import load_dataset
dataset = load_dataset("shiwk24/MathCanvas-Bench")
print(dataset)
📖 Introduction
MathCanvas-Bench is a challenging new benchmark designed to evaluate the intrinsic Visual Chain-of-Thought (VCoT) capabilities of Large Multimodal Models (LMMs). It serves as the primary evaluation testbed for the [MathCanvas] framework.… See the full description on the dataset page: https://huggingface.co/datasets/shiwk24/MathCanvas-Bench.TCM-Instruction-Tuning-ShizhenGPT
📚 Introduction
This dataset is a fine-tuning dataset for ShizhenGPT, a multimodal LLM for Traditional Chinese Medicine (TCM). We open-source 245K multimodal Chinese medicine instruction data, including text instructions, visual instructions, and signal instructions for TCM.
For details, see our paper and GitHub repository.
📊 Dataset Overview
The open-sourced fine-tuning dataset consists of three parts:
Modality
Data Quantity
TCM Text Instructions
📝 Text… See the full description on the dataset page: https://huggingface.co/datasets/FreedomIntelligence/TCM-Instruction-Tuning-ShizhenGPT.Shifaa_Arabic_Medical_Consultations
Shifaa Arabic Medical Consultations 🏥📊
Overview 🌍
Shifaa is revolutionizing Arabic medical AI by addressing the critical gap in Arabic medical datasets. Our first contribution is the Shifaa Arabic Medical Consultations dataset, a comprehensive collection of 84,422 real-world medical consultations covering 16 Main Specializations and 585 Hierarchical Diagnoses.
🔍 Why is this dataset important?
First large-scale Arabic medical dataset for AI applications.… See the full description on the dataset page: https://huggingface.co/datasets/Ahmed-Selem/Shifaa_Arabic_Medical_Consultations.shiftedx-bench
Shiftedx Bench
Shiftedx Bench is a reproducible qualification suite for local language-model deployments. It measures model quality, effective long-context use, tool protocol reliability, multi-turn agent behavior, vision, and runtime performance without collapsing them into a single “intelligence” score.
The project is designed for quantization and speculative-decoding decisions on Apple Silicon, but its API runner works with any OpenAI-compatible chat endpoint.… See the full description on the dataset page: https://huggingface.co/datasets/Shiftedx/shiftedx-bench.Shifaa_Arabic_Mental_Health_Consultations
🏥 Shifaa Arabic Mental Health Consultations 🧠
📌 Overview
Shifaa Arabic Mental Health Consultations is a high-quality dataset designed to advance Arabic medical language models.This dataset provides 35,648 real-world medical consultations, covering a wide range of mental health concerns.
📊 Dataset Summary
Size: 35,648 consultations
Main Specializations: 7
Specific Diagnoses: 123
Languages: Arabic (العربية)
Why This Dataset?
🔹 Lack of… See the full description on the dataset page: https://huggingface.co/datasets/Ahmed-Selem/Shifaa_Arabic_Mental_Health_Consultations.ReportQA
ReportQA: QA-Based Radiology Report Evaluation
This is the official dataset for ReportQA, including:
free-form reports,
knowledge trees,
structured reports,
QAs.
Dataset Statistics
Dataset
Modality
Anatomy
Language
Free-form report
QA
QA (filtered)
CTRG-Brain-zh
CT (3D)
Brain
Chinese
2,001
209,551
182,136
CT-RATE
CT (3D)
Chest
English
3,039
412,737
298,654
AMOS-MM
CT (3D)
Abdomen
English
400
45,128
25,433
MIMIC-CXR
X-ray
Chest
English
1,417
201… See the full description on the dataset page: https://huggingface.co/datasets/shiym2000/ReportQA.TCM-Instruction-Tuning-ShizhenGPT
📚 Introduction
This dataset is a fine-tuning dataset for ShizhenGPT, a multimodal LLM for Traditional Chinese Medicine (TCM). We open-source 245K multimodal Chinese medicine instruction data, including text instructions, visual instructions, and signal instructions for TCM.
For details, see our paper and GitHub repository.
📊 Dataset Overview
The open-sourced fine-tuning dataset consists of three parts:
Modality
Data Quantity
TCM Text Instructions
📝 Text… See the full description on the dataset page: https://huggingface.co/datasets/CarsonnnNN/TCM-Instruction-Tuning-ShizhenGPT.Mental-Health-Conversations
Dataset Card
This dataset consists of around 99k rows of mental health conversations. It is a cleaned version of "jerryjalapeno/nart-100k-synthetic".
Source
jerryjalapeno/nart-100k-synthetic
personal-info-unlearning
Synthetic Personal Information Unlearning Dataset
Dataset Description
This dataset is designed for research on large language model (LLM) unlearning in controlled synthetic personal-information settings.
It contains synthetic profiles and question-answer data for four personal attributes:
Year of birth
Blood type
Postcode
Social insurance number
The benchmark provides three forget-set sizes: N = 5, 20, 40.
All personal-profile data are synthetically generated… See the full description on the dataset page: https://huggingface.co/datasets/shichenghu/personal-info-unlearning.for-the-small-shield-chapters
Foreword
The datasets contain information I extracted from the first draft and only draft of a novel called For The Small Shield, on github, written by me, Kalab J. Oster.
I used Claude's LLM to extract information from each chapter in order, creating a Graph mapping to improve the storytelling ability of a model fine-tuned with this dataset: wordsum/for-the-small-shield-instruct
I've tested the Graph data with my story bots with NousResearch/Hermes-2-Pro-Llama-3-8B fine-tuned… See the full description on the dataset page: https://huggingface.co/datasets/wordsum/for-the-small-shield-chapters.shimbabomb-ai-benchmark
ShimbaBomb AI Benchmark
A GSM8K-style benchmark dataset for evaluating AI models on ShimbaBomb — an English-like scripting language that compiles to native C.
Overview
This dataset contains 81 problems with chain-of-thought reasoning for training and evaluating AI models on ShimbaBomb code generation, understanding, debugging, and explanation.
Each problem has:
question: A natural language description of a programming task or question about SB code
answer:… See the full description on the dataset page: https://huggingface.co/datasets/shimbaaa/shimbabomb-ai-benchmark.MSU-Benchmark
MSU-Bench: Towards Speaker-Centric Understanding in Conversational Multi-Speaker Scenarios
Interspeech 2026 · ASLP@NPU (Northwestern Polytechnical University), in collaboration with Li Auto.
Zhaokai Sun*, Shuai Wang*, Zhennan Lin*, Chengyou Wang, Dehui Gao, Yuang Cao, Chunjiang He, Pan Zhou, Lei Xie**
Audio, Speech and Language Processing Group (ASLP@NPU), School of Software, Northwestern Polytechnical University, China
School of Intelligent Science and Technology, Nanjing… See the full description on the dataset page: https://huggingface.co/datasets/Khalilah-Shields/MSU-Benchmark.WebWalkerQA📑 The paper of WebWalkerQA is available at arXiv.
📊 The dataset resource is a collection of 680 questions and answers from the WebWebWalker dataset.
🙋 The dataset is in the form of a JSON file.
The keys in the JSON include:
Question, Answer, Root_Url, and Info. The Info field contains
more detailed information, including Hop, Domain, Language,
Difficulty_Level, Source Website, and Golden_Path.
{
"Question": "When is the paper submission deadline for the ACL 2025 Industry Track, and what… See the full description on the dataset page: https://huggingface.co/datasets/shitianqi/WebWalkerQA.nctb-qa
NCTBench: When the Textbook Becomes the Oracle
A Dual-Language Retrieval-Augmented QA Benchmark for Low-Resource Educational NLP
12,403 validated QA pairs| 2 languages | 4 grade levels | 15 RAG configurations benchmarked
Bengali · English · ICT · Science · Classes 6–9 · NCTB Bangladesh
Overview
NCTBench is the first large-scale, dual-language question-answering benchmark derived from the
National Curriculum and Textbook Board (NCTB) pilot textbooks of… See the full description on the dataset page: https://huggingface.co/datasets/ShihabReza/nctb-qa.longmemeval-cn
LongMemEval-CN 500题中文子集｜识流公开结果
这是识流维护的 LongMemEval 500题中文翻译子集与逐题评测结果。500题结果快照版本为 2026.07.15；本存档修订版为 2026.07.18，仅完善许可、引用与跨平台存档元数据，results.jsonl 未改变。
该子集状态为 draft，不是 LongMemEval 官方发布的中文版本。源自 LongMemEval 的问题与参考答案继续遵循上游 MIT License；由识流新增且有权授权的中文译文、模型输出整理、评测结果、汇总数据与原创说明采用 CC BY 4.0。完整许可边界见 NOTICE.txt。
结果摘要
测试日期：2026-07-15
总题数：500
跳过：0
首轮通过：499/500（99.8%）
独立复判：唯一未通过题复判通过
复核后有效通过：500/500
exact-match guard：298题
DeepSeek deepseek-v4-flash 判分：202题… See the full description on the dataset page: https://huggingface.co/datasets/shiliu-memory/longmemeval-cn.MedCOT-Reason
Important Note: I do not claim this dataset as my own. The entire credit belongs to the sources shared below. This dataset is simply preprocessed and formatted to align with the task of fine-tuning meta-llama/Llama-3.2-3B-Instruct
Introduction
This dataset is used to fine-tune ShivomH/Vitalis-Llama3-Reason, a smart medical LLM designed for advanced medical reasoning. This dataset is constructed using GPT-4o.
Sources
FreedomIntelligence/medical-o1-reasoning-SFT
View… See the full description on the dataset page: https://huggingface.co/datasets/ShivomH/MedCOT-Reason.ambig-iac-sample
Ambig-IaC Random 50
A random subset of 50 rows sampled without replacement from all 300 rows of
the default/train split of znyang/ambig-iac.
Source revision: 96429693e7164b024333e6d02f8c6b2d017e4ecb.
Sampling: Python random.Random(42).sample(range(300), 50).
Rows are stored in random draw order. All seven original columns, their types,
values, and original IDs are preserved. No filtering or text changes were made.
The split is named train and contains exactly 50 rows. The… See the full description on the dataset page: https://huggingface.co/datasets/shihanlin/ambig-iac-sample.shinto-wikidata-qa
Shinto Wikidata QA
Instruction/QA pairs about the Shinto domain — Shinto shrines, kami (deities, with
genealogy), and key texts (Engishiki, Kojiki, Nihon Shoki) — generated from Wikidata
structured facts.
Built for the Adaption Labs AutoScientist Challenge (All Other Domains track).
Credit: Adaptive Data by Adaption.
Source & license
Source: Wikidata Query Service (https://query.wikidata.org). All statement data is
CC0 / public domain, so this derived dataset is… See the full description on the dataset page: https://huggingface.co/datasets/EmmaLeonhart/shinto-wikidata-qa.shippinglaw
Shipping Law Q&A Dataset Sample
The Shipping Law Q&A Dataset is a curated collection of approximately 1500 question and answer pairs on various topics within shipping law (Using ChatGPT and Claude). Each entry is structured to facilitate training of language models (LLaMA Chat) for the legal domain, particularly within the maritime law context.
Data Structure
Entries in the dataset are presented as JSON objects, each containing a text field with instructional tokens… See the full description on the dataset page: https://huggingface.co/datasets/zsmail/shippinglaw.manta-1m-seqlen-513-1024
MANTA-1M: Calibration Subset (Seq Len 513-1024)
Overview
This dataset is a length-specific subset of the LGAI-EXAONE/MANTA-1M dataset, curated specifically for Post-Training Quantization (PTQ) Calibration.
Following the insights from the MaCa (Matryoshka Calibration) paper, this dataset provides length-specific calibration samples to ensure that the quantization process accounts for the variable weight importance across different input scales. By focusing on the… See the full description on the dataset page: https://huggingface.co/datasets/haesol-shin/manta-1m-seqlen-513-1024.clin-auth-bench
ClinAuthBench
ClinAuthBench is a synthetic inpatient health authorization benchmark. V1 focuses on adult inpatient psychiatric authorization over dense 72-hour chart packets.
Each record contains a synthetic multi-form chart packet and structured gold labels for continued-stay reasoning, lower-level-of-care readiness, evidence grounding, risk reconciliation, and unsupported-claim avoidance.
Links
GitHub (evaluation code, baselines, generators):… See the full description on the dataset page: https://huggingface.co/datasets/Shivi1982/clin-auth-bench.MentalHealth-Support
Important Note
This dataset is created from merging two datasets from different sources and has been formatted according to the "messages", "role", "content" chat format. I do not claim any ownership of this dataset.
Keep in mind that this dataset is entirely synthetic. It is not fully representative of real therapy situations. If you are training an LLM therapist keep in mind the limitations of LLMs and highlight those limitations to users in a responsible manner.
Since Mental… See the full description on the dataset page: https://huggingface.co/datasets/ShivomH/MentalHealth-Support.ameba_faq_search
AMEBA Blog FAQ Search Dataset
This data was obtained by crawling this website.
The FAQ Data was processed to remove HTML tags and other formatting after crawling, and entries containing excessively long content were excluded.
The Query Data was generated using a Large Language Model (LLM). Please refer to the following blog for information about the generation process.
https://www.ai-shift.co.jp/techblog/3710
https://www.ai-shift.co.jp/techblog/3761
Column description… See the full description on the dataset page: https://huggingface.co/datasets/ai-shift/ameba_faq_search.Text2Space
Text2Space
Synthetic dataset of 20,000 spatial reasoning instances. Each instance pairs a natural-language description of a 2D layout with three ASCII renderings of the same scene and a query about the relative position of two objects. Designed to train and evaluate language and vision-language models on spatial reasoning.
Companion dataset for the paper Learning to Draw ASCII Improves Spatial Reasoning in Language Models (arXiv:2604.14641).
Quick Look
{… See the full description on the dataset page: https://huggingface.co/datasets/ShiyuanHuang/Text2Space.interview-coach-dataset
Interview Coach Dataset
Chat-format dataset for fine-tuning an AI interview coach on software engineering interview Q&A.
Dataset Summary
Each example is a single user/assistant turn in OpenAI-style messages format, suitable for instruction / chat fine-tuning (e.g. Unsloth, TRL, Hugging Face SFTTrainer).
Train: ~1,017 examples
Validation: ~114 examples
Total: ~1,131 examples
Split: 90% / 10% (seeded shuffle)
Data Structure
{
"messages": [… See the full description on the dataset page: https://huggingface.co/datasets/shimogerald/interview-coach-dataset.test1
Dataset Card for ICLR Papers with Reviews (2023-2025)
Dataset Description
This dataset contains paper submissions and review data from the International Conference on Learning Representations (ICLR) for the years 2023, 2024, and 2025. The data is sourced from OpenReview, an open peer review platform that hosts the review process for top ML conferences.
Focus on Review Data
This dataset emphasizes the peer review ecosystem surrounding academic papers. Each… See the full description on the dataset page: https://huggingface.co/datasets/shiweichn/test1.pii-masking-200k
👉 Looking for the newest release? The current flagship is ai4privacy/pii-masking-openpii-1m. 1.4M samples, 23 languages, 19 PII classes.
Ai4Privacy Community
Join our community at https://discord.gg/FmzWshaaQT to help build open datasets for privacy masking.
Purpose and Features
Previous world's largest open dataset for privacy. The current flagship is now pii-masking-openpii-1m
The purpose of the dataset is to train models to remove personally identifiable information… See the full description on the dataset page: https://huggingface.co/datasets/shivaniachary123/pii-masking-200k.rachelThis is a handmade dataset for making a Japanese chatbot. Conversations will continue to grow.