datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
soc-ratchakitcha
Royal Gazette Thailand (Ratchakitcha) Dataset
ชุดข้อมูลราชกิจจานุเบกษา (แบบ Machine Readable)
โครงการ Open Law Data Thailand ร่วมกับคณะกรรมาธิการการพาณิชย์และการอุตสาหกรรม วุฒิสภา ได้รับความอนุเคราะห์ข้อมูลจาก สำนักเลขาธิการคณะรัฐมนตรี (สลค.) เพื่อเผยแพร่ข้อมูลกฎหมายไทยสู่สาธารณะในรูปแบบที่ประมวลผลได้ด้วยคอมพิวเตอร์ (Machine Readable) เพื่อส่งเสริมนวัตกรรม Legal Tech และ AI ของประเทศไทย
Dataset Description
ชุดข้อมูลนี้รวบรวมรายการประกาศในราชกิจจานุเบกษา… See the full description on the dataset page: https://huggingface.co/datasets/open-law-data-thailand/soc-ratchakitcha.VietSuperSpeech
VietSuperSpeech
Vietnamese Speech Recognition Dataset
Dataset Information
Total samples: 32,267
Train samples: 29,041
Dev samples: 3,226
Total duration: 103.18 hours
Sample rate: 16000 Hz
Average segment length: ~12 seconds
Source Datasets
asr_dataset_nguoivietdailynews
asr_dataset_nguyenkhangofficial
asr_dataset_trinhlieu
Format
The dataset follows Icefall format:
train.json: Training samples
dev.json: Development samples
manifest.json:… See the full description on the dataset page: https://huggingface.co/datasets/thanhnew2001/VietSuperSpeech.thai_exam
Dataset Card for Thai_Exam
ThaiExam is a Thai knowledge benchmarking dataset, consisting of multiple-choice questions from examinations in Thailand. The dataset was originally developed for evaluating Typhoon (Thai LLM). This dataset contains 5 splits corresponding to 5 examinations as follows:
ONET: The Ordinary National Educational Test (ONET) is an examination for students in Thailand. This dataset is based on the grade-12 ONET exam, comprising 4 subjects and each question has 5… See the full description on the dataset page: https://huggingface.co/datasets/typhoon-ai/thai_exam.WangchanThaiInstruct_Multi-turn_Conversation_Dataset
WangchanThaiInstruct Multi-turn Conversation Dataset
We create a Thai multi-turn conversation dataset from airesearch/WangchanThaiInstruct (Batch 1) by LLM. It was created from synthetic method using open source LLM in Thai language.
Citation
Thammaleelakul, S., & Phatthiyaphaibun, W. (2024). WangchanThaiInstruct Multi-turn Conversation Dataset [Data set]. Zenodo. https://doi.org/10.5281/zenodo.13132633
or BibTeX
@dataset{thammaleelakul_2024_13132633,
author =… See the full description on the dataset page: https://huggingface.co/datasets/ThaiSyntheticQA/WangchanThaiInstruct_Multi-turn_Conversation_Dataset.SKILLRET
SkillRet Benchmark
📄 Technical report: SkillRet: A Large-Scale Benchmark for Skill Retrieval in LLM Agents (arXiv:2605.05726)
Dataset Overview
SkillRet is a retrieval benchmark for matching natural-language user requests to agent skills. It contains a curated library of public agent skills from GitHub with synthetic training and evaluation queries.
Dataset Statistics
Metric
Value
Total Records
218,157
Total File Size
714 MB
Total… See the full description on the dataset page: https://huggingface.co/datasets/ThakiCloud/SKILLRET.thai-aligner-bench
Thai Aligner Bench
🚧 Development in progress.
How accurately can a forced aligner place Thai token and word boundaries in
speech? This is a self-contained benchmark: one Python file
(aligner_bench.py) plus 1,572 clips of Thai speech with frame-exact timing
ground truth. No Thai NLP stack or other code is needed — just
numpy soundfile torch torchaudio transformers.
The ground truth is what makes the dataset useful: the audio was rendered by a
TTS model whose duration predictor… See the full description on the dataset page: https://huggingface.co/datasets/wayu-ai/thai-aligner-bench.embodied-spatial-reasoning
Embodied Spatial Reasoning Tasks
Dataset Description
This dataset is part of the embodied-spatial-reasoning project, where the agent has to actively explore the environment to determine if certain spatial relationships hold true. The tasks involve spatial reasoning with various objects and scenes. Each task includes a query about the spatial relationships between objects within a scene, which the agent must verify through exploration.
Dataset Structure
The… See the full description on the dataset page: https://huggingface.co/datasets/thanhqt2002/embodied-spatial-reasoning.taipydata25that-one-google-math-datasetapolocheese for poor format, it's because I Don't Care (i'm tired and still working)
data from: https://github.com/google-deepmind/mathematics_dataset
from huggingface_hub import snapshot_download
from datasets import load_dataset
import os
def get_all_files(directory):
file_paths = []
for root, dirs, files in os.walk(directory):
for name in files:
full_path = os.path.join(root, name)
file_paths.append(os.path.abspath(full_path))
return file_paths… See the full description on the dataset page: https://huggingface.co/datasets/midwestern-simulation/that-one-google-math-dataset.thai_exam-reformattedReformatted version of scb10x/thai_exam
Additional Changes:
Fix math incorrect answer
ถ้า \log_{\frac{1}{4}} 256 + \frac{2\log 625}{\log 5} = 3^a เมื่อ a เป็นจำนวนจริง แล้วคำตอบของ a เท่ากับเท่าใด?
a. \log_{3} 2
b. \log_{3} 4
c. \log_{3} \frac{33}{4}
d. \log_{3} 10
e. \log_{3} 12
# Original answer: d (\log_{3} 10)
# Correction : b (\log_{3} 4)
thai-exam-seacrowdthai-culturax-clean-dataset
Thai CulturaX Clean dataset
The data is sourced from the Thai subset of CulturaX dataset, which itself is sourced from mC4 and four OSCAR corpora.
It has about 8,748,575,684 words (without whitespace) and 16,768,585 lines (97 GB).
It was filtered content promoting gambling, adult content, and narcotics.
GitHub for clean: https://github.com/wannaphong/thai-filter-website
Considerations for Using the Data
This dataset is the cleaned version of the CulturaX datasets… See the full description on the dataset page: https://huggingface.co/datasets/pythainlp/thai-culturax-clean-dataset.amazon-2023-thai-1m
Amazon 2023 Thai 1M / ชุดข้อมูลสินค้า Amazon ภาษาไทย 1 ล้านรายการ
ไทย | English below
ชุดข้อมูลสินค้าอีคอมเมิร์ซภาษาไทย 1,000,000 รายการ แปลจากชุดข้อมูล Amazon Reviews '23 Extension ของ Google ด้วยโมเดล typhoon-translate-4b เพื่อใช้ทดสอบ/สาธิตระบบค้นหาเชิงความหมาย (semantic search) ภาษาไทย
แหล่งที่มา / Source & Attribution ⚠️ สำคัญ
ชุดข้อมูลนี้เป็นงานแปล (derivative work) จาก:
ชุดข้อมูลต้นทาง
google/extended_amazon_2023_dataset (Amazon Reviews '23… See the full description on the dataset page: https://huggingface.co/datasets/pnpkke/amazon-2023-thai-1m.banksft_training_corpusThaiExamFinetune Multiple-Choice Question Dataset in Thai
Details:
This dataset is designed for fine-tuning Thai language models, focusing on the Chain-of-Thought (COT) process, which aids in analyzing questions and deriving correct answers step by step. The dataset consists of multiple-choice questions divided into five categories:
O-NET: Ordinary National Educational Test
IC: Investment Consultant
TGAT: Thai General Aptitude Test
TPAT: Thai Professional Aptitude Test… See the full description on the dataset page: https://huggingface.co/datasets/EIRTHAIMED/ThaiExam.Vietnamese-Legal-QA
Vietnamese Legal QA — Question Specificity
Phân loại độ cụ thể của câu hỏi pháp luật dân sự Việt Nam: broad (hỏi khái
quát, phải tổng hợp nhiều chế định) hay narrow (hỏi vào một tình huống / một
điều luật xác định). Dùng để định tuyến truy vấn trong hệ RAG pháp luật.
Cấu trúc
Mỗi dòng là một câu hỏi kèm vết gán nhãn. Hai dòng cùng pair_id là một cặp
đối chứng sinh từ cùng một điều luật — một broad, một narrow.
Trường
Ý nghĩa
item_id, pair_id… See the full description on the dataset page: https://huggingface.co/datasets/ThanhVu101/Vietnamese-Legal-QA.kodialectbench
Code repository: https://github.com/sylvanus4/kodialectbench
English card: README.en.md
⛔ This repository contains zero AI Hub derivatives. No item text, no marker lexicon,
and no original item identifiers are included. Item ids are one-way hashes, and you
regenerate the text and the lexicon yourself from your own AI Hub account's data.
KoDialectBench v0.1.0
A benchmark that measures Korean regional dialects (사투리/방언). It covers five regions —
Gangwon, Gyeongsang, Jeolla… See the full description on the dataset page: https://huggingface.co/datasets/ThakiCloud/kodialectbench.marketing-benchmark-of-more-than-10-ai-models
Marketing Benchmark of 10+ AI Models
A 5,000-question benchmark for evaluating LLMs across six dimensions of modern
marketing — Meta Ads, Google Ads, SEO & Organic, Email & Lifecycle, Critical
Thinking, and Action-Based scenarios — graded through 10 distinct marketer personas.
Every question is independently authored by the AdsGPT Marketing Bench team.
Knowledge MCQs are hand-authored against 2026 platform documentation; open-ended
and action-based scenarios are built from… See the full description on the dataset page: https://huggingface.co/datasets/adsgpt/marketing-benchmark-of-more-than-10-ai-models.this-that-complex-decisions
this-that-complex-decisions
1,710 decisions where the answer follows from a stated policy applied to a state, and where no
single field of that state gives it away.
1,710 questions 19 decision types 40 domains chance rate 0.258
Each row is a state, a question, a closed set of options, and the index of the one option the
policy selects. The answer is determinate: given the state and the policy there is exactly one
correct choice, and it does not depend on anyone's… See the full description on the dataset page: https://huggingface.co/datasets/limberc/this-that-complex-decisions.MultiSensor-Home1A simple way to download the dataset:
# Make sure hf CLI is installed: pip install -U "huggingface_hub[cli]"
hf download thanhhff/MultiSensor-Home1 --repo-type=dataset --local-dir dataset/home1
The MultiSensor-Home2 dataset is available at: https://huggingface.co/datasets/thanhhff/MultiSensor-Home2/
MultiSensor-Home1: Benchmark for Multi-modal Multi-view Action Recognition in Home Environments
A wide-area multi-modal multi-view dataset for action recognition and… See the full description on the dataset page: https://huggingface.co/datasets/thanhhff/MultiSensor-Home1.superskillret-index
superskillret prebuilt index — full-context
Prebuilt embedding index for the superskillret Claude Code plugin.
Unlike the v1 index (which embedded only name + description), v2 encodes
the full skill body (name + description + body) up to
max_seq_length=32768 tokens. Larger index, much higher recall on
skills whose name/description don't capture every keyword in the body.
Version: 2
Corpus: ThakiCloud/SKILLRET (train+test)
Encoder: ThakiCloud/SkillRet-Embedding-0.6B
Skills… See the full description on the dataset page: https://huggingface.co/datasets/ThakiCloud/superskillret-index.ThaiQA-v1
ThaiQA v1
ThaiQA v1 is a Thai Synthetic QA dataset. It was created from synthetic method using open source LLM in Thai language.
We used Nvidia Nemotron 4 (340B) to create this dataset.
Topics:
Technology and Gadgets 100
Travel and Tourism 91
Food and Cooking 99
Sports and Fitness 50
Arts and Entertainment 24
Home and Garden 72
Fashion and Beauty 99
Science and Nature 100
History and Culture 91
Education and Learning 99
Pets and Animals 83
Relationships and Family 78
Personal… See the full description on the dataset page: https://huggingface.co/datasets/ThaiSyntheticQA/ThaiQA-v1.this-that-spatial-bench
spatial-decisions
7,305 multiple-choice decision questions over 6,525 distinct simulated states, in
15 families and two environments. Every answer is computed from the simulator, not
annotated by a person and not taken from a model. That is the point of the set: on a question whose
answer is derived from the rules of the environment, a disagreement is a mistake, and there is
nothing to argue about.
The set was built to replace a much narrower public artefact: a recording of 68… See the full description on the dataset page: https://huggingface.co/datasets/limberc/this-that-spatial-bench.kanitakorn-v23-thaiexam-clean-20260614
Qwen v23 ThaiExam Clean Mix
Audited fallback mix for Kanitakorn. It avoids v20/v21 replay, avoids v13+v17 double replay, uses v1 repair once, includes all normalized worker v2, adds clean worker v3, and keeps small IF/math/identity retainers.
Validation
Records: 8,214
inspect_generated_jsonl: 3,214 source rows valid, 0 invalid, 0 duplicate prompts
inspect_sft_mix: 0 role errors, 0 empty errors
Contamination scan: 0 issues with 8,608 benchmark texts loaded… See the full description on the dataset page: https://huggingface.co/datasets/Jnx03/kanitakorn-v23-thaiexam-clean-20260614.repro-fixed-budget-no-harder-than-fixed-confidence-bai-traces
Agent traces
Agent sessions published from a Trackio Logbook.
KhanomTanLLM-pretrained-dataset-thai-subset
KhanomTanLLM pretrained dataset (Thai subset)
This daataset collect all raw text for pretraining LLM. (Thai subset)
Codename: numfa v2
Repository: https://github.com/pythainlp/KhanomTanLLM
Thai
pythainlp/thai_food_v1.0
pythainlp/thailaw-v1.0
pythainlp/thai-tnhc2-books
pythainlp/thai-constitution-corpus
pythainlp/thai-it-books
pythainlp/prd_news_3011202
pythainlp/thailand-policy-statements
pythainlp/thai-cc-license
pythainlp/blognone_news
pythainlp/goethe-website… See the full description on the dataset page: https://huggingface.co/datasets/wannaphong/KhanomTanLLM-pretrained-dataset-thai-subset.vncompress-vi-v2
VNCompress-VI v2 — query-conditioned context compression for Vietnamese
⚠️ BẢN ĐANG REBUILD — WORK IN PROGRESS.
corpus/qa/qa_synthetic đã sạch và ổn định. compression.jsonl mới có
486 hàng thật (prompt v3, trích xuất) trên tổng số dự kiến 100.000+ —
quá trình sinh đang tạm dừng vì tỉ lệ drop "không đạt ngân sách token"
rất cao (90% ở batch gần nhất) chưa được xử lý gốc rễ. Đừng dùng
compression.jsonl để báo cáo kết quả benchmark hay train E5/E6 ở quy mô
lớn — số hàng hiện tại… See the full description on the dataset page: https://huggingface.co/datasets/thanthienhai/vncompress-vi-v2.seed-free-synthetic-instruct-thai-v1
Seed-Free Synthetic Instruct Thai v1 (F+C+D+)
This dataset is part of the research paper "Seed-Free Synthetic Data Generation Framework for Instruction-Tuning LLMs: A Case Study in Thai" submitted to ACL SRW 2024. It represents the best-performing synthetic dataset (F+C+D+) generated using our novel seed-free framework for low-resource languages, specifically Thai.
Dataset Details
Size: 5,000 instructions
Language: Thai
Task: Instruction-tuning for Large Language Models… See the full description on the dataset page: https://huggingface.co/datasets/parinzee/seed-free-synthetic-instruct-thai-v1.thai_instruction_sft