SPAISS6F1/spai-ss6-llm-1b-thai-corpus

Thai Medical And Health Corpus

Thai public medical and health web corpus collected for research and LLM dataset experimentation, with optional imported Thai medical/health datasets from Hugging Face stored as separate configs.

Public Web Corpus

• Config: default
• Split: train
• Records: 3660 deduplicated articles
• Columns: 16
• Format: Parquet
• Latest collection profile: free_1000
• Latest generated at: 2026-06-06T17:41:38.787978+00:00

Source And Method

The default corpus URLs are collected from public sitemap XML files on configured Thai public sources, then crawled with robots.txt checks, rate limiting, Thai text validation, PII-like redaction, and content deduplication.

Free sitemap collection does not use Google SERP ranking, so rank is 0 for sitemap-derived records.

Additional Hugging Face Sources

These sources are imported as separate configs and keep their original column schemas where possible. Check each upstream dataset card before redistribution, commercial use, or model release.

Config	Source	License	Rows	Columns
medical_o1_reasoning_sft_thai	ZombitX64/Medical-o1-Reasoning-SFT-Thai	cc-by-sa-4.0	16396	3
medical_o1_verifiable_problem_thai	ZombitX64/Medical-o1-verifiable-problem-Thai	cc-by-sa-4.0	40906	2
mental_health_thai	PahnKCN/mental_health_thai	unknown	21544	2
thai_medical_care_aiotx	AIoTx-Tech/thai_medicalCare_dataset	apache-2.0	3599	11

Imported source metadata is stored in metadata/external_sources.json.

Existing Mirrored Hub Sources

These Parquet files already exist in this repository and are exposed as separate configs for the Dataset Viewer. Source and license values are kept conservative when upstream provenance is not clear.

Config	Source	License	Rows	Files	Size
seapile	Existing seapile shards in this repo	unknown	16428048	54	22.18 GB
finance_business_pythainlp_thai_financial_dataset	pythainlp/thai-financial-dataset mirror	unknown	502942	11	1.68 GB
law	Existing law corpus mirror in this repo	unknown	208964	1	0.49 GB
nation	Existing nation corpus mirror in this repo	unknown	240945	1	0.69 GB
social	Existing social corpus mirror in this repo	unknown	22983	1	1.45 GB

Mirrored source metadata is stored in metadata/repo_existing_sources.json.

Massive Imported Sources

These larger upstream Hugging Face datasets are mirrored as separate configs. Some upstream Dataset Viewer conversions are partial; see metadata/massive_sources.json for source row estimates and import details.

Config	Source	License	Rows	Files	Size
wangchanlion_web	aisingapore/WangchanLION-Web	odc-by	557502	10	1.97 GB
wangchanlion_curated	aisingapore/WangchanLION-Curated	odc-by	401532	19	2.64 GB
khanomtanllm_thai_subset	wannaphong/KhanomTanLLM-pretrained-dataset-thai-subset	odc-by	464339	10	1.80 GB
thai_culturax_clean	pythainlp/thai-culturax-clean-dataset	odc-by	818727	10	1.96 GB
thai_wikipedia_clean_20230101	pythainlp/thai_wikipedia_clean_20230101	cc-by-sa-3.0	1436054	2	0.26 GB
scb_mt_en_th_2020_mt_opus	pythainlp/scb-mt-en-th-2020_mt-opus	cc-by-sa-3.0	4319905	3	0.45 GB
thaisum	pythainlp/thaisum	mit	391868	9	1.30 GB
prd_news_30112023	pythainlp/prd_news_30112023	cc0-1.0	242985	3	0.39 GB

Massive source metadata is stored in metadata/massive_sources.json.

Thai Instruction And QA Sources

These upstream Hugging Face datasets are mirrored as instruction, QA, and task-oriented Thai configs for SFT and evaluation experiments. Check each upstream dataset card before redistribution, commercial use, or model release.

Config	Source	License	Rows	Files	Size
thai_local_instruction_v2	pythainlp/thai-local-instruction-v2	cc-by-4.0	39829	1	4.95 MB
thai_instruction_sft_suraponn	Suraponn/thai_instruction_sft	apache-2.0	131907	3	294.75 MB
thai_synthetic_qa_v1	ThaiSyntheticQA/ThaiQA-v1	cc-by-4.0	12668	1	17.66 MB
thai_idioms_instruction	pythainlp/thai-idioms-instruction	cc-by-3.0	1152	1	0.13 MB
thai_tourist_attraction_instruction	wannaphong/thai-tourist-attraction-instruction	cc-by-3.0	51662	1	8.59 MB
thai_synonym_instruction	wannaphong/thai-synonym-instruction	cc0-1.0	167	1	0.02 MB
thaiqa_lst20_mit	SuperAI2-Machima/ThaiQA_LST20	mit	7643	1	5.23 MB

Instruction source metadata is stored in metadata/instruction_sources.json.

Thai Exam And Answer-Key Sources

These O-NET and Thai exam records are normalized into the thai_exam_qa_with_answers config with common fields for question, choices, answer, explanation, source license, and raw provenance. Only sources with explicit compatible license tags are imported in v1.

• Normalized rows: 8,191
• Config: thai_exam_qa_with_answers
• Review-only source list: metadata/exam_review_sources.json

Source	License	Raw Rows	Normalized	Duplicates
Thaweewat/onet-m6-social	cc-by-sa-3.0	510	510	0
TaengooTV/math_onet	mit	191	191	0
matichon/thai-onet-m6-exam	apache-2.0	2750	639	2111
openthaigpt/thai-investment-consultant-licensing-exams	apache-2.0	160	160	0
typhoon-ai/thai_exam	apache-2.0	590	590	0
EIRTHAIMED/ThaiExam	apache-2.0	1627	1618	0
biodatlab/thai_buddhist_studies_exam	apache-2.0	4100	3917	183
RJTPP/thai_exam-reformatted	apache-2.0	1163	566	597

Exam source metadata is stored in metadata/exam_sources.json.

Token Estimate

• Estimated tokens: 27,313,716,557
• Tokenizer/method: Qwen/Qwen2.5-1.5B
• Sampled rows: 1,760
• Generated at: 2026-06-06T21:02:15.852689+00:00

Token estimates are sample-based and should be treated as planning numbers, not final training-token accounting. Details are stored in metadata/token_estimate.json.

Intended Use

This dataset is intended for research, corpus inspection, preprocessing experiments, retrieval experiments, and LLM training/evaluation exploration.

Limitations

• This dataset is not medical advice.
• Web article quality varies by source.
• External configs may have different collection methods, schemas, licenses, and usage restrictions from the default public web corpus.
• Redistribution, commercial use, or model release requires separate rights and legal review because source website and upstream dataset licenses may differ.

Generation Artifacts

• data/train.parquet
• metadata/run_manifest.json
• metadata/serp_urls.parquet
• metadata/external_sources.json
• metadata/repo_existing_sources.json
• metadata/massive_sources.json
• metadata/instruction_sources.json
• metadata/exam_sources.json
• metadata/exam_review_sources.json
• metadata/token_estimate.json
• DATASET_AUDIT.md
• DATASET_REPOS.md