QuangDuy/unsupervised_data

HUIT-Embedding — unsupervised pairs (Vietnamese)

(query, document) pairs for contrastive pre-training of Vietnamese embedding models. 29 sources, 68,097,453 rows, of which 53,083,874 pass every filter (78.0%). Processed on 2026-10-21. Rows are never dropped: each row carries the reason it was filtered in filter_flag. The original data is kept at tag v0-raw.

Usage

from datasets import load_dataset

ds = load_dataset("QuangDuy/unsupervised_data", "bkai_newssapo_title_sapo", split="train")
train = ds.filter(lambda r: r["filter_flag"] == "pass" and not r["duplicate_flag"])

Recommended training view: filter_flag == "pass" and duplicate_flag == false. Sample one source per batch (probability ∝ size^0.5), cap news at 40 %, cap machine-translated and instruction data, and never put two rows with the same group_id in one batch.

Schema

Column	Type	Meaning
query	string	Title, question, sentence or instruction (cleaned)
document	string	Body, answer, paraphrase or response (cleaned)
source	string	Config name
pair_type	string	Kind of pair (title_body, query_passage, paraphrase_s, …)
translated	bool	Machine-translated text
group_id	string	Rows sharing a query, document or upstream group; keep them out of the same batch
similarity	float32	cos(query, document) with intfloat/multilingual-e5-small; null when not scored
rank	int32	Rank of the true document among 1,000 random documents of the same source; null when not scored
filter_flag	string	pass or the first filter that removed the row
duplicate_flag	bool	Query also appears in a higher-priority source

Sources

Config	Source	Paper	License	Label origin	Translated	Domain	Rows	Pass	Kept
bkai_newssapo_title_body	bkai-foundation-models/BKAINewsCorpus	paper	unknown (news sites)	weak	False	news	21,972,378	18,342,262	83.5%
bkai_newssapo_title_sapo	bkai-foundation-models/BKAINewsCorpus	paper	unknown (news sites)	weak	False	news	21,434,465	18,343,810	85.6%
vn_news_title_context	tmnam20/Vietnamese-News	–	unknown (news sites)	weak	False	news	2,323,628	1,370,210	59.0%
vn_news_title_description	tmnam20/Vietnamese-News	–	unknown (news sites)	weak	False	news	2,317,639	1,354,618	58.4%
msmarco-vn-pretrain	GreenNode/msmarco-vn	paper	see MS MARCO terms (GreenNode/msmarco-vn card states cc-by-sa-4.0)	weak	True	msmarco	5,115,817	3,671,412	71.8%
dbpedia_vn_title_text	GreenNode/dbpedia-vn	paper	cc-by-sa-4.0 (GreenNode card); upstream DBpedia is cc-by-sa-3.0	weak	False	wiki	4,632,996	4,376,528	94.5%
finewiki-vi	HuggingFaceFW/finewiki	paper	cc-by-sa-4.0	weak	False	wiki	1,279,087	1,244,189	97.3%
contrastive-pretraining-vi	orionweller/contrastive-pretraining	–	odc-by	weak	False	web	884,499	682,421	77.2%
crosslingual_law	bkai-foundation-models/crosslingual	paper	agpl-3.0	weak	False	bitext	521,697	489,235	93.8%
visp_paraphrase	ngwgsang/ViSP	paper	mit	human_paper	False	paraphrase	797,352	307,380	38.6%
viepaws	ngwgsang/ViePAWS	paper	unknown	synthetic	False	paraphrase	9,966	9,880	99.1%
vietnam_journal	web crawl of Vietnamese scientific journal websites	–	unknown	weak	False	academic	63,426	17,558	27.7%
vinmec_vn_hoi_dap	https://www.vinmec.com/vi/hoi-dap-bac-si/	–	unknown	weak	False	medical	17,253	16,661	96.6%
medlatec_vn_hoi_dap	https://medlatec.vn/hoi-dap-suc-khoe	–	unknown	weak	False	medical	5,994	3,916	65.3%
viqpc_question	ngwgsang/ViQPC	–	unknown	weak	False	paraphrase	12,244	0	0.0%
viqpc_sentence	ngwgsang/ViQPC	–	unknown	weak	False	paraphrase	49,286	0	0.0%
viqp	ngwgsang/ViQP	paper	mit	human_paper	False	paraphrase	34,506	64	0.2%
dichvucong_faqs	hieunguyen1053/dichvucong_faqs (config hoi-dap)	–	unknown	weak	False	public-service	1,074	286	26.6%
naturalqa_vi	5CD-AI/Vietnamese-NaturalQA-gg-translated-unrefined	paper	unknown (Natural Questions: cc-by-sa-3.0)	translated	True	qa	99,698	86,166	86.4%
math_vi	5CD-AI/Vietnamese-395k-meta-math-MetaMathQA-gg-translated, 5CD-AI/Vietnamese-microsoft-orca-math-word-problems-200k-gg-translated, 5CD-AI/Vietnamese-nvidia-OpenMathInstruct-1-50k-gg-translated	paper, paper, paper	unknown	synthetic	True	math	328,452	323,952	98.6%
tinycot_vi	5CD-AI/Vietnamese-mabryCodes-tiny-cot-alpaca-gg-translated	–	unknown	synthetic	True	instruction	599,093	541,086	90.3%
webtext_vi	5CD-AI/Vietnamese-nampdn-ai-tiny-webtext-gg-translated	–	unknown (upstream nampdn-ai/tiny-webtext: mit)	weak	True	web	4,550,405	1,046,384	23.0%
health200k_vi	https://github.com/doanthuan/bloom-gpt/raw/main/data/translated_health_200k.zip	paper	unknown	weak	True	medical	198,396	80,260	40.5%
medical_lc	mtue29/vietnamese-medical-dataset	–	apache-2.0	weak	False	medical	231,582	208,075	89.8%
medical_qa_hungnm	hungnm/vietnamese-medical-qa	–	apache-2.0	weak	False	medical	9,335	7,100	76.1%
medical_qa_dqdung	Dqdung205/medical-vietnamese-qa	–	cc-by-sa-4.0	weak	False	medical	13,594	11,679	85.9%
hoidap_tvpl	52100303-TranPhuocSang/hoidap-tvpl-500k, phuocsang/hoidap-thuvienphapluat-200k, vietgpt/tvpl_qa	–	unknown (phuocsang/hoidap-thuvienphapluat-200k: mit)	weak	False	legal	433,128	395,914	91.4%
vilexnorm	visolex/ViLexNorm	paper	cc-by-nc-sa-4.0 (per SEACrowd; the HF card states none)	human_paper	False	lexnorm	10,463	10,179	97.3%
diacritic_vi	yammdd/vietnamese-diacritic-restoration-corpus	–	unknown	weak	False	diacritics	150,000	142,649	95.1%

Notes:

• vn_news_title_context: Subset all.
• vn_news_title_description: Subset all.
• msmarco-vn-pretrain: Pairs are consecutive passages from the MS MARCO corpus, not query -> passage pairs.
• finewiki-vi: Vietnamese subset.
• contrastive-pretraining-vi: Vietnamese part.
• visp_paraphrase: Paraphrases are manually verified according to the ViSP card.
• viepaws: Card describes LLM-based word scrambling with LLM silver labels followed by human verification.
• vietnam_journal: Title -> abstract pairs from many Vietnamese journals; the crawler and full site list are not documented. Sampled rows trace back to sites such as sj.agu.edu.vn, ctujsvn.ctu.edu.vn, tapchiyhcd.vn and jfs.ulis.vnu.edu.vn.
• vinmec_vn_hoi_dap: Crawled doctor Q&A pages.
• medlatec_vn_hoi_dap: Crawled health Q&A pages.
• viqpc_question: Construction is undocumented; pairs are filtered with the provided sem/lex scores.
• viqpc_sentence: Construction is undocumented; pairs are filtered with the provided sem/lex scores.
• math_vi: Upstream solutions are model-generated; the Vietnamese text is a Google machine translation.
• health200k_vi: Bloom-GPT README says the file was collected from ChatDoctor and translated to Vietnamese.
• hoidap_tvpl: Decontaminated against GreenNode/TVPL-Retrieval-VN (VN-MTEB), which shares the same origin.

Rows removed per step

Rows flagged at each step (a row is counted at the first step that flags it). Token lengths use the HuITBERT tokenizer on a sample of kept rows.

Config	cleaning	language	statistics	dedup	decontamination	scoring	Duplicate flag	Query tokens p50/p95	Document tokens p50/p95
bkai_newssapo_title_body	33,124	168,922	15,119	2,381,665	27,994	1,003,292	18,555,763	14/23	546/1642
bkai_newssapo_title_sapo	49,846	197,691	16,801	1,416,182	2,108	1,408,027	0	14/23	43/91
vn_news_title_context	4,117	40,583	4,020	821,162	3,461	80,075	1,424,333	15/22	587/1648
vn_news_title_description	3,847	37,957	789	821,988	305	98,135	764,211	15/21	42/86
msmarco-vn-pretrain	7,238	28,213	5,004	23,337	86	1,380,527	3	33/68	46/90
dbpedia_vn_title_text	1,001	2,552	12,312	136,564	12,281	91,758	3,019	6/12	76/138
finewiki-vi	7	7,416	22,493	3,440	424	1,118	310,211	7/10	36/848
contrastive-pretraining-vi	58,650	48,705	5,658	33,669	1,576	53,820	0	13/22	404/2085
crosslingual_law	0	119	375	52	9,965	21,951	0	27/40	209/270
visp_paraphrase	562	638	52	400,429	88,291	0	15	28/55	29/55
viepaws	18	65	0	0	3	0	13	29/64	30/64
vietnam_journal	6,506	37,537	1,636	37	2	150	4	24/39	235/385
vinmec_vn_hoi_dap	1	0	0	9	285	297	0	93/183	168/403
medlatec_vn_hoi_dap	2	64	5	0	0	2,007	0	54/128	298/612
viqpc_question	0	80	3	9,655	2,506	0	0	0/0	0/0
viqpc_sentence	0	20	4	49,262	0	0	0	0/0	0/0
viqp	28	228	24	17,152	17,010	0	0	12/18	12/18
dichvucong_faqs	167	20	9	16	8	568	0	40/136	49/160
naturalqa_vi	1	5,951	6	630	15	6,929	8	11/17	147/329
math_vi	74	2,301	1,729	0	0	396	0	62/119	153/370
tinycot_vi	217	788	249	8,077	60	48,616	762	222/878	79/202
webtext_vi	155,070	344,463	121,986	2,243,971	47	638,484	181	17/50	144/260
health200k_vi	343	250	54	18	1	117,470	0	94/224	103/239
medical_lc	3,186	6,185	21	4,153	473	9,489	0	9/15	203/228
medical_qa_hungnm	159	160	0	3	314	1,599	0	55/148	107/366
medical_qa_dqdung	1	0	0	368	223	1,323	0	74/157	128/369
hoidap_tvpl	31	490	160	0	31,394	5,139	0	21/33	356/778
vilexnorm	1	253	25	5	0	0	0	11/32	11/31
diacritic_vi	79	2,076	46	5,150	0	0	0	15/51	14/48

Composition of kept rows

• By domain: news: 74.2%, wiki: 10.6%, msmarco: 6.9%, web: 3.3%, instruction: 1.0%, bitext: 0.9%, legal: 0.7%, medical: 0.6%, math: 0.6%, paraphrase: 0.6%, diacritics: 0.3%, qa: 0.2%, academic: 0.0%, lexnorm: 0.0%, public-service: 0.0%
• By label origin: weak: 97.6%, synthetic: 1.6%, human_paper: 0.6%, translated: 0.2%
• Machine-translated: False: 89.2%, True: 10.8%

Consistency scoring

Pairs are scored with intfloat/multilingual-e5-small. Keep rule: rank <= 3 or similarity >= threshold. Thresholds are set without hand labels: for each group, the 99% quantile of the similarity between a query and a random document of the same source (≈1 % false positives). Paraphrase, lexical-normalization and diacritic sources are human-checked and not scored.

Threshold group	Threshold	AUC (pair vs random)	Pairs above threshold
title_long	0.8567	0.9759	83.5%
title_short	0.8413	0.9902	89.1%
crosslingual	0.8615	0.9986	98.0%
title_entity	0.8053	0.999	98.7%
query_passage	0.8861	0.9391	64.3%
instruction	0.8621	0.9949	94.5%

Decontamination

Every row is checked against 68 evaluation sets (328,903 queries, 310,265 relevant documents): exact match after normalization (NFC, lower case, tone marks, spaces) and 13-gram containment ≥ 50 % against evaluation queries. Blocked: all of ViRE, the VN-MTEB retrieval/reranking/STS/pair-classification test data, and the test splits of the training sources.

Evaluation sets

csconda_test, hoidap_tvpl200k_test, hoidap_tvpl_test, isedsc_dev, isedsc_test, vianli_test, vicov19qa_test, vifactcheck_test, vihealthqa_test, vimedaqa_test, vimqa_test, vinli_test, viqp_test, viqpc_test, viquad2_test, vire4mrc_test, vire_alqac, vire_csconda, vire_educoqa, vire_uit-viquad2, vire_vihermes, vire_vimedaqa_v2, vire_vinewsqa, vire_vire4mrc_v2, vire_virhe4qa_v2, vire_vixstory, vire_vlogqa_2, vire_zalolegalqa, virhe4qa_test, viwikifc_test, vlogqa_test, vnmteb_GreenNode-Table-Markdown-Retrieval-VN, vnmteb_TVPL-Retrieval-VN, vnmteb_arguana-vn, vnmteb_askubuntudupquestions-reranking-vn, vnmteb_biosses-sts-vn, vnmteb_climate-fever-vn, vnmteb_cqadupstack-android-vn, vnmteb_cqadupstack-gaming-vn, vnmteb_cqadupstack-gis-vn, vnmteb_cqadupstack-mathematica-vn, vnmteb_cqadupstack-physics-vn, vnmteb_cqadupstack-programmers-vn, vnmteb_cqadupstack-stats-vn, vnmteb_cqadupstack-tex-vn, vnmteb_cqadupstack-unix-vn, vnmteb_cqadupstack-webmasters-vn, vnmteb_cqadupstack-wordpress-vn, vnmteb_dbpedia-vn, vnmteb_fever-vn, vnmteb_fiqa-vn, vnmteb_hotpotqa-vn, vnmteb_msmarco-vn, vnmteb_nfcorpus-vn, vnmteb_nq-vn, vnmteb_quora-vn, vnmteb_scidocs-reranking-vn, vnmteb_scidocs-vn, vnmteb_scifact-vn, vnmteb_sickr-sts-vn, vnmteb_sprintduplicatequestions-pairclassification-vn, vnmteb_stackoverflowdupquestions-reranking-vn, vnmteb_stsbenchmark-sts-vn, vnmteb_trec-covid-vn, vnmteb_twittersemeval2015-pairclassification-vn, vnmteb_twitterurlcorpus-pairclassification-vn, vnmteb_webis-touche2020-vn, vnmteb_zalo-ai-legal-text-retrieval-vn

Rows removed, by evaluation set (top 15): viqpc_test 118,662, vnmteb_TVPL-Retrieval-VN 34,752, vnmteb_dbpedia-vn 11,871, hoidap_tvpl_test 7,828, vire_alqac 5,245, vinli_test 4,627, hoidap_tvpl200k_test 4,614, vire_vinewsqa 3,846, vire_zalolegalqa 1,969, vianli_test 1,239, vihealthqa_test 863, vifactcheck_test 695, isedsc_dev 505, viwikifc_test 370, vnmteb_hotpotqa-vn 349

Excluded sources

Source	Reason
HuggingFaceFW/finepdfs (vie_Latn)	Not usable for Vietnamese.
proj-persona/PersonaHub	No Vietnamese data.
PaxiAI/Vietnamese-Embedding-Training-Data	Repository does not exist.
benhvien108 Q&A crawl	Dropped by user decision.
vksndtc Q&A crawl	Dropped by user decision.
Vietjack Q&A crawl	Dropped by user decision.
sentence-transformers/parallel-sentences-*	Bitext; dropped by user decision.
T2Ranking-vi	This pair type is already covered.
Wikipedia split by section	This pair type is already covered.
XL-Sum	This pair type is already covered.
ura-hcmut/ViMMRC	CC-BY-NC-ND license forbids derivative works; dropped by user decision.
vietgpt/daily_dialog_vi	Dropped by user decision; translation of DailyDialog with weak signal.
esdurmus/wiki_lingua	Dropped by user decision.
Linhz/Wikilingua_VN	Dropped by user decision.
GreenNode/amazon-reviews-multi-vn	Dropped by user decision.
ntphuc149/ViSpanExtractQA	Dropped by user decision.
BookingCare/ViHealthQA	Gated; content could not be inspected.
BoringSpeaker/hoidap247	Gated; content could not be inspected.
VietAI/vi_pubmed	English-Vietnamese bitext.
khoaliamle/vietnamese-medical-qa	Contexts are LLM-generated.
BlossomsAI/vietnamese-conversational-dataset	Machine translation of unknown origin.
hoang2004/vietnamese-qa-system	Low-quality chit-chat.
anti-ai/ViSTS	Test split only, translated from English STS sets, and overlaps the VN-MTEB STS evaluation.
hoailebads/VLSP-Legal-Reranker-Train	Built from Zalo with negatives mined by an AITeamVN model; duplicates the Zalo sources already used.
ContextSearchLM/context_search_vietnamese_prompt_97_minilmtok	News title to passage, the same pair type as the over-represented news data; negatives are random.

Citations

@misc{duc2024comprehensivevietnameseretrievalaugmentedgeneration,
      title={Towards Comprehensive Vietnamese Retrieval-Augmented Generation and Large Language Models}, 
      author={Nguyen Quang Duc and Le Hai Son and Nguyen Duc Nhan and Nguyen Dich Nhat Minh and Le Thanh Huong and Dinh Viet Sang},
      year={2024},
      eprint={2403.01616},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2403.01616}, 
}

@misc{bajaj2018msmarcohumangenerated,
      title={MS MARCO: A Human Generated MAchine Reading COmprehension Dataset}, 
      author={Payal Bajaj and Daniel Campos and Nick Craswell and Li Deng and Jianfeng Gao and Xiaodong Liu and Rangan Majumder and Andrew McNamara and Bhaskar Mitra and Tri Nguyen and Mir Rosenberg and Xia Song and Alina Stoica and Saurabh Tiwary and Tong Wang},
      year={2018},
      eprint={1611.09268},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/1611.09268}, 
}

@misc{vnmteb,
      title={VN-MTEB: Vietnamese Massive Text Embedding Benchmark}, 
      author={Loc Pham and Tung Luu and Thu Vo and Minh Nguyen and Viet Hoang},
      year={2025},
      eprint={2507.21500},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2507.21500}, 
}

@dataset{penedo2025finewiki,
  author    = {Guilherme Penedo},
  title     = {FineWiki},
  year      = {2025},
  publisher = {Hugging Face Datasets},
  url       = {https://huggingface.co/datasets/HuggingFaceFW/finewiki},
  urldate   = {2025-10-20},
  note      = {Source: Wikimedia Enterprise Snapshot API (https://api.enterprise.wikimedia.com/v2/snapshots). Text licensed under CC BY-SA 4.0 with attribution to Wikipedia contributors.}
}

@inproceedings{nguyen-etal-2025-improving,
    title = "Improving {V}ietnamese-{E}nglish Cross-Lingual Retrieval for Legal and General Domains",
    author = "Nguyen, Toan Ngoc  and
      Hai, Nam Le  and
      Hieu, Nguyen Doan  and
      Nguyen, Dai An  and
      Van, Linh Ngo  and
      Nguyen, Thien Huu  and
      Dinh, Sang",
    editor = "Chiruzzo, Luis  and
      Ritter, Alan  and
      Wang, Lu",
    booktitle = "Proceedings of the 2025 Conference of the Nations of the {A}mericas Chapter of the {A}ssociation for {C}omputational {L}inguistics: Human Language Technologies (Volume 2: Short Papers)",
    month = apr,
    year = "2025",
    address = "Albuquerque, New Mexico",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.naacl-short.12/",
    doi = "10.18653/v1/2025.naacl-short.12",
    pages = "142--153",
    ISBN = "979-8-89176-190-2"
}

@inproceedings{nguyen-nguyen-2025-large,
    title = "A Large-Scale Benchmark for {V}ietnamese Sentence Paraphrases",
    author = "Nguyen, Sang Quang  and
      Nguyen, Kiet Van",
    editor = "Chiruzzo, Luis  and
      Ritter, Alan  and
      Wang, Lu",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {NAACL} 2025",
    month = apr,
    year = "2025",
    address = "Albuquerque, New Mexico",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.findings-naacl.59/",
    doi = "10.18653/v1/2025.findings-naacl.59",
    pages = "1045--1060",
    ISBN = "979-8-89176-195-7"
}

@INPROCEEDINGS{viepaws,
  author={Nguyen, Sang Quang and Van Nguyen, Kiet and Nguyen, Ngan Luu-Thuy and Le, Duy-Dinh},
  booktitle={2026 International Conference on Multimedia Analysis and Pattern Recognition (MAPR)}, 
  title={ViePAWS: A Vietnamese Adversarial Dataset for Paraphrase Identification under LLM-based Word Scrambling}, 
  year={2026},
  volume={},
  number={},
  pages={43-48},
  keywords={Modeling;Labeling;Conferences;Printing;Accuracy;Tuning;Generative Pre-trained transformer;Poles and zeros;Surfaces;Large language models;Vietnamese NLP;paraphrase identification;adversarial word scrambling},
  doi={10.1109/MAPR72750.2026.11685740}
}

@INPROCEEDINGS{viqp,
  author={Nguyen, Sang Quang and Vo, Thuc Dinh and Nguyen, Duc P.A and Tran, Dang T. and Nguyen, Kiet Van},
  booktitle={2023 International Conference on Multimedia Analysis and Pattern Recognition (MAPR)}, 
  title={ViQP: Dataset for Vietnamese Question Paraphrasing}, 
  year={2023},
  volume={},
  number={},
  pages={1-6},
  doi={10.1109/MAPR59823.2023.10288738}
}

@article{kwiatkowski-etal-2019-natural,
    title = "Natural Questions: A Benchmark for Question Answering Research",
    author = "Kwiatkowski, Tom  and
      Palomaki, Jennimaria  and
      Redfield, Olivia  and
      Collins, Michael  and
      Parikh, Ankur  and
      Alberti, Chris  and
      Epstein, Danielle  and
      Polosukhin, Illia  and
      Devlin, Jacob  and
      Lee, Kenton  and
      Toutanova, Kristina  and
      Jones, Llion  and
      Kelcey, Matthew  and
      Chang, Ming-Wei  and
      Dai, Andrew M.  and
      Uszkoreit, Jakob  and
      Le, Quoc  and
      Petrov, Slav",
    editor = "Lee, Lillian  and
      Johnson, Mark  and
      Roark, Brian  and
      Nenkova, Ani",
    journal = "Transactions of the Association for Computational Linguistics",
    volume = "7",
    year = "2019",
    address = "Cambridge, MA",
    publisher = "MIT Press",
    url = "https://aclanthology.org/Q19-1026/",
    doi = "10.1162/tacl_a_00276",
    pages = "452--466"
}

@misc{yu2024metamathbootstrapmathematicalquestions,
      title={MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models}, 
      author={Longhui Yu and Weisen Jiang and Han Shi and Jincheng Yu and Zhengying Liu and Yu Zhang and James T. Kwok and Zhenguo Li and Adrian Weller and Weiyang Liu},
      year={2024},
      eprint={2309.12284},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2309.12284}, 
}

@misc{mitra2024orcamathunlockingpotentialslms,
      title={Orca-Math: Unlocking the potential of SLMs in Grade School Math}, 
      author={Arindam Mitra and Hamed Khanpour and Corby Rosset and Ahmed Awadallah},
      year={2024},
      eprint={2402.14830},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2402.14830}, 
}

@misc{toshniwal2024openmathinstruct118millionmath,
      title={OpenMathInstruct-1: A 1.8 Million Math Instruction Tuning Dataset}, 
      author={Shubham Toshniwal and Ivan Moshkov and Sean Narenthiran and Daria Gitman and Fei Jia and Igor Gitman},
      year={2024},
      eprint={2402.10176},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2402.10176}, 
}

@misc{li2023chatdoctormedicalchatmodel,
      title={ChatDoctor: A Medical Chat Model Fine-Tuned on a Large Language Model Meta-AI (LLaMA) Using Medical Domain Knowledge}, 
      author={Yunxiang Li and Zihan Li and Kai Zhang and Ruilong Dan and Steve Jiang and You Zhang},
      year={2023},
      eprint={2303.14070},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2303.14070}, 
}

@misc{doan2023efficientfinetuninglargelanguage,
      title={Efficient Finetuning Large Language Models For Vietnamese Chatbot}, 
      author={Vu-Thuan Doan and Quoc-Truong Truong and Duc-Vu Nguyen and Vinh-Tiep Nguyen and Thuy-Ngan Nguyen Luu},
      year={2023},
      eprint={2309.04646},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2309.04646}, 
}

@misc{vietnamese-medical-dataset,
  title  = {Vietnamese Medical Dataset},
  author = {Tue Phan Vo Minh},
  year   = {2025},
  url    = {https://huggingface.co/datasets/mtue29/vietnamese-medical-dataset}
}

@misc{vietnamese-medical-qa,
      title={Vietnamese Medical QA: Question Answering dataset for medical in Vietnamese},
      author={Hung Nguyen},
      year={2024},
}

@inproceedings{nguyen-etal-2024-vilexnorm,
    title = "{V}i{L}ex{N}orm: A Lexical Normalization Corpus for {V}ietnamese Social Media Text",
    author = "Nguyen, Thanh-Nhi  and
      Le, Thanh-Phong  and
      Nguyen, Kiet",
    editor = "Graham, Yvette  and
      Purver, Matthew",
    booktitle = "Proceedings of the 18th Conference of the {E}uropean Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
    month = mar,
    year = "2024",
    address = "St. Julian{'}s, Malta",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.eacl-long.85/",
    doi = "10.18653/v1/2024.eacl-long.85",
    pages = "1421--1437"
}

@inproceedings{vire,
    title = "Which Works Best for {V}ietnamese? {A} Practical Study of Information Retrieval Methods across Domains",
    author = "Nguyen, Long S. T.  and
      Quan, Tho T.",
    editor = "Demberg, Vera  and
      Inui, Kentaro  and
      Marquez, Llu{\'i}s",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EACL} 2026",
    month = mar,
    year = "2026",
    address = "Rabat, Morocco",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.findings-eacl.110/",
    doi = "10.18653/v1/2026.findings-eacl.110",
    pages = "2098--2119",
    ISBN = "979-8-89176-386-9"
}

No verified citation was found for: vn_news_title_context, vn_news_title_description, contrastive-pretraining-vi, vietnam_journal, vinmec_vn_hoi_dap, medlatec_vn_hoi_dap, viqpc_question, viqpc_sentence, dichvucong_faqs, tinycot_vi, webtext_vi, medical_qa_dqdung, hoidap_tvpl, diacritic_vi.

Limitations and licenses

Each source keeps its own license (column above); this collection is license: other. Sources that are non-commercial or have no stated license: bkai_newssapo_title_body, bkai_newssapo_title_sapo, vn_news_title_context, vn_news_title_description, viepaws, vietnam_journal, vinmec_vn_hoi_dap, medlatec_vn_hoi_dap, viqpc_question, viqpc_sentence, dichvucong_faqs, naturalqa_vi, math_vi, tinycot_vi, webtext_vi, health200k_vi, hoidap_tvpl, vilexnorm, diacritic_vi. Check the original license before any use beyond research. Q&A sources crawled from the web had person names masked with a Vietnamese NER model and phone/ID/account numbers masked by regex (<name>, <phone>, …).