Meddies/meddies-embedding-seed-corpora

Meddies Vietnamese Embedding Seed Corpora

Normalized Parquet subsets of Vietnamese clinical, legal, and government text, exported CPU-only on Modal for later chunking, teacher query generation, and license-aware curation.

Derived artifacts. Every subset is a normalized derivative of a public upstream dataset (listed below). This repository hosts only the derived Parquet; the upstream datasets are the source of truth for content and licensing. Verify licensing against the upstream source, not this mirror.

• Source rows streamed: 21,870,263
• Corpus rows: 41,963,062
• Pair rows: 42,936,095

training_ready is false on every row. No subset is cleared for training or redistribution until license review and quality review pass.

Licensing and allowed use

Subset	License status	Allowed use
clinical_vi_pubmed_parallel	needs_review	internal_only_until_license_review
gov_vi_dichvucong_procedures	needs_review	internal_only_until_license_review
gov_vi_nso_catalog	needs_review	eval_only_until_license_review
gov_vi_sapnhap_bando	needs_review	eval_only_until_license_review
legal_vi_court_judgments	needs_review	internal_only_until_license_review
legal_vi_metadata_th1nhng0	needs_review	metadata_only_until_body_text_found
legal_vi_pbgdpl_qna	needs_review	internal_only_until_license_review
legal_vi_phapdien_articles	needs_review	internal_only_until_license_review
legal_vi_precedents_anle	needs_review	internal_only_until_license_review
legal_vi_terms_thuvienphapluat_bilingual	needs_review	internal_only_until_license_review
legal_vi_vbpl_documents	needs_review	internal_only_until_license_review

Upstream sources

Subset	Upstream dataset	Config	Domain	Language
clinical_vi_pubmed_parallel	VietAI/vi_pubmed	default	medical	vi-en
gov_vi_dichvucong_procedures	tmquan/dichvucong-gov-vn	procedures	general_vietnamese	vi
gov_vi_nso_catalog	tmquan/nso-gov-vn	catalog	general_vietnamese	vi
gov_vi_sapnhap_bando	tmquan/sapnhap-bando-vn	all	general_vietnamese	vi
legal_vi_court_judgments	tmquan/congbobanan-toaan-gov-vn	documents	legal	vi
legal_vi_metadata_th1nhng0	th1nhng0/vietnamese-legal-documents	metadata	legal	vi
legal_vi_pbgdpl_qna	tmquan/pbgdpl-vn-legal-qna	qa	legal	vi
legal_vi_phapdien_articles	tmquan/phapdien-moj-gov-vn	articles	legal	vi
legal_vi_precedents_anle	tmquan/anle-toaan-gov-vn	documents	legal	vi
legal_vi_terms_thuvienphapluat_bilingual	tmquan/thuvienphapluat-vn-tnpl	bilingual	legal	vi-en
legal_vi_vbpl_documents	tmquan/vbpl-vn	documents	legal	vi

Subset statistics

Subset	Source rows	Corpus rows	Pair rows
clinical_vi_pubmed_parallel	20,087,006	40,164,459	40,163,554
gov_vi_dichvucong_procedures	3,927	15,480	7,854
gov_vi_nso_catalog	502	501	0
gov_vi_sapnhap_bando	6,712	6,712	10,067
legal_vi_court_judgments	1,370,726	1,370,726	2,399,288
legal_vi_metadata_th1nhng0	153,420	153,418	0
legal_vi_pbgdpl_qna	4,593	4,593	9,186
legal_vi_phapdien_articles	65,967	65,451	130,895
legal_vi_precedents_anle	1,963	1,968	3,083
legal_vi_terms_thuvienphapluat_bilingual	16,625	32,465	64,912
legal_vi_vbpl_documents	158,822	147,289	147,256

Build

• Run: seed-corpora-full-20260630-2227-meddiesresearch-r1
• Platform: Modal CPU (no GPU)
• Bulk data was streamed and written remotely; no large local download.
• Two subsets carry only corpus rows (no pairs): gov_vi_nso_catalog, legal_vi_metadata_th1nhng0.