mozarcik/clinical-en-smpc-awq-calibration

Clinical-EN SmPC AWQ Calibration Corpus

An English clinical-domain text corpus used as a calibration set for AWQ / AutoAWQ / GPTQ post-training quantization of English (or multilingual) clinical large language models. The corpus is dense, domain-specific clinical English (pulmonology + thoracic oncology), drawn from the English Summary of Product Characteristics (SmPC, Annex I of the EMA Product Information).

This is the cross-language counterpart to the Polish corpus `mozarcik/clinical-pl-smpc-awq-calibration`: the same medicines, but the text is the English Annex I instead of the Polish ChPL, extracted from the same EMA EPAR documents. It is intended for English-side and cross-lingual calibration experiments alongside the mozarcik/Llama-PLLuM-70B-*-awq series.

Streszczenie (PL). Angielski odpowiednik korpusu kalibracyjnego `mozarcik/clinical-pl-smpc-awq-calibration`. Te same leki, ale tekst pochodzi z angielskiej ChPL (Annex I Charakterystyki Produktu Leczniczego) z tych samych dokumentów EPAR EMA — nie z polskiej wersji. 412 fragmentów (chunków) gęstego klinicznego angielskiego (pulmonologia + onkologia klatki piersiowej) z 61 leków EMA. Identyczny schemat per-chunk jak w korpusie PL, z language: "en". Przeznaczenie: dane kalibracyjne do kwantyzacji potreningowej (AWQ / AutoAWQ / GPTQ) anglojęzycznych i wielojęzycznych modeli LLM, oraz strona angielska eksperymentów kalibracji międzyjęzykowej. To nie jest zbiór treningowy ani ewaluacyjny.

Contents / Zawartość

412 text chunks (soft window ~150 words, hard cap 255, min 80; ~512-token budget) extracted from the English Annex I SmPC / Product Information of 61 EMA centrally-authorised medicines in pulmonology and thoracic oncology. Each record carries per-chunk source provenance with language: "en":

{
  "text": "4.1 Therapeutic indications TAGRISSO as monotherapy is indicated for: ...",
  "source_authority": "EMA",
  "source_document_type": "SmPC / Product Information",
  "source_url": "https://www.ema.europa.eu/en/documents/product-information/tagrisso-epar-product-information_en.pdf",
  "medicine": "osimertinib",
  "brand_name": "TAGRISSO",
  "language": "en",
  "retrieved_at": "2026-06-06",
  "chunk_id": "EMA_osimertinib_en_0001",
  "license_note": "EMA reproduction policy; source attribution required"
}

All 412 chunks have source_authority = "EMA", source_document_type = "SmPC / Product Information" and language = "en". The SmPC sections most represented are §4.2 (posology), §4.8 (undesirable effects), §5.1 (pharmacodynamics / clinical efficacy), §4.1 (indications), §4.6, §4.4 and §5.2 — the same clinical-efficacy / pharmacology bias as the Polish corpus.

The medicine list is shared with the Polish corpus (61 INNs realized here, brand → EPAR slug map in manifest.json). Files in this dataset repository:

corpus.jsonl SHA-256: 820b7a4de75f20baddf3d89fa73a3d8f348ae22086f1ed52f0278ec6040aac85

How it was built / Jak powstał

Structurally identical pipeline to the Polish corpus; only language-bound tokens and an explicit fetch step differ. For each of the 61 medicines, the English Product Information PDF was fetched from EMA (.../product-information/<epar_slug>-epar-product-information_en.pdf, retrieved 2026-06-06), parsing restricted to Annex I clinical prose (Annex II/III labelling, the package leaflet, and the Section 6 pharmaceutical-particulars tail are excluded). Text was chunked at a soft ~150-word window (sentence-boundary preferred, hard cap 255, min 80) and sampled per drug proportionally to clinical-prose volume, with a section weighting biased toward clinical-efficacy (§5.1), pharmacokinetics (§5.2), dosing (§4.2) and special-warnings (§4.4) content. The extraction is deterministic and reproducible (script + RUNBOOK live in the navimed-umb repository under calibration/clinical-en/).

Result shape closely matches the Polish corpus: 412 vs 418 chunks; words/chunk min 80 / median 161 / max 255 (PL median also 161); all 61 medicines covered.

Intended use / Przeznaczenie

Calibration data for post-training quantization (AWQ / AutoAWQ / GPTQ) of English and multilingual clinical LLMs, and the English side of cross-lingual calibration experiments paired with the Polish corpus. Reuse as a corpus-controlled calibration set enabling like-for-like quantization-quality comparison across model scales and languages.

This is NOT a training set and NOT an evaluation / benchmark set. It is not a question-answering, instruction-tuning, or clinical-decision-support dataset and must not be used to fine-tune clinical behaviour or to evaluate clinical accuracy.

No patient data / no PHI. SmPC documents describe medicinal products (indications, dosing, adverse reactions, pharmacokinetics, aggregate clinical-trial data), not individuals. Every chunk's text is verbatim extracted from a real fetched EMA English PI PDF — verified programmatically (412/412 chunks are a contiguous span of their source document after line-cleaning; zero fabricated or paraphrased text), and absence of PHI was confirmed by automated pattern scan. The corpus is not clinical advice and confers no clinical authority; the canonical, legally-authoritative product information remains the EMA-published SmPC.

License and source attribution

This dataset is a calibration corpus for AWQ / AutoAWQ / GPTQ quantization of clinical LLMs. It contains text chunks derived from official medicinal product information documents. The licence is `other` (EMA public-reproduction policy and source-specific reuse terms) — the same terms as the Polish corpus.

Source text

Source text is derived from EMA-published English SmPC / Product Information documents.

EMA source text: © European Medicines Agency. EMA-published documents are reproduced and distributed under EMA's content-reproduction policy, which permits reproduction and/or distribution, in whole or in part, for non-commercial and commercial purposes, provided that EMA is always acknowledged as the source.

Compilation

Compilation, drug selection, extraction workflow, chunking, dataset structuring and metadata: Łukasz Minarowski / navimed-umb.

No claim is made that the underlying SmPC source text is licensed under CC-BY-4.0, CC-BY-NC-4.0, MIT, Apache-2.0 or any other open-source software license. Users are responsible for preserving EMA source attribution and for checking source-specific restrictions before redistribution or downstream use.

Context

Built for **navimed-umb** — a local-LLM benchmarking and clinical-decision-support feasibility project (Medical University of Białystok). Cross-language counterpart to `mozarcik/clinical-pl-smpc-awq-calibration`.