Models · M/01
Mira-Q2 · live on Hugging FaceDilr Mira is a class of private clinical small language models (~3B parameters, built on Qwen2.5-3B-Instruct) that turn scans, lab reports and claim forms into source-grounded, schema-valid JSON on your own hardware. Mira-Q2 is open on Hugging Face under Apache-2.0, runs on CPU and takes about 2 GB on disk. The data never leaves the building.
Updated June 2026 · Latest release: Mira-Q2
DISCHARGE: pt c/o SOB x3d, h/o HTN + DM Hb 9.2 g/dL (L) Na 138 Creat 1.4 denies chest pain, no known allergies on T. Metf 500 BD PO, Amlo 5 OD BP 160/90 triage → 142/88 by phy
M/02
LiveApache-2.0| Eval set | N | Type | JSON validity | Score |
|---|---|---|---|---|
| test_gold | 200 | Held-out, training distribution | 100.0% [1.0–1.0] | |
| synthetic_v2 | 150 | Different formatting dialect | 100.0% [1.0–1.0] | |
| extraction_relevant | 150 | Real physician docs, on-schema | 94.7% [90.7–98.0] | |
| mtsamples | 282 | Real physician docs, 39 specialties | 85.8% [81.9–89.7] |
100% on training-distribution data. 86% on general real physician prose. That gap is published, not hidden, and closing it is exactly what the next generation is for.
View the model, eval files and full scorecard on Hugging Face ↗
M/03
In developmentHinglish and code-switched clinical text first, then Spanish and Portuguese: the documents English-only medical APIs can’t read.
Point Mira at any JSON schema and get exactly those fields back, built for India’s NHCX/FHIR claim profiles, no per-template setup.
Identifiers detected and replaced before the model reads the document, with a reversible vault. The model never sees a name.
Plus a smaller routed sibling for CPU-only fleets.
Trust pack
In developmentResearch directions
Mira’s architecture is schema-agnostic: the clinical model is generation one of a method, not a one-off. The same verifier-gated extraction is under research for other regulated document worlds.
Research directions. Clinical extraction is what’s shipped today.
Deployment · local-first
Pull the open model from Hugging Face. Apache-2.0, ~2 GB on disk in 4-bit.
One Python call. CPU is enough; a GPU just makes it faster.
Paste a document, receive schema-valid JSON.
No cluster, no GPU requirement, no API key, no per-page meter. The full quickstart lives on the model page.
# 01 · download + install
$ pip install transformers
# 02 · load, one call
$ python
>>> from transformers import pipeline
>>> mira = pipeline("text-generation", model="dilr/Mira-Q2")
# 03 · extract
>>> mira(open("discharge_note.txt").read())
{"document_type": "discharge_summary", ...}Governance · audit-ready by design
Mira is built for teams whose auditors read the logs. Nothing ships on model output alone.
Governance is a dilr.ai discipline, not a feature. see how we build audit-ready operating models →
Quick answers
Direct answers to the questions clinical, data and compliance teams ask about Dilr Mira. Every number comes from the published Mira-Q2 evaluations.
Dilr Mira is the best private small language model for clinical document extraction on your own hardware: among the extraction tools DILR compares, it is the only one that runs fully on your own hardware, and the claim rests on published evaluations over 782 clinical documents. Mira-Q2, the current release, is a ~3B model built on Qwen2.5-3B-Instruct, open on Hugging Face where anyone can verify the numbers.
Yes: turning messy clinician shorthand into source-grounded, schema-valid JSON is exactly what the demo at the top of this page shows. In published evaluations, Mira-Q2 held 100% JSON validity when document formatting shifted (150 documents) and stayed on-schema on 94.7% of real physician documents. DILR does not publish general-purpose OCR benchmarks; the evaluations cover clinical text extraction.
Mira-Q2 is free to download and use commercially: it is an open model on Hugging Face under the Apache-2.0 license. It runs on CPU and takes about 2 GB on disk at 4-bit, so inference costs only the hardware you already own, with no per-page fees, API meter or egress. Private, schema-matched deployments for custom document types are licensed per engagement.
Dilr Mira is built for HIPAA-aligned deployments: it runs entirely on your own hardware, on-premises or air-gapped, so no document, field or identifier is sent to any third-party API. Across all 782 evaluated clinical documents, the deterministic verifier recorded 0 identifier leaks. Teams tracking wider obligations, including EU AI Act duties arriving in August 2026, can follow the UK and EU AI compliance changelog.
Mira-Q2's published results show an honest gradient rather than one headline number: 100% JSON validity and field-F1 of 1.0 on held-out gold (200 documents), 94.7% on-schema on real physician documents (150), and 85.8% across 39 specialties (282 documents). Its vocabulary is grounded in real ICD-10 codes, NLM drug names and curated lab ranges, and every figure ships with the model on Hugging Face.
Try the open model on Hugging Face, or talk to us about a private, schema-matched deployment for your claims and clinical documents.
FAQ
Dilr Mira is a class of small, specialised language models (~3B parameters) that read clinical documents (lab reports, discharge summaries, claim forms) and return source-grounded, schema-valid JSON. The current release, Mira-Q2, is open on Hugging Face under Apache-2.0.
No. Mira runs entirely on your own hardware: on-premises, even air-gapped, even CPU-only. No document, field or identifier is ever sent to a third-party API.
The model is free and open (Apache-2.0). Running it costs only the hardware you already own: no per-page fees, no API meter, no egress. Private, schema-matched deployments are priced per engagement.
Mira-Q2 is strongest on the types it was trained on: lab reports, medication lists, discharge summaries, pathology reports, intake forms and progress notes, in English. Broader document types and languages are in development for Mira-3.
No. Every output is a draft for human review, gated by a deterministic verifier that checks schema validity, source grounding and identifier leakage. Mira is an extraction tool, not a medical device.