Alpha
Contact
Implement · MeditronFO

Can you run it?

Ownership levelPartialnone·limited·partial·substantial·fullAnalytical input D ยท 51.2/100

This page is a projection of the one entry record, the Reliability and Data control factors that Implement covers. The full verdict is set by all four factors together, floor-weighted so the weakest caps the whole.

Which domain expands which factor
  • AssessUse & modify + Transparency
  • ImplementData control + Reliability
  • UseReliability
  • SupportTransparency

Install & run

The checkpoints are standard safetensors on Hugging Face under EPFLiGHT. The Apertus architecture needs transformers 4.56 or later (ev-base-card). Pin the commit, because the weights on main were replaced twice in autumn 2026 (ev-hf-meta).

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "EPFLiGHT/Apertus-70B-MeditronFO"   # or Apertus-8B-MeditronFO
rev = "d13151ac9ff56b966dd5257d5064876685a750e3"  # 70B revision; do not float on main
tok = AutoTokenizer.from_pretrained(model_id, revision=rev)
model = AutoModelForCausalLM.from_pretrained(
    model_id, revision=rev, torch_dtype=torch.bfloat16, device_map="auto")

The card's own example uses sampling with temperature 0.7 and max_new_tokens=8192, and the repo's generation_config.json sets the same defaults (ev-card-70b, ev-hf-meta). Verify the per-shard sha256 against the Hugging Face file record after download. The code repository (ev-github-readme) is a separate Apache-2.0 project for reproducing the paper and is not an inference package.

Hardware & VRAM requirements

The publisher gives training compute only (70B: 16 nodes of 4 GH200 for 3 h 06 min, 199 GPU-hours) and no serving guidance (ev-card-70b). Sizes come from the file listing (ev-hf-meta):

CheckpointParametersBF16 weight sizePractical minimum (OneHill estimate, untested)
Apertus-70B-MeditronFO70.6B in 30 shardsabout 141 GBtwo to four 80 GB GPUs for BF16 plus KV cache, or a quantised build
Apertus-8B-MeditronFO8.05B in 4 shardsabout 16 GBone 24 GB GPU with short contexts

KV-cache memory grows with context. The fine-tuning used 8,192-token sequences, so plan around that length.

Serving stacks

The MeditronFO cards document only transformers. The Apertus base card lists transformers, vLLM, SGLang and MLX as supported (ev-base-card). That is a statement about the base architecture and not a test of these fine-tunes. Community GGUF conversions (llama.cpp) and an 8B MLX conversion exist (ev-quants). OneHill has not run any of them.

Safe-deployment controls & Deployment Ceiling

Deployment Ceiling: T1 (internal assistant), conditional, and only for non-clinical use. Clinical decision use is outside any ceiling because the publisher prohibits it (ev-project-page). T2 and above are not recommended: no guard model, no independent safety evaluation and low-agreement safety judging mean the playbook's model layer is missing.

Controls required to reach T1 on non-patient work:

  1. Pin the commit and verify shard hashes (supply-chain layer).
  2. Keep a qualified person reading every output, and feed no output into any patient pathway.
  3. Run a clinical safety evaluation and a red-team for the use case. The publisher says its judge is unsuitable as a deployment-readiness signal for safety criteria (ev-paper) and recommends independent safety evaluation (ev-data-card).
  4. Add input and output guardrails and a separate guard or classifier model; none ships with the release.
  5. Constrain answer length, because answers run longer than the base model's (ev-project-page).
  6. Resolve the licence chain with counsel before any commercial use (see the licence item in assess), and consider the Apertus acceptable-use and personal-data output-filter advice (ev-base-card).
  7. If the product could influence a clinical decision, treat medical-device and AI Act high-risk assessment as a precondition to development.

Available quantizations

There is no official quantisation. Community GGUF and i1-GGUF conversions of both sizes (mradermacher), an MLX conversion of the 8B (loleg) and Q4 and Q5 GGUF of the 8B (BrazosDeDios) are listed (ev-quants). None is opened, and quantisation can change clinical answer quality in ways nobody has measured for this model. Prefer the canonical safetensors for anything under evaluation.

Fine-tuning & adaptation

Fine-tuning is the strongest part of the release. The GitHub repository publishes the Axolotl training configs for every reported model, the data-generation, resampling and decontamination scripts, and the evaluation harness fork. The recipe is published: 70B, 1 epoch, learning rate 1e-5, cosine decay, 10 percent warmup, effective batch 64 packed sequences of 8,192 tokens, AdamW, seed 42, DeepSpeed ZeRO-3, bf16 (ev-github-readme, ev-card-70b).

The corpus loads with datasets.load_dataset("EPFLiGHT/fully-open-meditron") and each row carries its source dataset, gold label and teacher (ev-data-card). The README warns that Axolotl silently reuses a stale tokenised cache unless each config has its own dataset_prepared_path. The paper reports an optional 10 percent Tulu 3 replay that recovers most of the general-capability loss (ev-paper). The intermediate data and per-item evaluation outputs are not yet released, so a full reproduction from a fresh clone is not possible (ev-github-readme).

API / OpenAI-compatible integration

No MeditronFO-specific API documentation exists. A serving stack that supports the Apertus architecture (the base card names vLLM and SGLang) can expose an OpenAI-compatible endpoint, but OneHill has not tested that for these checkpoints (ev-base-card). The chat template is bundled in the repository (ev-hf-meta).

How this scores

The ownership factors this domain covers, drawn from the one entry record.

3

ReliabilityIs it reliable and good enough for the job?

Weak

Safety is 2: no safety tuning, publisher-only low-agreement safety judging and no guard or red-team, and the publisher says not to use it for patient decisions, so it is not safe to deploy as presented.

How this scores (AOI sub-dimensions)
Operational3/5how practical it is to run, serve and maintain in productionStandard transformers architecture class supported since transformers 4.56 per the base card, BF16 safetensors, a chat template and recommended sampling settings in generation_config, and community GGUF and MLX conversions for both sizes.
Safety2/5whether misuse risks are evaluated and guardrails are providedThe models are supervised fine-tunes for medical accuracy and not safety tuning.
4

Doesn't extract your dataDoes running it keep your knowledge and data yours?

Strong

Self-hosted BF16 safetensors run on the deployer's own hardware with no telemetry or licence clawback in the declared Apache-2.0 grant.

How this scores
Not a scored AOI dimension. For a self-hosted model, data-control is a structural property of running the weights yourself, strong by default unless the model phones home or the licence claws back rights. For a hosted API this factor is the retention + train-on-inputs + residency read, scored from the binding terms.
What this means for adoptionOwnership is partial. The release is an inspectable medical fine-tune that can be run, adapted and audited on the deployer's own hardware, and the pipeline behind it is public. It is not a clinically usable or clearly licensed product: the publisher bars patient decisions, there is no LICENSE file, the corpus licence is stated two ways, and safety evaluation is thin. Ownership moves to substantial if EPFL ships a LICENSE file and reconciles the corpus licence, and a deployer adds an independent clinical safety evaluation.

Sources

The same evidence records as the entry sheet. Read means the text was verified; unverified means it is known to exist but not yet read.

Model cardread2026-11-19
The Apertus-70B-MeditronFO model card declares licence tag apache-2.0 and an Apache 2.0 link in the header.
Model cardread2026-11-19
The Apertus-8B-MeditronFO model card (partial extracts) has the same template and licence tag, base swiss-ai/Apertus-8B-Instruct-2509, a five-benchmark average of 48.26 against 42.39 for the base, and preference over the base in 74.2 percent of AutoMOOVE comparisons.
Documentationread2026-11-19
Hugging Face records show both repositories are ungated, safetensors only (30 shards for 70B, 4 for 8B, each with an LFS sha256) and without a LICENSE file (raw LICENSE is HTTP 404).
Licenceread2026-11-19
The GitHub code repository EPFLiGHT/FullyOpenMeditron carries an Apache License 2.0 LICENSE file added 2026-08-03.
Documentationread2026-11-19
The project page states the MeditronFO models are not medical devices, have not been tested in clinical practice and must not be used to make decisions about a patient.
Documentationread2026-11-19
The GitHub README says the repository holds the data pipeline, training configs and evaluation to reproduce the paper.
Documentationread2026-11-19
The dataset card gives an Apache-2.0 tag, 533,888 rows in four components and every answer by gpt-oss-120b.
Technical_reportread2026-11-19
arXiv 2605.16215 v2 (29 May 2026) says the corpus is released under a research-use licence and that assets are restricted to research purposes.
Model cardread2026-11-19
The Apertus-70B-Instruct-2509 base card (the 8B card is identical apart from the name) declares licence tag apache-2.0.
Model cardread2026-11-19
gpt-oss-120b, the teacher that wrote every training answer, is Apache-2.0 with a one-sentence usage policy requiring compliance with applicable law.
Vendor announcementread2026-11-19
The EPFL AI Center news item of 17 June 2026 describes MeditronFO as the first fully open framework for medical LLMs and says every stage of development is publicly available.
Vendor announcementread2026-11-19
The Meditron website gives a family history from Meditron 1 (2023) to MeditronFO (2026), a contact address for the lab, and a chat disclaimer saying it is a research model, not a doctor, that can be wrong.
Documentationread2026-11-19
The Hugging Face listing of the EPFLiGHT organisation shows five MeditronFO checkpoints with the licence tag apache-2.0 (Apertus 70B and 8B, OLMo-2-32B, EuroLLM-22B and 9B).
Third-party analysisread2026-11-19
A Hugging Face search for MeditronFO lists community GGUF and i1-GGUF conversions of both Apertus sizes by mradermacher, an MLX conversion of the 8B by loleg and Q4 and Q5 GGUF of the 8B by BrazosDeDios.
Documentationunverified2026-11-19
The Apertus card links an EU AI Act public summary and a Code of Practice document for the base model.