BioJev
Research framework for biomedical decision models on Qwen3.5, studying domain pretraining, NLI training, calibration and abstention.
Overview
BioJev is a biomedical decision-model research framework built on the Qwen3.5 family at LITIS, Université de Rouen Normandie. It studies how three kinds of adaptation contribute to biomedical decisions: domain-adaptive pretraining on PubMed and PMC, general decision training on natural-language inference datasets, and biomedical decision specialization on biomedical and clinical NLI datasets.
Beyond accuracy, it looks at cross-task transfer, calibration, confidence, selective prediction and component ablations, and is built as a reproducible pipeline rather than a single checkpoint. Public checkpoints include BioJev-Nano for ablations and BioJev-4B for the main evaluation, using a three-class sequence-classification head. It is evaluated on five biomedical datasets against baselines including OpenJev-style models, and exposes a System One API.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Does the premise entail, contradict or say nothing about the hypothesis? (biomedical NLI)(inferred) | entailment · contradiction · neutral |
How it works
- Domain-adaptive pretraining on PubMed/PMC
- General NLI decision training
- Biomedical NLI specialization
- Evaluate accuracy, calibration and selective prediction
- Run ablations
Strengths
- Reproducible pipeline
- Studies calibration and selective prediction
- Public checkpoints
Caveats
- Research code
- Three-class head limits question types
Getting started
Download BioJev-4B or BioJev-Nano from Hugging Face (Gabriel382).At a glance
- Type
- Benchmark
- Status
- research
- Language
- Python
- Runtime
- training pipeline + System One API
- Has code
- yes
Use cases and integrations
How curated lists describe it
BioJev is a biomedical decision-model benchmark and adaptation framework built…
wh000wh000/awesome-jev-live
Frequently asked questions
What is BioJev?
A research framework and models for biomedical decisions built on Qwen3.5.
Which checkpoints exist?
BioJev-Nano and BioJev-4B on Hugging Face.
Is it clinical software?
No, it is a research project.
Related projects
Evaluation suite for decision models with 23,900 pinned rows across nine task families and three primitives, plus a leaderboard.
ProjectbetaPythonJev-compatible open System One engine with a Claude Code-style terminal REPL showing live probabilities, plus an experiment harness.
ProjectprototypePythonLlamaIndex notebook comparing Jev with open decoders, encoders and specialised tools on document-pipeline Choice tasks.
ProjectresearchPythonPre-registered study testing Jev's calibration claims on clinical diagnosis cases against Claude Opus 5 and Opus 5.5.
ProjectresearchPythonBenchmark harness comparing Jev with OpenRouter models and open Jev-shaped models on accuracy, calibration, latency and cost.
ProjectresearchPythonEvaluation of Jev versus DeepSeek and Qwen on healthcare decision tasks: triage, diagnosis, medication, nursing and trial screening.
ProjectresearchPython