Modelpublic

kayodekosi/medical-clinical-decision-support-7b

sourceHugging Faceapache-2.0updated 1mo agoView on Hugging Face

0likes

Model Card

kayodekosi/medical-clinical-decision-support-7b

Production-oriented Clinical Decision Support (CDS) assistant Symptom / history → differential diagnosis, urgency triage, and guideline-aware recommendations with strong refusal & escalation behaviour.

Brief description: given a patient case, produces urgency triage, a ranked differential diagnosis, recommended next steps, and explicit red-flag/escalation notes, refusing a definitive read when the input is insufficient. A ready-to-deploy Gradio demo lives in `space/` — see the Deploying the demo Space section below.

Intended Hospital Use-Case

—Assist clinicians with rapid differential generation and triage urgency scoring.
—Surface relevant guideline snippets (when provided in context) and recommended next investigations.
—Explicitly refuse or escalate when information is insufficient or the case is high-risk.
—Always framed as decision support — never autonomous diagnosis or treatment.

Model Details

Item	Value
Base	Qwen/Qwen2.5-7B-Instruct
Method	QLoRA SFT (Unsloth, with automatic fallback to plain PEFT + TRL) + optional DPO on clinician preference / safety pairs
Adapter type	LoRA (PEFT) — load on top of the base model
Data	Public medical QA + symptom→diagnosis + triage style corpora, chat-format pairs
Safety features	Refusal on incomplete data, escalation phrases, no treatment orders without human confirmation

Training Recipe

bash

pip install -r requirements-train.txt
python train.py --config train_config.yaml
python train.py --config train_config.yaml --push_to_hub --hub_repo_id kayodekosi/medical-clinical-decision-support-7b

See `train_config.yaml` and `train.py` for the exact recipe.

Quickstart Inference

bash

pip install -r requirements.txt
python inference.py --case sample_data/example_case.txt

This pulls the base model plus the fine-tuned LoRA adapter from the Hub by default. Point --adapter at a local checkpoint directory to test a fresh training run, or pass --adapter none for base-model-only output.

Output Structure (example)

Urgency: Immediate / Urgent / Routine / Insufficient information
Differential (ranked):
1. ... (likelihood, key supporting features)
2. ...
Recommended next steps: ...
Escalation / Red flags: ...
Disclaimer: This is decision support only. Licensed clinician must confirm.

HF Path

kayodekosi/medical-clinical-decision-support-7b

Safety Statement

This model is not a medical device and must not be used for autonomous diagnosis or treatment decisions. It is intended for use by licensed healthcare professionals as an assistive tool inside a controlled hospital environment with full human oversight.

Deploying the demo Space

`space/` is a complete, self-contained Gradio Space (app.py, requirements.txt, README.md with the Space metadata front matter). Deploy it with the portfolio-level script:

bash

pip install -r ../../scripts/requirements.txt
huggingface-cli login
python ../../scripts/deploy_spaces.py --only clinical-decision-support

This creates kayodekosi/medical-clinical-decision-support-demo if it doesn't exist and uploads the whole space/ folder — nothing to copy by hand.

Repository Layout

03-clinical-decision-support/
├── README.md                # this file / HF model card
├── train_config.yaml
├── train.py
├── inference.py
├── requirements.txt
├── requirements-train.txt
├── requirements-serving.txt
├── sample_data/example_case.txt
└── space/                    # Gradio demo Space (app.py + requirements.txt + README.md)