NurseCitizenDeveloper/nursing-llama-3-8b-fons

🏥 Relational Ai for Nursing: Nursing Llama-3 (8B)

The first open-source LLM fine-tuned on Foundation of Nursing Studies (FONS) literature for person-centred, equitable clinical documentation.

🔗 Live Demo: opennursingcoreig.com/relation-ai.html

---

📄 Model Description

Relational Ai for Nursing is a fine-tuned version of Llama-3 (8B) designed specifically to assist nursing professionals with documentation, care planning, and clinical reasoning. Unlike general-purpose models, Relational Ai for Nursing has been trained on 6,698 instruction-response pairs derived from open-access nursing practice development literature, ensuring that its outputs align with the FONS Principles of person-centredness, relational care, and practice development.

• Developed by: NurseCitizenDeveloper / Open Nursing Core Team
• Base Model: unsloth/llama-3-8b-bnb-4bit
• Fine-Tuning Method: LoRA (Low-Rank Adaptation) via Unsloth
• Language: English
• License: CC BY-NC 3.0

📜 Dataset & Licensing

This model was fine-tuned on data derived from the International Practice Development Journal (IPDJ) and other FONS publications, which are licensed under CC BY-NC 3.0.

• Attribution: Foundation of Nursing Studies (FONS)
• Non-Commercial: This model should be used for research and educational purposes only.

🌟 Key Features

• Person-Centred Language: Generates documentation that respects patient dignity and preferences.
• Health Equity Focus: Specifically trained to include skin tone documentation in pressure ulcer risk assessments (Braden Scale), addressing a critical gap in standard AI models.
• Structured Nursing Process: Follows the ADPIE key framework (Assessment, Diagnosis, Planning, Implementation, Evaluation).
• FONS Alignment: Prioritizes relational care and checking for understanding over generic medical jargon.

📊 Evaluation Results

We evaluated Relational Ai for Nursing using Azure GPT-4o typically as an "Expert Judge" across clinically relevant test cases.

Metric	Score (1-10)	Description
Clinical Accuracy	6.6	Clinically sound interventions and assessments
Person-Centred Language	7.6	High degree of respect, dignity, and personalization
FONS Alignment	6.0	Good adherence to relational care principles

🏆 Spotlight Performance

• Equity / Skin Tone Assessment:

  • Accuracy: 8/10
  • Person-Centredness: 9/10
  • The model successfully identifies the importance of documenting skin tone nuances often missed by generic models.

• Nursing Process (ADPIE):

  • Accuracy: 9/10
  • Strong capability in structuring clinical reasoning.

💻 How to Use

Installation

Relational Ai for Nursing is optimized with unsloth for faster inference but supports standard Hugging Face transformers.

pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
pip install --no-deps xformers trl peft accelerate bitsandbytes

Inference Code

from unsloth import FastLanguageModel

# 1. Load Model
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "NurseCitizenDeveloper/nursing-llama-3-8b-fons",
    max_seq_length = 2048,
    dtype = None,
    load_in_4bit = True,
)
FastLanguageModel.for_inference(model)

# 2. Define Prompt (Alpaca Format)
alpaca_prompt = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
{}

### Input:
{}

### Response:
{}"""

# 3. Generate
inputs = tokenizer(
[
    alpaca_prompt.format(
        "Summarize the key nursing interventions for a patient with delirium.", # Instruction
        "Patient is an 85-year-old male presenting with acute confusion...", # Input
        "", # Leave blank for generation
    )
], return_tensors = "pt").to("cuda")

outputs = model.generate(**inputs, max_new_tokens = 128)
print(tokenizer.batch_decode(outputs)[0])

⚠️ Limitations & Ethical Considerations

• Clinical Assistant, Not Replacement: This model is a support tool. All outputs must be verified by a registered nurse.
• Training Data: While focused on FONS literature, the model may still hallucinate facts or reflect biases present in the base Llama-3 model.
• Scope: Optimized for UK/NHS context but applicable broadly.

📚 Citation

If you use this model in your research or practice, please cite:

@misc{fons_ai_2025,
  author = {Lincoln Gombedza},
  title = {Relational Ai for Nursing: A Person-Centred Nursing Documentation Model},
  year = {2025},
  publisher = {Hugging Face},
  journal = {Hugging Face Hub},
  howpublished = {\url{https://huggingface.co/NurseCitizenDeveloper/nursing-llama-3-8b-fons}}
}

---

Created with ❤️ by the Open Nursing Core Team.