README.md

metadata

license: cc-by-nc-3.0
datasets:
  - custom
language:
  - en
base_model: google/gemma-2-9b-it
pipeline_tag: text-generation
library_name: peft
tags:
  - nursing
  - healthcare
  - person-centred-care
  - gemma2
  - lora
  - medical

Relational Intelligence - Gemma 2 9B for Nursing

A fine-tuned Gemma 2 9B model specialised for person-centred nursing care and relational practice.

Model Details

Model Description

Relational Intelligence is a LoRA adapter fine-tuned on nursing care literature, designed to support nurses in understanding and applying person-centred care principles. Trained on open access journal articles from:

• Foundation of Nursing Studies (FONS) - https://www.fons.org/
• International Practice Development Journal (IPDJ) - ISSN: 2046-9292

This model respects the CC BY-NC 3.0 license of the source materials.

• Developed by: Nursing Citizen Development
• Funded by: Self-funded research project
• Shared by: NurseCitizenDeveloper
• Model type: LoRA Adapter (Causal Language Model)
• Language(s) (NLP): English
• License: Creative Commons Attribution Non-Commercial 3.0-
• Finetuned from model: google/gemma-2-9b-it

Model Sources

• Repository: NurseCitizenDeveloper/relational-intelligence-gemma2-9b
• Base Model: google/gemma-2-9b-it

Uses

Direct Use

• Educational support for nursing students and practitioners
• Care planning inspiration and person-centred approaches
• Reflective practice guidance and prompts
• Understanding nursing theoretical frameworks
• Research assistance in nursing literature

Downstream Use

Can be integrated into:

• Clinical decision support tools
• Nursing education platforms
• Care planning applications
• Reflective practice journals

Out-of-Scope Use

⚠️ NOT for:

• Clinical diagnosis
• Prescribing medications or treatments
• Emergency medical decisions
• Replacing professional clinical judgement
• Direct patient care without human oversight

Bias, Risks, and Limitations

• Not a diagnostic tool - Cannot diagnose or recommend treatments
• Knowledge limitations - May not reflect latest evidence
• Potential hallucinations - May generate incorrect information
• Cultural context - Primarily trained on English Western nursing literature
• No patient context - Cannot examine patients

Recommendations

• Always verify against current clinical guidelines
• Use as supportive tool, not replacement for expertise
• Never input identifiable patient information
• Follow your organisation's AI policies

How to Get Started with the Model

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "google/gemma-2-9b-it",
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

# Load LoRA adapter
model = PeftModel.from_pretrained(
    base_model,
    "NurseCitizenDeveloper/relational-intelligence-gemma2-9b"
)

tokenizer = AutoTokenizer.from_pretrained("google/gemma-2-9b-it")

# Generate
prompt = "<start_of_turn>user\nWhat are the key principles of person-centred care?<end_of_turn>\n<start_of_turn>model\n"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=500)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training Details

Training Data

• Source: FONS, IPDJ, person-centred care literature
• Size: 15,234 Q&A pairs
• Topics: Person-centred care, therapeutic relationships, reflective practice

Training Procedure

Preprocessing

Q&A pairs generated from nursing literature using Gemini 2.5 Flash, formatted in Gemma chat template.

Training Hyperparameters

• Training regime: bf16 mixed precision, QLoRA (4-bit)
• LoRA rank: 64
• LoRA alpha: 16
• Learning rate: 2e-4
• Batch size: 1 (gradient accumulation: 4)
• Epochs: 3
• Max sequence length: 512
• Optimizer: paged_adamw_8bit

Speeds, Sizes, Times

• Training time: ~11 hours
• Trainable parameters: 216M (2.28%)
• Total parameters: 9.4B

Environmental Impact

• Hardware Type: NVIDIA A100 GPU (40GB)
• Hours used: 11
• Cloud Provider: Google Colab
• Compute Region: US

Technical Specifications

Model Architecture and Objective

• Architecture: Gemma 2 9B decoder-only transformer
• Adapter: LoRA (Low-Rank Adaptation)
• Objective: Causal language modeling

Compute Infrastructure

Hardware

NVIDIA A100 GPU, 40GB VRAM

Software

• transformers >= 4.42.0
• peft >= 0.11.0
• trl >= 0.8.6
• bitsandbytes >= 0.43.0

Acknowledgments & Attribution

This model was trained on content from:

Foundation of Nursing Studies (FONS)

• Website: https://www.fons.org/
• The Foundation of Nursing Studies supports nurses and nursing to improve care

International Practice Development Journal (IPDJ)

• Website: https://www.fons.org/library/journal/
• An open access journal published under CC BY-NC 3.0 license
• ISSN: 2046-9292

We gratefully acknowledge FONS for making nursing research openly accessible to support evidence-based practice and education.

License Compliance

The training data derives from open access publications licensed under Creative Commons Attribution Non-Commercial 3.0 (CC BY-NC 3.0).

Accordingly, this model and its outputs:

• ✅ May be used for educational purposes
• ✅ May be used for research
• ✅ May be adapted with attribution
• ❌ May NOT be used for commercial purposes

Citation

BibTeX:

@misc{relational-intelligence-2025,
  author = {Lincoln Gombedza},
  title = {Relational Intelligence: A Fine-Tuned Gemma 2 Model for Nursing},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/NurseCitizenDeveloper/relational-intelligence-gemma2-9b}
}

APA:

Lincoln Gombedza. (2025). Relational Intelligence: A Fine-Tuned Gemma 2 Model for Nursing. Hugging Face. https://huggingface.co/NurseCitizenDeveloper/relational-intelligence-gemma2-9b

Model Card Authors

Lincoln Gombedza

Model Card Contact

• HuggingFace: @NurseCitizenDeveloper
• Email: info@nursingcitizendevelopment.com