NINA-Qwen3-4B: Nursing Intelligent Network Assistant
NINA-Qwen3-4B is a specialized 4 billion parameter language model, fine-tuned from the Qwen/Qwen3-4B architecture. Its core purpose is to assist with nursing documentation tasks, providing a focused and safe AI tool for healthcare professionals.
Key Capabilities & Differentiators
- Strict Guardrails: NINA is engineered with absolute, non-overridable limits to ensure patient safety and regulatory compliance. It will never provide clinical diagnoses, recommend treatments or medications, alter physician orders, or impersonate medical personnel.
- Adversarial Training: The model has been rigorously trained against 22 adversarial techniques, with P1 guardrail testing before every push, enhancing its robustness against misuse.
- Specialized Fine-tuning: Trained using SFT with LoRA (r=16 RSLoRA) on approximately 6,400 balanced samples, optimizing its performance for nursing-specific language and workflows.
Ideal Use Cases
- Nursing Documentation: Primarily designed as a documentation aid for qualified registered nurses.
- Safe AI Integration: Suitable for environments requiring highly constrained and safety-critical AI applications in healthcare.
Important Disclaimer: NINA is strictly a documentation tool and is not intended for diagnostic or clinical decision-making purposes.