llama3-medical-3b-4bit

Llama 3.2 3B Instruct fine-tuned for first-aid question answering, quantized to 4-bit with bitsandbytes. I built this as the model component of kit-ai, an offline-first emergency first-aid PWA.

Training

• Base model: meta-llama/Llama-3.2-3B-Instruct
• Method: [FILL IN: LoRA/QLoRA? framework? rough training setup]
• Data: [FILL IN: what the fine-tuning data was and how you built it]
• Quantization: 4-bit via bitsandbytes

Try it

The model runs in the offline-medical-assistant Space.

Limitations

This is not a medical device. It must not be used for diagnosis or treatment decisions, and in an emergency you should call your local emergency number. Outputs can be wrong or incomplete. The kit-ai app grounds its answers in a vetted first-aid knowledge base rather than trusting model output alone.

Evaluation

No formal evaluation yet. An eval harness comparing model answers against the vetted first-aid guidelines used by kit-ai is planned.