problem-list-generator is an experimental local LLM trained to generate a list of active clinical problems from a History and Physical (H&P) section of a clinical note.

This is part of a small personal experiment to train small, fast, local language models that can assist clinicians when writing medical notes. It can be run completely offline, without any data leaving your local machine.

For more information on trying this out, please see the corresponding GitHub repository.

Technical Details

This model was finetuned using synthetic clinical data generated with Llama 3.1 405b. The model itself is a LORA finetune of Meta-Llama-3.1-8B-Instruct.

This repository contains a basic Python script to inference the model via Ollama.

Setup

Clone this repository
Download the GGUF model from here and place in same directory as the repo.
Add the model to Ollama: ollama create problemlist -f ProblemList.Modelfile
Run the Python script: python problemlist.py