openai-tutorialslab

Biomedical Informatics lab tutorials for how to use the OpenAI API.

Project Structure

openai-tutorialslab
├── .env # contains your OPENAI_API_KEY
├── Dockerfile # recipe for latest JupyterLab datascience-notebook and openai 
├── NHLBI-BDC-openai-logprobs.ipynb
└── README.md

The .env file should look like this:

OPENAI_API_KEY=sk-letslearntheopenaiAPI

Using the openAI Python library, we can connect to the client using this snippet:

from openai import OpenAI
from dotenv import load_dotenv

load_dotenv()
client = OpenAI()

Tutorials

National Heart Lung Blood Institute (NHLBI) BioData Catalyst (BDC)

NHLBI-BDC-openai-logprobs.ipynb

1. Classify cardiac research article abstracts into a set of predefined categories and evaluate the log probability. Demonstrates chat prompt engineering using chat.completions.create and model=gpt-3.5-turbo with the logprobs and top_logprobs parameters set for evaluating the confidence of the model's predictions.
2. Retrieve confidence scoring for congenital heart disease prompt questions using an article's abstract result section to reduce hallucinations.
3. Autocomplete each predicted token in a sentence using logprobs to assess how confident a model is in its prediction for the top_logprobs=3.
4. Highlight tokens using the logprobs and bytes parameter. Evaluate the logprobs for a token and ASCII.

Install

sudo docker pull apblair/openai-tutorialslab:nhlbi-bdc

Next run the JupyterLab container, open a browser to localhost:8888, and copy/paste the paste the JupyerLab password.

sudo docker run it -p 8888:8888 -v $(pwd):/home/jovyan/work apblair/openai-tutorialslab:nhlbi-bdc