GPT experiments on Jupyter/Colab noteooks

These are simple experiements to call OpenAI's GPT APIs via Python notebooks.

The notebooks can be run on Jypter or on Google Colab.

Med GPT

The first notebook med_gpt_notebook walks through how to configure and use GPT APIs. Right now it is using gpt-3.5-turbo model but can be easily extended to use gpt-4 apis when they are released.

We are using ChatCompletion API for the demo, which takes a prompt and returns completions.

We have designed the system to accept an input context(system context) in the config variable DEFAULT_CONTEXT_MESSAGE which allows for the GPT to respond with relevant the context.

Parameters to fine tune

We need to experiment with the responses of GPT by finetuning the parameters we pass to GPT.

These primary parameters:

• MAX_TOKENS
• TEMPERATURE
• NUMBER_OF_CHOICES

[]

How to run

You can run this on Jupyter or Google Colab.

Using Jupyter on local / command line:

• $ jupyter notebook
• navigate to med_gpt_notebook
• Replace OPENAI_API_KEY ='##ADD_YOUR_KEY_HERE##' with your OpenAPI key. If you don't have one - signup here: platform.openai.com
• Run

You should see an output like this:

[]

Play with different prompts / context and have fun!