A small, experimental playground with Jupyter magics to use OpenAI's GPT-3 models inside Jupyter environments.
This isn't even alpha software - it's just a quick and dirty prototype. But it works, if you have the python openai
package installed and the environment variable OPENAI_API_KEY
defined with your personal API key as per the docs.
Yes, the name is a silly play on the pronunciation of GPT and Jupyter.
pip install jupytee
To load the extension, use
%load_ext jupytee
The package provides a few core magics, %chat
, %pic
and %code
(along with some extra helpers), that wrap your interactions around the OpenAI Python API. They target the main entry points of the API: text completion, image generation and code completion, respectively.
You can see them in action in this notebook contained in the examples
directory (which you can run immediately by clicking on the Binder button above).
Briefly, these are some examples of how you can use them. A simple question (note the extra space before ?
, needed b/c IPython's help will otherwise trigger by accident):
%chat What is 100F in Celsius ?
You can control the sampling temperature with -T
:
%chat -T 1 --raw Write an uplifting short poem
Markdown (with math) works:
%%chat
Produce the LaTeX form of Maxwell's equations.
For pic
, you can control image size and the number of images returned:
%pic -s m -n 2 an astronaut knitting
When generating code, you can iteratively refine the answer by referring to the CODE
special word (in all CAPS):
%code A function to add two numbers.
and then:
%code Update CODE to have a docstring explaining its use.
BSD 3-Clause.