language-model-notebooks

Quickstart for interacting with language models and APIs via a notebook-like interface

George Box: "All models are wrong, but some are useful."

Quickstart

Look at the demo notebook in a colab:

Or in a browser: getting-started.ipynb

Installing prerequisities

python3 -m venv .venv 
source .venv/bin/activate
pip install -r requirements.txt

Then after activating (source .venv/bin/activate) run jupyter lab jupyter lab. You will need to make sure to activate the virtual environment every time you restart your computer or shell session.

Running the notebooks

After installing the prerequisites, you can run the notebooks by running the following command:

jupyter lab

If a browser window doesn't open automatically, you can navigate to http://localhost:8888/lab in your browser and follow the instructions there.

Testing different closed-source language models via a third-party API

You'll need to add a file called .env to this repository, with the following content:

PROVIDER_API_KEY=your-api-key

Replacing your-api-key with your actual OpenAI/Anthropic/Microsoft API key.

Applications that are helpful for working with large language models

On a Mac, you can install the following applications to make working with large language models easier:

• Maccy: clipboard manager (https://maccy.app/): brew install maccy
• Rectangle: window manager (https://rectangleapp.com/): brew install rectangle
• Amethyst: tiling window manager (https://ianyh.com/amethyst/): brew install amethyst

There are equivalents on Windows, that can be installed with chocolatey.org, such as copyq: choco install copyq.

Enabling word wrapping in Jupyter Lab

Inside a virtual environment, run the following command:

which jupyter-lab

This will give you a path:

/Users/me/projects/language-model-notebooks/.venv/bin/jupyter-lab

Alternatively:

jupyter --path

You can then edit the settings file:

code ~/.jupyter/lab/user-settings/@jupyterlab/notebook-extension/tracker.jupyterlab-settings

And add the following to make sure line wrapping and code cell wrapping is on:

{
  "MarkdownCell": {
    "cm_config": {
      "lineWrapping": true
    }
  },
  "CodeCell": {
    "cm_config": {
      "lineWrapping": true
    }
  }
}

Enabling GitHub Copilot support in Jupyter Lab

• TODO