LLM Magics for Jupyter Notebooks

Interact seamlessly with Large Language Models (LLMs) like OpenAI's GPT models directly from your Jupyter Notebook using IPython magics.

https://pypi.org/project/llm-magics/

pip install llm-magics

Features

• Interactive Chat Interface: Engage in a conversational exchange with LLMs within your notebook cells.
• Customizable Models: Switch between different OpenAI chat models (e.g., gpt-3.5-turbo, gpt-4).
• Set System Messages: Define system prompts to guide the behavior of the LLM.
• Persistent Chat History: Maintain context across multiple interactions.
• Rich Rendering: Receive responses with proper formatting, including syntax-highlighted code blocks.
• Easy Clearing: Reset the conversation when needed.

Installation

Prerequisites

• Python 3.10 or higher
• Jupyter Notebook or JupyterLab
• An OpenAI API key

Steps

1. Clone the Repository

   git clone https://github.com/kmkolasinski/ipython-llm-magics
   cd llm-magics
   

2. Install Dependencies

   Install the required Python packages:

   pip install -r requirements.txt
   

3. Install the Package

   pip install .
   

Configuration

Setting the OpenAI API Key

Before using the magics, ensure your OpenAI API key is accessible to the application.

Option 1: Environment Variable

Set the OPENAI_API_KEY environment variable in your shell:

export OPENAI_API_KEY='your-api-key-here'

Option 2: Within the Notebook

Alternatively, set the API key within your notebook:

import os
os.environ['OPENAI_API_KEY'] = 'your-api-key-here'

Usage

Loading the Extension

In your Jupyter Notebook, load the llm_magics extension:

%load_ext llm_magics

Setting the Model

Specify the OpenAI model you want to use:

%llm_set_model gpt-4o

Available models include gpt-3.5-turbo, gpt-4o, etc.

Setting a System Message (Optional)

Define a system message to guide the assistant's behavior:

%llm_set_system_message "You are a helpful assistant."

Starting a Conversation

Use the %%llm_chat cell magic to send a message to the LLM:

%%llm_chat

Write a Python function that generates a random integer between 1 and 100.

Response:

The assistant will provide a Python function as per your request.

Continuing the Conversation

Maintain context across multiple %%llm_chat cells:

%%llm_chat

Now modify the function to generate a random floating-point number between 0 and 1.

Response:

The assistant will update the function accordingly.

Inserting local variables into the chat

not_sorted_list = [3, 1, 4, 1, 5, 9, 2, 6, 5, 3, 5]
expected_output = sorted(not_sorted_list)

%%llm_chat

Write python code to sort the list in the descending order using merge sort algorithm.
So that I can write:
input = $not_sorted_list
assert $expected_output == my_merge_sort(input)

Clearing the Conversation History

Reset the chat history when needed:

%llm_clear

Rendering and Syntax Highlighting

The extension includes rich rendering of the LLM's responses:

• Code blocks are syntax-highlighted using Prism.js and Highlight.js.
• Copy buttons are added to code blocks for convenience.
• Markdown formatting is preserved in the responses.

History

• v1.1.0: Initial release with basic chat functionality and rendering.