Build an LLM from Scratch 1: Set up your code environment

Introduction to the Coding Along Video Series

Overview of the Video Series

• Sebastian Raschka introduces himself as the author of "Build a Large Language Model (from Scratch)" and explains that he will create supplementary videos for each chapter.

• The videos aim to provide a casual yet interesting perspective on code examples, complementing the book's content.

Focus of Chapter One

• Chapter one lacks code examples; thus, this video focuses on setting up a Python environment for running code from chapters two to seven.

• Raschka emphasizes his personal preference for setting up Python and aims to keep it simple while using current recommendations.

Setting Up Your Python Environment

GitHub Repository Resources

• A GitHub repository (RASBT/LLM-from-scratch) is available with setup recommendations and resources for readers.

Hardware Compatibility

• The author mentions using an older MacBook Air, ensuring that the code runs well across various hardware configurations, including GPUs and older systems.

Alternatives to Local Setup

Cloud Computing Options

• If local setup fails, Raschka recommends cloud resources like Lightning Studio and Google Colab for executing Jupyter Notebooks easily in a browser.

Personal Preferences in Python Setup

Optional Setup Preferences

• The video demonstrates how Raschka would set up Python on a fresh macOS environment, highlighting optional preferences that users can choose to follow or not.

Package Management Tools

• He discusses his previous use of Conda as a package manager but now prefers UV due to its speed and design advantages over traditional methods like PIP.

Installing Python

Installation Process Overview

• Raschka outlines steps for installing Python, starting by checking if it's already installed via terminal commands.

Version Awareness

How to Install Python and Set Up a Virtual Environment

Installing Python

• The recommended method for installing Python is by visiting the official website, python.org, where users can download versions suitable for different operating systems.

• It is advisable to install an older version of Python (e.g., 3.12 or 3.11) as some packages like PyTorch may not support the latest releases immediately.

• Using slightly older versions ensures compatibility with scientific computing tasks without missing out on essential features.

• Homebrew is mentioned as a personal preference for installation; it’s a package manager for macOS that simplifies the installation of various tools, including Python.

• Users can also install specific versions of Python using Homebrew commands (e.g., brew install python@3.11).

Setting Up Package Managers

• After installing Python, the next step involves setting up UV, a faster alternative to pip, which can be installed via pip install uv.

• UV enhances convenience in managing packages compared to pip and is recommended for better performance.

Creating a Virtual Environment

• A virtual environment is crucial as it creates an isolated space on your computer where all packages are stored separately from system-wide installations.

• This isolation allows users to experiment without affecting their main system setup; if issues arise, they can simply delete and recreate the environment.

Downloading Project Files

• Users should download necessary project files from GitHub repositories and organize them conveniently on their desktops for easy access during setup.

Activating the Virtual Environment

• To create a virtual environment using UV, users execute commands that specify the desired version of Python (e.g., uv env -p python3.11).

• Once created, this folder contains its own version of Python and can be deleted or recreated if needed without impacting other projects.

Managing Visibility of Folders

• Note that folders starting with a dot are hidden by default in Finder; enabling visibility settings will allow users to see these folders when necessary.

Final Steps in Setup

Setting Up Python Environment for Deep Learning

Virtual Environment Setup

• The speaker demonstrates using a virtual environment in macOS to execute Python, ensuring that the correct version (Python 3.11) is utilized from the specified path.

• Installation of packages can be done using uv pip install, with an example given for installing a package called "packages" and PyTorch, which is referred to as "torch" in Python.

Managing Dependencies

• The speaker mentions that all necessary packages for the book are listed in a requirements.txt file, allowing users to install them all at once using pip install -r requirements.

• A note on potential issues when running TensorFlow on Windows due to its requirement for accessing GPT weights, even though the book primarily uses PyTorch.

Handling TensorFlow Issues

• Users may encounter problems with TensorFlow installation on Windows; it’s suggested to either remove TensorFlow from the requirements file or run PIP install without UV.

• An alternative solution involves using Google Colab, which provides a cloud-based environment where notebooks can be executed without local setup complications.

Using Google Colab

• In Google Colab, users can directly install required packages by copying the link from requirements.txt and executing it within their notebook.

• Since Google Colab operates as a virtual environment itself, creating an additional virtual environment is unnecessary; users can simply use system commands.

Running Jupyter Lab

• After installing packages, users can launch Jupyter Lab to work with notebooks. It opens in the primary browser and allows access to different chapters or creation of new notebooks.

• The speaker recommends clearing outputs of existing notebooks before starting fresh coding sessions for better learning experiences rather than just copying code.

Conclusion and Support