Llm Inference Arithmetics: The Theory Behind Model Serving

Source code for the talk Llm Inference Arithmetics: The Theory Behind Model Serving at PyCon DE 2025.

• 📽️ Live

🛩️ How to run

Prerequisites

Make sure you have installed the following:

• git.
• bun (but you can use any other JavaScript runtime).
• just, a command runner.
  • The justfile contains a bunch of utilities to ease development. Though optional, it's highly recommended.
• gh, the GitHub CLI.
  • This is optional too, but recommended.

1. Clone the repo

# with github CLI
gh repo clone baggiponte/llm-inference-arithmetics

# with git
git clone [email protected]:baggiponte/llm-inference-arithmetics

Note

🎬 How to configure SSH

Cloning over SSH is safer. Follow this guide. Alternatively, you can follow the steps in this workshop of GitHub's.

2. Install the dependencies

# with the command runner
just install

3. Open the slideshow locally

# with the command runner
just preview

4. Visit http://localhost:3030

🤗 contributing

Prerequisites

• Install all of the the dependencies above and commitizen. commitizen is a release management tool. It's used to release new versions.

Release workflow

1. Fork the repo:

gh repo fork baggiponte/llm-inference-arithmetics

2. Create your own branch.

git checkout -b your-branch-name

3. Commit following the conventional commits specification.

4. Before pushing your changes, run the following command to check if the version bump is possible.

just test-release

This will format the slides, try to build them and test whether a version bump is possibile.

🙏 Credits

• slidev is an amazing framework to build slides from markdown and host them.
• just is just a command runner.
• commitizen is a release management tool.