A hands-on beginner workshop on what actually happens inside an LLM.

⚡ Sponsored by Modal

You’ve used ChatGPT. Maybe you’ve played with prompts, APIs or open models.

But what does it actually mean to train a language model?

In this 4-hour workshop, we’ll go from a tiny model that barely knows how to speak to a real pretrained LLM that we can fine-tune ourselves.

No AI magic. No 80-slide lecture. We’ll actually build things.

What we’ll do

  • Turn text into tokens and see how next-token prediction works

  • Train a tiny language model from scratch

  • See how a Transformer improves on our first model

  • Experiment with temperature, training time and model size

  • Run a real open-source LLM on a GPU

  • Understand CPU vs. GPU, memory and quantization

  • Fine-tune a pretrained LLM with LoRA

  • Compare the model before and after training

You’ll leave understanding

Training vs. inference. Pretraining vs. fine-tuning. Tokens, parameters, Transformers, GPUs, LoRA and why any of it matters.

More importantly, you’ll have actually used these concepts yourself instead of just hearing about them.

Who is this for?

Developers, students, data/product people and technically curious humans who want to understand LLMs beyond prompting.

You don’t need previous machine-learning experience.

Basic Python familiarity will help, but the exercises are guided.

Format

4 hours · Beginner-friendly · Hands-on

GPU compute for the workshop is sponsored by Modal. We’ll use a prepared cloud environment, so you won’t need to install CUDA, own a powerful laptop or spend half the workshop debugging dependencies.

Bring your laptop and curiosity.