How To Train Your LLM 🐉

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.