← Tutorials22 / 22
What is an LLM?
Real LLMs train on thousands of GPUs. Can you learn this without one?Tap to reveal
Yes. The ideas (tokens, embeddings, next-token prediction, loss, gradient descent, sampling) work the same at tiny scale. GPUs only make the matrix multiplications fast for billions of parameters.
In the next tutorial you'll build a character-level language model in plain Rust that trains on a laptop CPU in under a second.