The GPT-3 Architecture, on a Napkin
A hand-drawn, step-by-step diagram tour of how GPT-3 turns words into predictions, covering embeddings, attention and layers with napkin-style sketches and minimal math.
No account needed · sign in to keep your taste everywhere
More like this
Pages that share The GPT-3 Architecture, on a Napkin's themes. Or browse all of Tech.
Training a 125M-Parameter Model to Autocomplete Piano
A write-up of training a small transformer to autocomplete piano MIDI performances in real time on-device, covering the…
microgpt
A visualization of a tiny GPT language model running in your browser, so you can watch how tokens flow through the layer…
Jay Alammar's Illustrated Guides
Illustrated guides that break down how transformers, GPT, and other AI models work using detailed step-by-step diagrams…
Sergiy Bondaryev
Interactive articles that build intuition for machine learning, such as a diffusion transformer turning random dots into…
MicroGPT Explained Interactively
An interactive walkthrough of a 200-line GPT, from tokenizing names to stepping through backpropagation and attention he…
LLM Attention Visualization
An interactive visualization with a short explanation of how attention lets a language model draw selectively on earlier…
Stumble onto more like this
One click takes you to a page you didn't know you were looking for. Rate it and the next one lands closer to your taste.
Add to Browser — it's freeWorks in Chrome, Edge and Brave. No account needed.