GPT Models Explained with 3D Visualization by BrendanBycroft
3D visualization showing how GPT models work
BrendanBycroft created a web page to visualize the internal structure and workings of a large language model (LLM), similar to the Transformer model behind ChatGPT.
This page, rendered in 3D, shows all the steps a large language model like GPT takes to run a single token inference and how it works.
This project created by Brendan Bycroft can be used to demonstrate artificial intelligence models of different sizes, such as the relatively small model of GPT-2.
But because larger models (such as GPT-3) require processing and storing a lot of data, this project does not include all the data for these large models.
Left click to drag, right click to rotate, scroll to zoom. and hover the mouse over the tensor unit.
Blue cells are weights/parameters and green cells are intermediate values.
Every cell is a number!
🔧 And he also open sourced the project
This project also has a CPU Simulation display.
It provides a 2D environment in which you can edit and simulate how the CPU (Central Processing Unit) works.
You can create and modify digital circuits in this editor that simulate how a real CPU works.
Portal
GitHub
More AI News
Artificial Intelligence Article
New AI Technology
