Skip to content

Latest commit

 

History

History
20 lines (12 loc) · 846 Bytes

README.md

File metadata and controls

20 lines (12 loc) · 846 Bytes

LLM Visualization (TR)

This is a fork of the original project. You can find the original repository here: LLM-Viz Original Repository

This project displays a 3D model of a working implementation of a GPT-style network. That is, the network topology that's used in OpenAI's GPT-2, GPT-3, (and maybe GPT-4).

The first network displayed with working weights is a tiny such network, which sorts a small list of the letters A, B, and C. This is the demo example model from Andrej Karpathy's minGPT implementation.

The renderer also supports visualizing arbitrary sized networks, and works with the smaller gpt2 size, although the weights aren't downloaded (it's 100's of MBs).

Running Locally

  1. Install dependencies: yarn
  2. Start the dev server: yarn dev