bbycroft/llm-viz

3D Visualization of an GPT-style LLM

What it solves

It provides a way to visualize the internal workings of a GPT-style Large Language Model (LLM) in 3D, making the abstract process of inference more tangible and understandable.

How it works

The project implements a 3D interactive renderer that displays the network topology used in models like GPT-2 and GPT-3. It can visualize a tiny network (using weights from Andrej Karpathy's minGPT) that sorts letters, as well as larger networks like gpt2, though the weights for larger models are not pre-downloaded due to their size.

Who it’s for

People interested in understanding the architectural structure and data flow of transformer-based LLMs through interactive visualization.

Highlights

  • 3D interactive model of a GPT-style network.
  • Visualizes the network topology used in OpenAI's GPT-2 and GPT-3.
  • Supports visualizing arbitrary sized networks.
  • Includes a working demo using a tiny network that sorts letters.

Related

  • Project
    poloclub/transformer-explainer

    An interactive visualization tool that runs a live GPT-2 model in the browser to help users learn how Transformer-based models predict text.

  • Dispatch
    OpenAI GPT-5 First Look

    OpenAI has provided a first look at GPT-5, marking a new generation of their large language model series.

  • Project
    willyfh/visualtorch

    A visualization tool for PyTorch neural network architectures that generates flow, graph, and LeNet-style diagrams and animated reveals.

  • Dispatch
    OpenAI GPT-3: Language Models as Few-Shot Learners

    OpenAI introduced GPT-3, a 175 billion parameter autoregressive language model that demonstrates strong few-shot performance across diverse NLP tasks without requiring task-specific fine-tuning.

  • Project
    NVIDIA/TensorRT-LLM

    An open-source optimization library from NVIDIA that maximizes inference performance, throughput, and efficiency for LLMs and visual generation models on NVIDIA GPUs.