Skip to content

Latest commit

 

History

History
40 lines (31 loc) · 1.69 KB

README.md

File metadata and controls

40 lines (31 loc) · 1.69 KB

VQ-BeT TinyGrad Notebook

This repository contains a Jupyter notebook implementing VQ-BeT (Vector Quantized Behavior Transformers) using TinyGrad.

What is VQ-BeT?

VQ-BeT is an advanced AI model for robot learning. It combines vector quantization and transformer architectures to help robots learn complex behaviors from demonstrations. VQ-BeT processes sequences of actions and observations, encoding them into discrete tokens. These tokens are then used to train a transformer model, which can generate new sequences of actions for the robot to perform. This approach allows robots to learn and generalize from a small number of demonstrations, making it efficient for various robotic tasks. The link to the original repo is here.

Notebook Structure

This notebook interleaves explanatory text with executable code blocks:

  1. Detailed explanations of VQ-BeT concepts and implementation details
  2. Code snippets implementing various components of the VQ-BeT model
  3. Visualizations and examples to illustrate key concepts
  4. Step-by-step implementation of the VQ-BeT algorithm using TinyGrad

Quick Start

To run this notebook:

  1. Ensure you have Jupyter Lab installed
  2. Install TinyGrad version 0.9.2: 
    pip install tinygrad==0.9.2
    Note: TinyGrad 0.9.2 is required as the Tensor.realize function in version 0.10.0 is incompatible with this notebook.
  3. Clone this repository: 
    git clone https://github.com/mdaiter/vbet-tinygrad-notebook.git
  4. Navigate to the repository directory: 
    cd vbet-tinygrad-notebook
  5. Start Jupyter Lab: 
    jupyter-lab
  6. Open the VQ-BeT notebook and start exploring!