whatisgithub

What is deep-vector-quantization?

karpathy/deep-vector-quantization — explained in plain English

Analysis updated 2026-07-17 · repo last pushed 2021-11-20

★ 647Jupyter NotebookAudience · researcherComplexity · 4/5DormantSetup · hard

In one sentence

A training toolkit that teaches an AI model called a VQVAE to compress images into a small set of discrete codes that can be reconstructed back, useful for feeding images into generative models like GPT.

Mindmap

mindmap
  root((repo))
    What it does
      Compress images
      Discrete codebook
      Reconstruct images
    Implementations
      DeepMind version
      Gumbel Softmax
      DALL-E attempt
    Use cases
      Generative image models
      Compression research
      Learning resource
    Requirements
      GPU training
      CIFAR-10 dataset
      Still evolving

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

What do people build with it?

USE CASE 1

Train a VQVAE to compress images into discrete codes on a dataset like CIFAR-10

USE CASE 2

Build the image-encoding stage for a system that feeds compressed image codes into a GPT-style generative model

USE CASE 3

Compare three implementations (DeepMind, Gumbel Softmax, and an in-progress DALL-E recreation) to learn their tradeoffs

USE CASE 4

Study discrete image compression as a learning resource before building your own generative image system

What is it built with?

PythonPyTorchJupyter Notebook

How does it compare?

karpathy/deep-vector-quantizationllsourcell/how-to-predict-stock-prices-easily-demonvidia/cuopt-examples
Stars647771452
LanguageJupyter NotebookJupyter NotebookJupyter Notebook
Last pushed2021-11-202022-06-23—
MaintenanceDormantDormant—
Setup difficultyhardmoderatemoderate
Complexity4/52/53/5
Audienceresearchervibe coderdeveloper

Figures from each repo's GitHub metadata at analysis time.

How do you get it running?

Difficulty · hard Time to first run · 1h+

Requires a GPU for training, the DALL-E-style implementation is still incomplete.

License terms are not described in the explanation, check the repository directly before use.

Copy-paste prompts

Prompt 1
Walk me through training the DeepMind-style VQVAE in this repo on the CIFAR-10 dataset using a GPU.
Prompt 2
Explain how the Gumbel Softmax variant in this repo differs from the original DeepMind VQVAE approach.
Prompt 3
Show me how to avoid the 'catastrophic index collapse' issue mentioned in this repo when training the original VQVAE.
Prompt 4
How do I take the discrete codes this repo produces and feed them into a GPT-style model for image generation?
Prompt 5
Compare the three VQVAE implementations in this repo and recommend which one to start learning with.

Frequently asked questions

What is deep-vector-quantization?

A training toolkit that teaches an AI model called a VQVAE to compress images into a small set of discrete codes that can be reconstructed back, useful for feeding images into generative models like GPT.

What language is deep-vector-quantization written in?

Mainly Jupyter Notebook. The stack also includes Python, PyTorch, Jupyter Notebook.

Is deep-vector-quantization actively maintained?

Dormant — no commits in 2+ years (last push 2021-11-20).

What license does deep-vector-quantization use?

License terms are not described in the explanation, check the repository directly before use.

How hard is deep-vector-quantization to set up?

Setup difficulty is rated hard, with roughly 1h+ to a first successful run.

Who is deep-vector-quantization for?

Mainly researcher.

Open on GitHub → Ask about another repo

This repo across BitVibe Labs

Verify against the repo before relying on details.

What are similar repos?