Vector Quantize Pytorch

Understand the project

Practical overview

Use it when a PyTorch model needs a configurable quantization layer with codebook-based representations and returned quantization losses or indices.

Before you choose

The repository is public, active and MIT-licensed.

Choose VectorQuantize for a basic quantization layer.
Choose ResidualVQ when recursively quantizing residuals with multiple quantizers.
Choose GroupedResidualVQ when applying residual quantization across feature groups.
Use k-means initialization when the codebook should start from centroids of the first batch.
Consider shared codebooks or stochastic code sampling when following the documented RQ-VAE-style configuration.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Add the documented VectorQuantize layer to a PyTorch project and run it on a tensor.

Install the package with the command below.
Documented command
```
pip install vector-quantize-pytorch
```
Import torch and VectorQuantize from vector_quantize_pytorch.
Create VectorQuantize with the documented dim, codebook_size, decay and commitment_weight parameters.
Create an input tensor and call the layer to obtain quantized, indices and commit_loss.

A configured vector-quantization layer returns quantized vectors, codebook indices and commitment loss.

The documented example demonstrates layer invocation, not a complete training pipeline or model deployment.
The example uses tensor dimensions and parameter values that may need adjustment for a particular model.

Evidence and transparency

Source record details

Source availability: Active
Sources checked: 30 September 2026
Original website title: lucidrains/vector-quantize-pytorch · Official homepage title · retained in its source language