This library implements vector and scalar quantization as PyTorch modules, originally transcribed from Deepmind’s TensorFlow implementation. It uses exponential moving averages to update codebook dictionaries and provides both standard VectorQuantize and ResidualVQ classes for recursive quantization of residuals. The library enables high-quality image generation (VQ-VAE-2) and music generation (Jukebox) by compressing latent representations into discrete codebook entries.