{"categories":[{"label":"Artificial Intelligence","url":"https://skillfed.io/packages/category/scientific-engineering-artificial-intelligence/3"}],"enrichment":{"capability":"Implements vector quantization layers for PyTorch, enabling discrete codebook-based compression of continuous embeddings used in generative models.","skillfed_tags":["generative-models","neural-compression"],"use_cases":["Building VQ-VAE models for image compression and generation with discrete latent codes","Implementing audio codecs that use residual quantization across multiple stages","Constructing RQ-VAE architectures with shared codebooks and stochastic sampling for image synthesis","Experimenting with grouped quantization to reduce codebook overhead while maintaining reconstruction quality","Training models with kmeans-initialized codebooks to improve early convergence and codebook utilization"],"what_it_does":"Vector-quantize-pytorch provides PyTorch implementations of vector quantization techniques originally from DeepMind's TensorFlow codebase. It includes the core VectorQuantize layer for single-stage quantization and ResidualVQ for multi-stage hierarchical quantization, both using exponential moving averages to update learned codebooks. The library also offers variants like GroupedResidualVQ and supports modern techniques such as DiVeQ gradient-based updates, kmeans initialization, rotation-trick gradient estimation, and strategies to combat dead codebook entries.\n\nThe package is designed for researchers and practitioners building generative models that require discrete latent representations\u2014particularly image and audio generation systems. It handles the forward pass mapping continuous vectors to nearest codebook entries and provides configurable loss terms and gradient computation methods. Dependencies on einops, einx, and torch-einops-utils suggest heavy use of tensor reshaping and Einstein notation for flexible multi-dimensional operations.","worth_installing":"Yes. The package is actively maintained, has no known vulnerabilities, carries a permissive MIT License, and offers low install friction. It is well-suited for anyone implementing vector quantization in PyTorch, particularly for generative modeling. The Beta status suggests research-grade rather than production-hardened, so verify compatibility with your specific PyTorch and Python versions before relying on it in production."},"id":"vector-quantize-pytorch","links":{"html":"https://skillfed.io/packages/vector-quantize-pytorch","md":"https://skillfed.io/packages/vector-quantize-pytorch.md","pypi":"https://pypi.org/project/vector-quantize-pytorch/"},"maintenance":{"status":"active"},"meta":{"latest_release":"2026-08-02","license_spdx":null,"license_treatment":"permissive","name":"vector-quantize-pytorch","python_support":"supports_current","summary":"Vector Quantization - Pytorch"},"popularity":{"monthly_downloads":1830185,"position":3511,"tier":"top_5000"},"security":{"n_vulnerabilities":0},"version":"1.31.1"}