# Quantization

> Browse 35 items tagged with Quantization.

## Items

- [CommVQ](https://vectordb.ever.works/ko/items/commvq) — A commutative vector quantization method for KV cache compression that reduces FP16 cache size by 87.5% with 2-bit quantization and enables 1-bit quantization, allowing LLaMA-3.1 8B to run with 128K context on a single RTX 4090 GPU.
- [VectorChord](https://vectordb.ever.works/ko/items/vectorchord) — PostgreSQL extension for scalable, high-performance vector search, successor to pgvecto.rs. Features RaBitQ quantization enabling 6x cost savings vs Pinecone. Fully compatible with pgvector. This is an OSS extension.
- [faiss-quickeradc](https://vectordb.ever.works/ko/items/faiss-quickeradc) — Optimized variant of Faiss with faster ADC quantization for GPU-accelerated vector search via CUDA, achieving higher throughput and lower latency than CPU Faiss on large-scale similarity tasks. Designed for real-time AI applications, CV inference, and high-QPS workloads requiring NVIDIA hardware acceleration. Outperforms standard CPU Faiss and baselines like Annoy in GPU environments.
- [Hora](https://vectordb.ever.works/ko/items/hora) — High-performance vector search library with product quantization.
- [QBit](https://vectordb.ever.works/ko/items/qbit) — ClickHouse column type for query-time vector precision tuning in hybrid analytical SQL+vector searches. Enables flexible recall/speed tradeoff with ACID-like features on columnar relational infra.
- [nanopq](https://vectordb.ever.works/ko/items/nanopq) — nanopq is a lightweight product quantization library for efficient vector compression and similarity search, which is an important feature for vector databases that need to store and query large-scale vector data efficiently.
- [ScaNN Library](https://vectordb.ever.works/ko/items/scann-library) — Scalable Nearest Neighbors library by Google Research that provides efficient vector similarity search at scale. Uses anisotropic vector quantization and advanced compression techniques to handle twice as many queries per second compared to alternatives.
- [ruvllm](https://vectordb.ever.works/ko/items/ruvllm) — Local LLM inference engine supporting GGUF models with hardware acceleration on Metal, CUDA, ANE, WebGPU. Features Flash Attention, MicroLoRA, RoPE, quantization (Q4-Q8, π-Quantization), MoE routing, and streaming tokens for browser and edge deployment.
- [Boosting Deep Vector Quantization with Progressive Distribution Transformation](https://vectordb.ever.works/ko/items/boosting-deep-vector-quantization-with-progressive-distribution-transformation) — SIGKDD 2025 paper proposing a progressive distribution transformation approach for boosting deep vector quantization, improving quantization accuracy by progressively adapting data distributions during training.
- [Optimizing Clusters for Billion-Scale Quantization-Based NNS](https://vectordb.ever.works/ko/items/optimizing-clusters-for-billion-scale-quantization-based-nns) — TKDE 2024 paper on optimizing the number of clusters for billion-scale quantization-based nearest neighbor search, providing methods to determine optimal clustering for quantized vector indexing.
- [RaBitQ — Quantizing High-Dimensional Vectors with Theoretical Error Bound for ANNS](https://vectordb.ever.works/ko/items/rabitq-quantizing-high-dimensional-vectors-with-theoretical-error-bound-for-anns) — SIGMOD 2024 paper introducing RaBitQ, a quantization method for high-dimensional vectors with provable theoretical error bounds for approximate nearest neighbor search in Euclidean space.
- [Residual Quantization with Implicit Neural Codebooks](https://vectordb.ever.works/ko/items/residual-quantization-with-implicit-neural-codebooks) — ICML 2024 paper presenting a novel residual quantization approach using implicit neural codebooks for vector compression in high-dimensional similarity search, replacing traditional fixed codebooks with learned representations.
- [Routing-Guided Learned Product Quantization for Graph-Based ANNS](https://vectordb.ever.works/ko/items/routing-guided-learned-product-quantization-for-graph-based-anns) — ICDE 2024 paper proposing a routing-guided learned product quantization method that enhances graph-based approximate nearest neighbor search by learning optimal quantization guided by graph routing information.
- [ScaNN — Accelerating Large-Scale Inference with Anisotropic Vector Quantization](https://vectordb.ever.works/ko/items/scann-accelerating-large-scale-inference-with-anisotropic-vector-quantization) — ICML 2020 paper introducing ScaNN (Scalable Nearest Neighbors), a system for accelerating large-scale vector similarity search using anisotropic vector quantization, combining quantization with asymmetric distance computation for high-performance ANN search.
- [Binary Quantization for Vector Search](https://vectordb.ever.works/ko/items/binary-quantization-for-vector-search) — Compression technique that converts full-precision vectors to binary representations, achieving 32x storage reduction while maintaining 90-95% recall for efficient large-scale vector search.
- [Qdrant 1.5-bit Quantization](https://vectordb.ever.works/ko/items/qdrant-15-bit-quantization) — Middle-ground quantization introduced in Qdrant v1.15.0 that provides better precision than binary quantization while being more aggressive than scalar quantization.
- [IVF-SQ8 Index](https://vectordb.ever.works/ko/items/ivf-sq8-index) — A quantization-based vector indexing algorithm that combines Inverted File Index (IVF) with 8-bit scalar quantization (SQ8). Designed to tackle large-scale similarity search challenges, achieving faster searches with a much smaller memory footprint compared to exhaustive search methods by using 8-bit integers instead of 32-bit floats.
- [VQKV](https://vectordb.ever.works/ko/items/vqkv) — A training-free vector quantization method for KV cache compression in Large Language Models that achieves 82.8% compression ratio on LLaMA3.1-8B while retaining 98.6% baseline performance and enabling 4.3x longer generation length on the same memory footprint.
- [Statistical Binary Quantization](https://vectordb.ever.works/ko/items/statistical-binary-quantization) — Compression method developed by Timescale researchers that improves on standard Binary Quantization, reducing vector memory footprint by 32x while maintaining high accuracy for filtered searches.
- [BBQ Binary Quantization](https://vectordb.ever.works/ko/items/bbq-binary-quantization) — Elasticsearch and Lucene's implementation of RaBitQ algorithm for 1-bit vector quantization, renamed as BBQ. Provides 32x compression with asymptotically optimal error bounds, enabling efficient vector search at massive scale with minimal accuracy loss.
- [Leech Lattice Vector Quantization](https://vectordb.ever.works/ko/items/leech-lattice-vector-quantization) — Advanced vector quantization technique that explores the Leech lattice's optimal sphere packing properties at 24 dimensions. Delivers state-of-the-art LLM quantization performance, outperforming recent methods like Quip#, QTIP, and PVQ for extreme vector compression.
- [Locally-Adaptive Vector Quantization](https://vectordb.ever.works/ko/items/locally-adaptive-vector-quantization) — Advanced quantization technique that applies per-vector normalization and scalar quantization, adapting the quantization bounds individually for each vector. Achieves four-fold reduction in vector size while maintaining search accuracy with 26-37% overall memory footprint reduction.
- [Anisotropic Vector Quantization](https://vectordb.ever.works/ko/items/anisotropic-vector-quantization) — An advanced quantization technique introduced by Google's ScaNN that prioritizes preserving parallel components between vectors rather than minimizing overall distance. Optimized for Maximum Inner Product Search (MIPS) and significantly improves retrieval accuracy.
- [Binary Quantization](https://vectordb.ever.works/ko/items/binary-quantization) — Extreme vector compression technique converting each dimension to a single bit (0 or 1), achieving 32x memory reduction and enabling ultra-fast Hamming distance calculations with acceptable accuracy trade-offs.
- [Product Quantization (PQ)](https://vectordb.ever.works/ko/items/product-quantization-pq) — Vector compression technique that splits high-dimensional vectors into subvectors and quantizes each independently, achieving significant memory reduction while enabling approximate similarity search.
- [Scalar Quantization](https://vectordb.ever.works/ko/items/scalar-quantization) — Vector compression technique reducing precision of each vector component from 32-bit floats to 8-bit integers, achieving 4x memory reduction with minimal accuracy loss for vector search.
- [TreeAH](https://vectordb.ever.works/ko/items/treeah) — Vector index type based on Google's ScaNN algorithm combining tree-like structure with Asymmetric Hashing quantization, optimized for batch queries with 10x faster index generation and smaller memory footprint.
- [AWQ](https://vectordb.ever.works/ko/items/awq) — Activation-aware Weight Quantization method that preserves model accuracy at 4-bit quantization by identifying and skipping important weights. Maintains 99%+ of original performance with moderate inference speed improvements.
- [GGUF](https://vectordb.ever.works/ko/items/gguf) — GPT-Generated Unified Format for storing quantized model weights, designed for CPU inference and consumer hardware. Enables running LLMs on laptops and edge devices with flexible layer offloading to GPU.
- [GPTQ](https://vectordb.ever.works/ko/items/gptq) — Post-training quantization method for 4-bit weight compression that focuses on GPU inference performance. First quantization method to compress LLMs to 4-bit range while maintaining accuracy, minimizing mean squared error to weights.
- [Product Quantization Compression](https://vectordb.ever.works/ko/items/product-quantization-compression) — Lossy vector compression dividing vectors into subvectors for independent quantization. Achieves 8-64x storage reduction while enabling fast approximate distance computation via lookup tables.
- [voyage-4](https://vectordb.ever.works/ko/items/voyage-4) — Latest Voyage AI embedding model family featuring shared embedding space with MoE architecture, supporting flexible output dimensions and advanced quantization options for cost optimization.
- [IVF-PQ (Inverted File with Product Quantization)](https://vectordb.ever.works/ko/items/ivf-pq-inverted-file-with-product-quantization) — Vector indexing method combining inverted file index with product quantization for memory-efficient search. Reduces storage from 128x4 bytes to 32x1 bytes (1/16th) while maintaining search quality.
- [PQ (Product Quantization)](https://vectordb.ever.works/ko/items/pq-product-quantization) — Product Quantization is a compression and indexing technique for vector search that splits vectors into subspaces and quantizes each part separately, allowing vector databases to store large-scale embeddings compactly while supporting efficient ANN search.
- [Quantization](https://vectordb.ever.works/ko/items/quantization) — Resources and tools on quantization techniques for vectors, which are essential for optimizing storage and retrieval in vector databases.

---

_Canonical page: https://vectordb.ever.works/ko/tags/quantization_