patann 0.1.3

PatANN is a massively parallel, distributed, and scalable in-memory/on-disk vector database library for efficient nearest neighbor search across large-scale datasets by finding vector patterns.

PatANN - Pattern-Aware Vector Database and ANN Framework

Overview

PatANN is a pattern-aware, massively parallel, and distributed vector search framework designed for scalable and efficient nearest neighbor search, operating both in-memory and on-disk. Unlike conventional algorithms, PatANN leverages macro and micro patterns within vectors to drastically reduce search space before performing costly distance computations.

Refer to the website for technical details, algorithm overview, key innovations, benchmarks, and tutorials.

https://patann.dev

While still in beta, PatANN's pattern-first approach delivers unprecedented performance advantages. As shown in our benchmarks (Figure 1), PatANN consistently outperforms leading ANN libraries including HNSW (hnswlib), Google ScaNN, Facebook FAISS variants, and others in the critical recall-throughput tradeoff.

Installation

Install PatANN using pip:

pip install patann

Platforms

• Linux
• macOS (Apple Silicon)
• Windows
• Android
• iOS

Code Examples

Code examples for all platforms (Python, Kotlin, Java, Objective-C, Swift) in both asynchronous and synchronous modes are available at

https://github.com/mesibo/patann

You can also download python sample code using patann command line interface:

patann --help

Key Distinguishing Features

• Novel pattern-based probing for ANN search
• In-Memory, On-Disk, and Hybrid Index
• Fully asynchronous operations with built-in support for conventional synchronous execution
• Refined search, filtering, and pagination algorithms
• Unlimited scalability without pre-specified capacity
• Dynamic sharding to load balance across servers
• Cloud (in-progress) and Serverless
• SIMD-Accelerated for both x86_64 (SSE*, AVX2, AVX-512) and ARM (NEON, SVE) platforms
• OS-optimized I/O—huge pages (Linux), large pages (Windows), and super pages (macOS)
• NUMA-aware architecture