Redis 8 以新的数据类型针对向量相似性的 AI 应用程序

发布于 4 月 20 日

Redis Announces Vector Set: Redis has recently added Vector Set, a new data type for vector similarity and an option for AI applications. It is the first major contribution from Salvatore Sanfilippo (aka 'antirez') since rejoining the company.
- Data Type Details: Similar to Sorted Sets but with string elements associated with a vector instead of a score. Allows adding items and retrieving similar ones. Supports filtered search with vector similarity and scalar filters simultaneously.
- Based on HNSW: Based on the hnsw.c implementation with extensions for speed and functionality.
CEO's Praise: Rowan Trollope, CEO of Redis, praises Sanfilippo's contribution, stating that his expertise led to a simple and intuitive API reflecting Redis's philosophy.
Use Cases: Critical for GenAI applications like retrieving semantically relevant information for RAG. Other use cases include semantic caching for chatbots, recommender systems, and face recognition.
- Face Recognition Example: Face recognition is about modeling known faces with an embedding model and storing them in Redis for vector search.
Sanfilippo's Work: Sanfilippo reimplemented HNSW from scratch and introduced modifications like multithreading for all vector similarity requests and support for 8-bit and binary quantization. Highlights the unique data model and API.
Other Redis Feature: LangCache, a semantic caching service for AI apps and agents, is designed to reduce calls to LLMs by caching responses.
Availability: Vector Sets are available in preview with Redis 8 RC1 under RSALv2 or SSPLv1 license.

阅读 7