skills.rest
Clay-HHK
Clay-HHKCOMMUNITY
Explore all skills by this author →
GitHub

faiss

Implements billion-scale vector similarity search using FAISS index types and GPU acceleration.

1|Updated Mar 5, 2026
Search Tags:#embeddings#rag#vector search#similarity search#hnsw#faiss#k-nn
One-click install
npx skills add https://github.com/Clay-HHK/claude-skills --skill faiss-clay-hhk
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: faiss
Source: https://github.com/Clay-HHK/claude-skills/tree/main/faiss
Command: npx skills add https://github.com/Clay-HHK/claude-skills --skill faiss-clay-hhk

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires faiss-cpu, faiss-gpu, numpy, and includes references (resource) components.

What problem does it solve? Performing nearest-neighbor search over millions or billions of dense embedding vectors is computationally expensive with naive approaches, and this Skill provides guidance for building fast, memory-efficient similarity search indexes with FAISS. ## Core Features & Use Cases - Index Type Selection: Covers Flat, IVF, HNSW, and Product Quantization indexes with guidance on choosing based on dataset size, accuracy, and memory constraints. - GPU Acceleration: Shows how to move indexes to single or multiple GPUs for 10-100x faster search on large datasets. - Framework Integration: Includes LangChain and LlamaIndex integration patterns for RAG pipelines, plus index save/load for persistence. - Use Case: When building a semantic search system over 5 million document embeddings, use this Skill to create an HNSW index, tune efSearch for the right speed/accuracy balance, and persist the trained index to disk. ## Quick Start Ask the AI to create a FAISS index for your embedding vectors and run a k-nearest-neighbor search on a sample query.

Frequently Asked Questions about faiss

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I perform similarity search on vectors with FAISS?▼

Create an index such as faiss.IndexFlatL2(d) for your vector dimension, add your float32 vectors with index.add(), then call index.search(query, k) to get the k nearest neighbors with distances. Larger datasets benefit from IVF or HNSW indexes.

Which FAISS index type should I use for my dataset size?▼

Use Flat for under 10K vectors with exact results, IVF for 10K-1M vectors with fast approximate search, HNSW for 1M-10M vectors with the best quality, and IVF+PQ for over 10M vectors when memory is limited.

FAISS vs Chroma or Pinecone for vector search?▼

FAISS is a pure similarity search library without metadata filtering or database features, making it ideal for high-performance offline or embedded search. Choose Chroma or Pinecone when you need metadata filtering, managed storage, or full database capabilities.

Does FAISS support GPU acceleration?▼

Yes, install faiss-gpu and use faiss.index_cpu_to_gpu(res, 0, index) to move an index to a GPU, or faiss.index_cpu_to_all_gpus(index) for multi-GPU. GPU execution typically delivers 10-100x speedup over CPU.

How do I use cosine similarity with FAISS?▼

Use faiss.IndexFlatIP(d) for inner product search and normalize your vectors with faiss.normalize_L2() before adding and querying. Inner product on L2-normalized vectors is equivalent to cosine similarity.

Why does my FAISS IVF index return poor results?▼

IVF indexes require training on representative data via index.train() before adding vectors, and results depend on the nprobe parameter. Increase nprobe to search more clusters, trading speed for higher recall.

Related Skills

Explore More Software Engineering Skills→
DoanNgocCuongDoanNgocCuong

sentence-transformers

Generate sentence, text, and image embeddings using pre-trained transformer models.

Community
Intermediate
trueratruera
3.5k

trulens-blocking-guardrails

Configure TruLens feedback functions as runtime guardrails to block unsafe inputs, outputs, and irrelevant context.

Official
Intermediate
TechCorp25TechCorp25

vector-index-tuning

Optimize vector index performance for latency, recall, and memory usage.

Community
Advanced
vecearvecear

vector-db-setup

Configure vector databases for semantic search across Pinecone, Chroma, pgvector, and Qdrant.

Community
Advanced