Key Features
Similarity Search
Find nearest neighbors in high-dimensional vector spaces using efficient indexing.
Indexing Strategies
Supports flat, IVF, HNSW, PQ, and hybrid indexes for speed and memory trade-offs.
GPU Acceleration
Run large-scale searches with CUDA support for massive speedups.
Python & C++ APIs
Use FAISS in Python for prototyping or C++ for production performance.
Clustering & Quantization
Cluster vectors and compress indexes for memory-efficient search.
How It Works
Install FAISS
Use `pip install faiss-cpu` or `faiss-gpu` depending on your hardware.
Prepare Vectors
Generate embeddings using models like OpenAI, Hugging Face, or custom encoders.
Build Index
Choose an index type (e.g., Flat, IVF, HNSW) and train if needed.
Add Vectors
Insert vectors into the index using `add()` or `add_with_ids()`.
Query Neighbors
Use `search()` to find top-k similar vectors for a given query.
Code Example
import faiss
import numpy as np
# Create sample data
d = 128 # vector dimension
nb = 10000 # database size
np.random.seed(1234)
xb = np.random.random((nb, d)).astype('float32')
# Build index
index = faiss.IndexFlatL2(d)
index.add(xb)
# Query
xq = np.random.random((5, d)).astype('float32')
D, I = index.search(xq, k=5)
print(I)Use Cases
Semantic Search
Search documents, FAQs, or transcripts using vector similarity.
Recommendation Systems
Suggest items based on user or item embeddings.
Image Retrieval
Find visually similar images using CNN or CLIP embeddings.
RAG Pipelines
Retrieve relevant context for LLMs using FAISS indexes.
Anomaly Detection
Detect outliers by comparing vectors to known distributions.
Integrations & Resources
Explore FAISS’s ecosystem and find the tools, platforms, and docs to accelerate your workflow.
Popular Integrations
- Hugging Face Transformers for embeddings
- OpenAI and Cohere for semantic vectors
- LangChain and LlamaIndex for RAG
- FastAPI and Flask for deployment
- Streamlit and Gradio for demos
- NumPy and PyTorch for preprocessing
Helpful Resources
FAQ
Common questions about FAISS’s capabilities, usage, and ecosystem.