Faiss

orchestra-research/ai-research-skills/15-rag/faiss

作者 orchestra-research773a52944ba4MIT13K 个星标收录于 2026年10月8日更新于 2026年10月8日仓库3个月前更新

Facebook's library for efficient similarity search and clustering of dense vectors. Supports billions of vectors, GPU acceleration, and various index types (Flat, IVF, HNSW). Use for fast k-NN search, large-scale vector retrieval, or when you need pure similarity search without metadata. Best for high-performance applications.

仅含说明AI & Agents

AI 生成的概览

使用 FAISS 构建和调优向量相似度搜索索引的参考指南。

功能: 该技能是 Facebook AI 的 FAISS 库的参考指南，用于稠密向量的相似度搜索与聚类。它说明安装方式、索引类型（Flat、IVF、HNSW、乘积量化）、索引的保存与加载、GPU 加速以及与 LangChain/LlamaIndex 的集成，并附有代码示例和性能对比。它产出的是指导说明和示例代码，而非可执行产物。
适用场景: 当你需要快速 k 近邻搜索或大规模向量检索（包括十亿级数据集或 GPU 加速搜索）时使用。它也适用于在 FAISS 各索引类型之间做选择，或将 FAISS 与 Chroma、Pinecone、Weaviate、Annoy 等替代方案进行比较。
运行要求: 需要 faiss-cpu 或 faiss-gpu Python 包以及 numpy；GPU 加速需要 faiss-gpu 和相应硬件。可选集成涉及 LangChain、LlamaIndex 和 OpenAI 嵌入。该技能不附带脚本，只有说明文档和一份参考文档。

FAISS - Efficient Similarity Search

Facebook AI's library for billion-scale vector similarity search.

When to use FAISS

Use FAISS when:

Need fast similarity search on large vector datasets (millions/billions)
GPU acceleration required
Pure vector similarity (no metadata filtering needed)
High throughput, low latency critical
Offline/batch processing of embeddings

Metrics:

31,700+ GitHub stars
Meta/Facebook AI Research
Handles billions of vectors
C++ with Python bindings

Use alternatives instead:

Chroma/Pinecone: Need metadata filtering
Weaviate: Need full database features
Annoy: Simpler, fewer features

Quick start

Installation

Basic usage

python

import faissimport numpy as np
# Create sample data (1000 vectors, 128 dimensions)d = 128nb = 1000vectors = np.random.random((nb, d)).astype('float32')
# Create indexindex = faiss.IndexFlatL2(d)  # L2 distanceindex.add(vectors)             # Add vectors
# Searchk = 5  # Find 5 nearest neighborsquery = np.random.random((1, d)).astype('float32')distances, indices = index.search(query, k)
print(f"Nearest neighbors: {indices}")print(f"Distances: {distances}")

import faissimport numpy as np
# Create sample data (1000 vectors, 128 dimensions)d = 128nb = 1000vectors = np.random.random((nb, d)).astype('float32')
# Create indexindex = faiss.IndexFlatL2(d)  # L2 distanceindex.add(vectors)             # Add vectors
# Searchk = 5  # Find 5 nearest neighborsquery = np.random.random((1, d)).astype('float32')distances, indices = index.search(query, k)
print(f"Nearest neighbors: {indices}")print(f"Distances: {distances}")

Index types

1. Flat (exact search)

python

# L2 (Euclidean) distanceindex = faiss.IndexFlatL2(d)
# Inner product (cosine similarity if normalized)index = faiss.IndexFlatIP(d)
# Slowest, most accurate

# L2 (Euclidean) distanceindex = faiss.IndexFlatL2(d)
# Inner product (cosine similarity if normalized)index = faiss.IndexFlatIP(d)
# Slowest, most accurate

2. IVF (inverted file) - Fast approximate

python

# Create quantizerquantizer = faiss.IndexFlatL2(d)
# IVF index with 100 clustersnlist = 100index = faiss.IndexIVFFlat(quantizer, d, nlist)
# Train on dataindex.train(vectors)
# Add vectorsindex.add(vectors)
# Search (nprobe = clusters to search)index.nprobe = 10distances, indices = index.search(query, k)

# Create quantizerquantizer = faiss.IndexFlatL2(d)
# IVF index with 100 clustersnlist = 100index = faiss.IndexIVFFlat(quantizer, d, nlist)
# Train on dataindex.train(vectors)
# Add vectorsindex.add(vectors)
# Search (nprobe = clusters to search)index.nprobe = 10distances, indices = index.search(query, k)

3. HNSW (Hierarchical NSW) - Best quality/speed

python

# HNSW indexM = 32  # Number of connections per layerindex = faiss.IndexHNSWFlat(d, M)
# No training neededindex.add(vectors)
# Searchdistances, indices = index.search(query, k)

# HNSW indexM = 32  # Number of connections per layerindex = faiss.IndexHNSWFlat(d, M)
# No training neededindex.add(vectors)
# Searchdistances, indices = index.search(query, k)

4. Product Quantization - Memory efficient

python

# PQ reduces memory by 16-32×m = 8   # Number of subquantizersnbits = 8index = faiss.IndexPQ(d, m, nbits)
# Train and addindex.train(vectors)index.add(vectors)

# PQ reduces memory by 16-32×m = 8   # Number of subquantizersnbits = 8index = faiss.IndexPQ(d, m, nbits)
# Train and addindex.train(vectors)index.add(vectors)

Save and load

python

# Save indexfaiss.write_index(index, "large.index")
# Load indexindex = faiss.read_index("large.index")
# Continue usingdistances, indices = index.search(query, k)

# Save indexfaiss.write_index(index, "large.index")
# Load indexindex = faiss.read_index("large.index")
# Continue usingdistances, indices = index.search(query, k)

GPU acceleration

python

# Single GPUres = faiss.StandardGpuResources()index_cpu = faiss.IndexFlatL2(d)index_gpu = faiss.index_cpu_to_gpu(res, 0, index_cpu)  # GPU 0
# Multi-GPUindex_gpu = faiss.index_cpu_to_all_gpus(index_cpu)
# 10-100× faster than CPU

# Single GPUres = faiss.StandardGpuResources()index_cpu = faiss.IndexFlatL2(d)index_gpu = faiss.index_cpu_to_gpu(res, 0, index_cpu)  # GPU 0
# Multi-GPUindex_gpu = faiss.index_cpu_to_all_gpus(index_cpu)
# 10-100× faster than CPU

LangChain integration

python

from langchain_community.vectorstores import FAISSfrom langchain_openai import OpenAIEmbeddings
# Create FAISS vector storevectorstore = FAISS.from_documents(docs, OpenAIEmbeddings())
# Savevectorstore.save_local("faiss_index")
# Loadvectorstore = FAISS.load_local(    "faiss_index",    OpenAIEmbeddings(),    allow_dangerous_deserialization=True)
# Searchresults = vectorstore.similarity_search("query", k=5)

from langchain_community.vectorstores import FAISSfrom langchain_openai import OpenAIEmbeddings
# Create FAISS vector storevectorstore = FAISS.from_documents(docs, OpenAIEmbeddings())
# Savevectorstore.save_local("faiss_index")
# Loadvectorstore = FAISS.load_local(    "faiss_index",    OpenAIEmbeddings(),    allow_dangerous_deserialization=True)
# Searchresults = vectorstore.similarity_search("query", k=5)

LlamaIndex integration

python

from llama_index.vector_stores.faiss import FaissVectorStoreimport faiss
# Create FAISS indexd = 1536faiss_index = faiss.IndexFlatL2(d)
vector_store = FaissVectorStore(faiss_index=faiss_index)

from llama_index.vector_stores.faiss import FaissVectorStoreimport faiss
# Create FAISS indexd = 1536faiss_index = faiss.IndexFlatL2(d)
vector_store = FaissVectorStore(faiss_index=faiss_index)

Best practices

Choose right index type - Flat for <10K, IVF for 10K-1M, HNSW for quality
Normalize for cosine - Use IndexFlatIP with normalized vectors
Use GPU for large datasets - 10-100× faster
Save trained indices - Training is expensive
Tune nprobe/ef_search - Balance speed/accuracy
Monitor memory - PQ for large datasets
Batch queries - Better GPU utilization

Performance

Resources

GitHub: https://github.com/facebookresearch/faiss ⭐ 31,700+
Wiki: https://github.com/facebookresearch/faiss/wiki
License: MIT

来源与署名

来源：orchestra-research/ai-research-skills位于15-rag/faiss提交773a529

许可证： MIT

内容归原作者所有。SourceWeft 从公开仓库中收录这些内容。

举报或申请下架

Index Type	Build Time	Search Time	Memory	Accuracy
Flat	Fast	Slow	High	100%
IVF	Medium	Fast	Medium	95-99%
HNSW	Slow	Fastest	High	99%
PQ	Medium	Fast	Low	90-95%

Faiss

FAISS - Efficient Similarity Search

When to use FAISS

Quick start

Installation

Basic usage

Index types

1. Flat (exact search)

2. IVF (inverted file) - Fast approximate

3. HNSW (Hierarchical NSW) - Best quality/speed

4. Product Quantization - Memory efficient

Save and load

GPU acceleration

LangChain integration

LlamaIndex integration

Best practices

Performance

Resources

来源与署名

更多来自 orchestra-research/ai-research-skills 的技能

Speculative Decoding

Systems Paper Writing

Moe Training

Presenting Conference Talks

Model Pruning

Outlines

更多AI & Agents技能

Ppt Template Creator

Skill Development

Plugin Structure

Command Development

Claude Md Improver

Find Skills