How Do I Install Faiss on Linux?

by

If you’re working on large-scale similarity search or machine learning tasks involving nearest neighbor search, you’ve probably heard of Faiss (Facebook AI Similarity Search). Developed by Facebook AI Research, Faiss is a powerful open-source library designed for efficient similarity search and clustering of dense vectors, especially when dealing with high-dimensional data.

Whether you’re building a recommendation engine, search engine, or large-scale NLP application, Faiss can be a game-changer. But one common question many developers and data scientists ask is:
“How do I install Faiss on Linux?”

In this comprehensive guide, we’ll walk you through various ways to install Faiss on Linux in 2025 — whether you’re using pip, conda, or building from source. This guide is beginner-friendly, up to date, and includes tips for troubleshooting and optimizing your installation.

What Is Faiss and Why Use It?

Faiss is a library that allows for fast nearest neighbor search in high-dimensional spaces. It is widely used in natural language processing (NLP), computer vision, and recommendation systems.

Key features of Faiss:

  • Highly optimized for performance
  • Supports both CPU and GPU
  • Works with float32 and int8 vector types
  • Scales to billions of vectors
  • Provides support for quantization and clustering

Now that we understand why Faiss is useful, let’s look at how to get it running on a Linux system.

Pre-Installation Checklist

Before you begin, make sure your Linux system has the following:

  • Python 3.7+ (Python 3.8 or newer recommended)
  • pip or conda installed
  • GCC / G++ (for building from source)
  • CUDA Toolkit (if using GPU-enabled Faiss)
  • CMake (required for source build)

You can check if Python and pip are installed by running:

python3 --version
pip3 --version

And for conda:

conda --version

If you’re missing any of these, install them using your system’s package manager or download them from the official websites.

Method 1: Installing Faiss with pip (CPU-only)

The simplest way to install Faiss on Linux — especially for CPU-only use — is via pip.

Step-by-Step Instructions

  1. Create a virtual environment (recommended):
python3 -m venv faiss-env
source faiss-env/bin/activate
  1. Upgrade pip:
pip install --upgrade pip
  1. Install Faiss (CPU-only):
pip install faiss-cpu

Pros:

  • Easy and fast
  • No need to compile anything

Cons:

  • Does not support GPU

Method 2: Installing Faiss with Conda (CPU and GPU)

If you want GPU acceleration or need more stability, conda is a better choice. Faiss is available through the official conda-forge channel.

Step-by-Step Instructions

  1. Create a new conda environment:
conda create -n faiss-env python=3.9 -y
conda activate faiss-env
  1. Add the conda-forge channel:
conda config --add channels conda-forge
  1. Install Faiss (CPU):
conda install faiss-cpu -c conda-forge

Or, for GPU support:

conda install faiss-gpu cudatoolkit=11.3 -c conda-forge

⚠️ Make sure the CUDA version you install matches the version on your system (nvidia-smi can help you check).

Pros:

  • Supports both CPU and GPU
  • Stable and well-maintained builds
  • Easier dependency management

Cons:

  • Larger installation footprint
  • Requires conda

Method 3: Building Faiss from Source (Advanced)

This method is best if you need the latest features from the Faiss GitHub repository or want custom modifications.

Step-by-Step Instructions

  1. Install system dependencies:
sudo apt update
sudo apt install build-essential cmake libopenblas-dev libomp-dev git python3-dev python3-pip
  1. Clone the Faiss repository:
git clone https://github.com/facebookresearch/faiss.git
cd faiss
  1. Set up a Python environment:
python3 -m venv venv
source venv/bin/activate
pip install numpy cython
  1. Build Faiss (CPU-only):
cmake -B build -DFAISS_ENABLE_PYTHON=ON -DFAISS_ENABLE_GPU=OFF .
make -C build -j
cd build/faiss/python
python setup.py install
  1. For GPU Support (optional):

Install CUDA (if not already installed), then build with:

cmake -B build -DFAISS_ENABLE_PYTHON=ON -DFAISS_ENABLE_GPU=ON -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc .
make -C build -j
cd build/faiss/python
python setup.py install

Pros:

  • Full control over the build
  • Access to latest GitHub features

Cons:

  • More complex
  • Requires compilation and troubleshooting

Post-Installation: Verifying Your Installation

After installing, verify that Faiss is working:

import faiss
print(faiss.__version__)

You should see the installed version without any errors.

You can also run a quick test:

import faiss
import numpy as np

d = 64  # dimension
nb = 1000  # database size
nq = 5  # number of queries

xb = np.random.random((nb, d)).astype('float32')
xq = np.random.random((nq, d)).astype('float32')

index = faiss.IndexFlatL2(d)  # L2 distance index
index.add(xb)
D, I = index.search(xq, 5)

print(I)  # indices of nearest neighbors

Common Installation Issues and Fixes

Issue: ModuleNotFoundError: No module named 'faiss'
Fix: Double-check your environment and make sure you installed the correct package (faiss-cpu or faiss-gpu).

Issue: CUDA not detected or mismatched
Fix: Use nvidia-smi to verify your installed CUDA version and match it with the one in your conda or build settings.

Issue: Compilation errors during source build
Fix: Ensure that dependencies like CMake, OpenBLAS, and Python headers are properly installed.

Conclusion

So, how do I install Faiss on Linux? The answer depends on your system setup, performance requirements, and preference for ease versus customization.

Here’s a quick summary:

Installation MethodGPU SupportDifficultyRecommended For
pip install faiss-cpu❌EasyBeginners, CPU-only projects
conda install faiss-gpu✅ModerateMost users, GPU-enabled apps
Build from source✅AdvancedDevelopers, Custom builds

No matter which method you choose, Faiss offers blazing-fast performance and versatility for vector similarity search tasks. Now that you have it installed, you’re ready to build applications that scale to millions (or billions) of vectors!

Leave a Comment