Skip to content

Latest commit

 

History

History
184 lines (141 loc) · 5.08 KB

README.md

File metadata and controls

184 lines (141 loc) · 5.08 KB

VectorSearchForge

Setting Up Environment on a GPU

  1. Install miniconda on the machine using link: https://docs.anaconda.com/free/miniconda/#quick-command-line-install
  2. Get latest conda dependencies list from https://github.com/rapidsai/raft file conda/environments/all_cuda-122_arch-x86_64.yaml in your local, or if you need all the correct dependenices you can use the linux_x64_cuda12_raft24_06.yml
  3. Run the below command to create then env with name raft_246, you can use any name here.
conda env create --name raft_246 --file all_cuda-122_arch-x86_64-24-06.yaml

or

conda env create --name raft_246 --file linux_x64_cuda12_raft24_06.yml

3.(Not required if you use linux_x64_cuda12_raft24_06.yml in step 2) Install right version of libraft in the env. Use rapidsai-nightly to get latest build and rapidsai for stable builds. Use Optional-Libraft-Version to set version of libraft you need.

conda install -y -q libraft=<Optional-Libraft-Version> -c rapidsai-nightly  -c conda-forge
  1. Activate the env.
conda activate raft_246

Updating Submodule

git submodule update --remote

Build Package on GPU

Common for both C++ and Python

Replace ENV_NAME with conda env name in the command.

cmake -B build . -DCMAKE_BUILD_TYPE=Release -DFAISS_ENABLE_RAFT=ON  -DBUILD_SHARED_LIBS=ON -DFAISS_ENABLE_PYTHON=ON -GNinja -DCMAKE_CUDA_ARCHITECTURES=80 -DCMAKE_INSTALL_PREFIX=${CONDA_PREFIX} -DFAISS_ENABLE_GPU=ON -DCUDAToolkit_ROOT="/home/ubuntu/miniconda3/envs/<ENV_NAME>/lib"

Python

pip install numpy  swig==4.0.0 h5py psutil

for versions >=4.2.0 the faiss swig build will fail. GH issue: facebookresearch/faiss#3239

ninja -C build -j10 install faiss swigfaiss


cd build/external/faiss/faiss/python && python3 setup.py build

Now go to the root of the project.

export PYTHONPATH="$(ls -d `pwd`/build/external/faiss/faiss/python/build/lib*/):`pwd`/"

Setting up env for CPU Machine

  1. Run the below commands to install the correct packages. Some of these things are optional(like tmux)
sudo yum install gcc-c++ gcc g++ tmux git zlib-devel openblas-devel gfortran python3-devel -y
  1. Install miniconda on the machine using link: https://docs.anaconda.com/free/miniconda/#quick-command-line-install
  2. Run the below command to create then env with name faiss-cpu, you can use any name here.
conda env create --name faiss-cpu --file linux_arm_cpu.yml
  1. Now activate the env
conda activate faiss-cpu

Build Package on CPU

Common for both C++ and Python

Replace ENV_NAME with conda env name in the command.

cmake -B build . -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DFAISS_ENABLE_PYTHON=ON -DFAISS_ENABLE_GPU=OFF -DFAISS_ENABLE_RAFT=OFF"

Python

pip install numpy  swig==4.0.0 h5py psutil

for versions >=4.2.0 the faiss swig build will fail. GH issue: facebookresearch/faiss#3239

make -C build -j faiss swigfaiss

cd build/external/faiss/faiss/python && python3 setup.py build

Now go to the root of the project.

export PYTHONPATH="$(ls -d `pwd`/build/external/faiss/faiss/python/build/lib*/):`pwd`/"

Running Benchmarks

Setup

export PYTHONPATH="$(ls -d `pwd`/build/external/faiss/faiss/python/build/lib*/):`pwd`/"

Indexing GPU

python python/main.py --workload=sift-128 --index_type=gpu --workload_type=index

Indexing CPU

python python/main.py --workload=sift-128 --index_type=cpu --workload_type=index

Run Both indexing and search

GPU

python python/main.py --workload=gist-960 --index_type=gpu --workload_type=index_and_search

CPU

python python/main.py --workload=gist-960 --index_type=cpu --workload_type=index_and_search

Exporting All results as CSV

python python/results.py --workload=all --index_type=gpu --workload_type=index_and_search

After this command the results will be stored under results/all/all_results.csv

Old

Setup C++

Building all the CPP files on CPU and Run simple test

cmake -B cmake-build-debug -DCMAKE_BUILD_TYPE=Release -GNinja .  -DFAISS_ENABLE_PYTHON=OFF

cmake --build cmake-build-debug --target faiss-test -j 10

./cmake-build-debug/cpp/faiss-test

CPU Python

conda create -n faiss-cpu python=3.11.8

Not working For CPU

cmake -B build -DCMAKE_BUILD_TYPE=Release -GNinja .  -DFAISS_ENABLE_PYTHON=ON -DBUILD_SHARED_LIBS=ON -DFAISS_ENABLE_GPU=OFF -GNinja

/Applications/CLion.app/Contents/bin/ninja/mac/x64/ninja -C cmake-build-debug -j4 install


cd build/external/faiss/faiss/python && python3 setup.py build

cd cmake-build-debug/external/faiss/faiss/python

Working for CPU

cmake -B build . -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DFAISS_ENABLE_PYTHON=ON -DFAISS_ENABLE_GPU=OFF -DFAISS_ENABLE_RAFT=OFF
make -C build -j faiss swigfaiss

cd build/external/faiss/faiss/python && python3 setup.py build

export PYTHONPATH="$(ls -d `pwd`/build/external/faiss/faiss/python/build/lib*/):`pwd`/:$(/usr/local/dcgm/bindings/python3)"