Optimized Quantized Graph (OQG)

Install

First, install Anaconda.
Then start to install oqglib:

sudo apt install gcc-14
git clone https://github.com/TheDatumOrg/OQG.git
conda create -n oqg python==3.10 
conda activate oqg
conda install -c conda-forge openblas  
pip install cmake==3.28.3
conda install -c pytorch/label/nightly faiss-cpu  
pip install pybind11==3.0.1
pip install tqdm==4.67.1

cd python_bindings
pip install --no-build-isolation .

Benchmarks

First, please check the taskset command in example/train_test.sh, and adjust it with the number of available CPU cores you have.

cd ../example
bash train_test.sh

Now you can check the result at train.csv and test_k100.csv.
Please check example/train.py and example/test.py for details.

Set your own datasets

cd dataset
mkdir $YOUR_DATASET_NAME
cd $YOUR_DATASET_NAME

Then upload your base.fvecs, groundtruth.ivecs, query.fvecs into $YOUR_DATASET_NAME.
Then open example/dataset/dataset_config.py and config there.

Search Performance