minimal gpu custom embedded endpoint for weaviate vectorsearch with multi-qa-MiniLM-L6-cos-v1
conda create --name vsearch
conda activate vsearch
pip install "fastapi[all]"
conda install -c conda-forge sentence-transformers
conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia
this will take a while...
git clone https://github.com/saeneas/custom-vsearch.git
cd custom-vsearch
uvicorn main:app --reload
Query with
curl localhost:8000/vectors/ -H "Content-Type: application/json" -d '{"text":"give me vecs"}'