Setup

1. Git clone the repository.

git clone repo_url

2. Optional: download uv and install it in your environment. You can also use pip to install the requirements.

pip install uv

3. Create a virtual environment and activate it.

uv venv # or python -m venv .venv
source .venv/bin/activate

4. Compile requirements based on your environment.

uv pip compile builder/requirements.in -o builder/requirements.txt # uv is optional, but recommended

5. Install the requirements.

uv pip sync builder/requirements.txt

6. Download the models from huggingface and save them in the models_hub directory before building. See src/download_models.py for more details.

python3 src/download_models.py

7. Run the service locally using the following command.

python3 src/handler.py --rp_serve_api

8. The Embedding service is now running on http://localhost:8000/. You can test it using the following command.

curl --request POST \
  --url http://localhost:8000/runsync \
  --header 'Content-Type: application/json' \
  --data '{"input": {"task": "query","input_data": ["hello"]}}'  

Production

The service is deplyed via Docker. The Dockerfile is located in the root directory. Due to GPU requirements, the service works well either on a mac chips with MPS backend (local development) or on a linux machine with CUDA installed (production).

Commands

1. Test locally without docker (recommended on MacOS)

python src/handler.py --rp_serve_api

2. Build and publish image version is usually date e.g. 20240930 or 20240930

./build_publish.sh {org} {repo} 

4. Push image to docker hub manually (instead of using buld_publish.sh - aka, you want a specific version)

docker push {org}/{repo}:version

Hosting

We use runpod serverless endpoint. The Docker image is hosted on Docker Hub under: tjmlabs/colivare