quotify-backend

About

API for the Quotify webapp to generate quotes using a finetuned GPT2 model. The model can be downloaded from the releases page. The backend was built using FastAPI and deployed with Docker on Google Cloud Platform.

This was the best method we managed to find to deploy large models (>500MB) to the cloud.

API Docs: https://quotify-engine-l6lhxur2aq-uc.a.run.app/docs

Set up Locally

1. Clone Repo git clone https://github.com/Quotify-Bot/quotify-backend.git
2. Change directory cd quotify-backend
3. Download the model named pytorch_model.bin from releases and add it to the finetuned_models directory
4. Install virtual environment virtualenv env
5. Activate environment env\Scripts\activate
6. Install requirements pip install -r requirements.txt
7. Install pytorch cpu pip install torch==1.7.1+cpu -f https://download.pytorch.org/whl/torch_stable.html
8. Start the server uvicorn main:app --host 0.0.0.0 --port 8080

Deploy to Google Cloud Platform (GCP)

1. Create docker image docker build -t <image_name>:<tag_name> .
2. Login using gcloud CLI gcloud auth login
3. Tag the image in the correct format for deployment docker tag <image_name>:<tag_name> gcr.io/<project_name>/<image_name>:<tag_name>
4. Push to GCP container registry docker push gcr.io/<project_name>/<image_name>:<tag_name>
5. Go to GCP container registry and deploy using cloud run