ResultOn_hackaton

Project arhitecture

Backend

For local development and further deployment there is a docker-compose.yml.

It reads parameters from an .env file. As it includes secrets, it is only an example file and parameters should be added manually before running services.

Currently next services are configured:

• NGINX
• PostreSQL
• MinIO
• RabbitMQ
• MLFlow
• ChromaDB

NGINX

Dummy proxy for testing server accessibility.

PostgreSQL

Stores MLflow backend data.

MinIO

S3 server for development phase. Will be migrated to Arnes Shramba.

Used for saving artifacts from MLFlow, as such model training data and model registry.

RabbitMQ

Message broker, used for log sending from ML scripts.

MLFlow

MLOps platform, used for saving trainings from classification model and model registry.

ChromaDB

Save image embeddings generated by Img2Vec model. Can use cosine singularity to find similarities between images (their embeddings)

Arnes HPC

Part of the project to run ML scripts. Uses Apptainer for building and running containers.

To build container run build_apptainer_python.sh.

Bash scripts to run ML scripts are located here.

Alpaca

Stable diffusion

MobileNevV3

Train and save MobileNetV3 fine tuned model

Extra documentation

Navodila za oddajo izdelkov iz 1. kroga HackathONa