What is it ?

A tribe project to build a Model Platform

Intentions

1. Build a project as a team
2. Test, identify, consolidate convictions on model management at scale
3. Be as portable as possible
4. Use opensource technologies

Vision and key features of a Model Platform

Vision
For models developers, so that they can focus on building the best model, we offer a model platform that can version, deploy, host, govern models with as few configuration as possible
For application developers, so that they can integrate models seamlessly, we offer a model platform that can provide a simple API to call.

How to run

- Setup your K8S env

A working minikube is needed for dev purposes

brew install minikube kubectl

Install helm

brew install helm

- Start and setup Cluster

Start minikube cluster with recommended specs

#recommended configuration to avoid freezing/timeouts
minikube start --cpus 2 --memory 7800 --disk-size 50g

(You may need to start colima with custom config)

colima start --cpu 4 --memory 8

Activate Ingress controller add-on

minikube addons enable ingress

Activate ssh tunnel to you minikube cluster

minikube tunnel

- "DNS" setup

Add the following line to your /etc/hosts

# Mac
127.0.0.1 model-platform.com
# Linux
minikube ip
IP.RESULT model-platform.com

- Fill .env file

Some values are used (pgsql password etc...) in the init scripts !!

Use the .env.example 

- Setup Namespaces, NGINX, Ingress

make k8s-network-conf

- Deploy a PGSQL instance with helm

Add bitnami pgsql repo to helm

helm repo add bitnami https://charts.bitnami.com/bitnami
helm repo update

make k8s-pgsql

It will output a lot of logs, don't worry.

- Prepare a custom MLFLow Docker image

With integrated psycopg2 package

    make build-mlflow

You should now have a working k8s environment

Deploy the model platform en k8s

  make k8s-modelplatform

Connect to model platform

Via fronted

http://model-platform.com

or

mp login --username XXXXX --password XXXX

Root account is the one you set in the .env file

TROUBLESHOOTING

Backend cannot find docker executable

Error executing docker build: [Errno 2] No such file or directory: 'docker'

Run

minikube docker-env
#Check if same in .env
if not update and run 
k8s-modelplatform

Dev expérience

Running CI locally

You will need to install nektos act https://nektosact.com/installation/

#mac ARM
make run-ci-arm

or

#Intel processors
make run-ci-amd

Access the pgsql via local db client

Run

kubectl port-forward svc/modelplatform-pgsql-postgresql 5432:5432 -n pgsql

Add to you db client

host: localhost
port: 5432
user: postgres
password: your_postgres_password
db: model_platform_db

you can now access the model_platform_db database