Breakfast

Breakfast is a kubernetes custom resource operator. It manages the life cycle of running Machine Learning pods through custom controllers.

Get Started

Preparations

• Kubernetes v1.15+
• SCV has been deployed.
• Yoda-Scheduler has been deployed.

Deploy Breakfast

kubectl apply -f https://raw.githubusercontent.com/NJUPT-ISL/Breakfast/master/deploy/breakfast.yaml

Create Machine Learning Task

• Create a ML task with Tensorflow Framework

apiVersion: core.run-linux.com/v1alpha1
kind: Bread
metadata:
  name: test
  namespace: root
spec:
  scv:
    gpu: "1"
    memory: "4000"
    level: "Medium"
  framework:
    name: "tensorflow"
    version: "2.0"
  task:
    type: train
    path: "/root"
    command: "python /root/test.py"

• Create a ML task with ssh service

apiVersion: core.run-linux.com/v1alpha1
kind: Bread
metadata:
  name: test
spec:
  scv:
    gpu: "1"
    memory: "4000"
    level: "Medium"
  framework:
    name: "tensorflow"
    version: "2.0"
  task:
    type: ssh
    path: "/root"
    command: ""

How to Dev

• Build the Breakfast

make 

• Build the Docker Image

make docker-build

• Debug the Breakfast Controller

make run

Contact us