README.md

Kepler Power Model

Get started with Kepler Model Server.

This repository contains source code related to Kepler power model. The modules in this repository connects to core Kepler project and kepler-model-db as below.

For more details, check the component diagram.

Model server and estimator deployment

Using Kepler Operator

apiVersion: kepler.system.sustainable.computing.io/v1alpha1
kind: KeplerInternal
metadata:
  name: kepler
spec:
..
  modelServer:
    enabled: <true|false>
  estimator:
    node:
      components:
        sidecar: <true|false>
        initUrl: <static model URL>
      total:
        sidecar: <true|false>
        initUrl: <static model URL>

Using manifests with setup script

Deploy with estimator sidecar

OPTS="ESTIMATOR" make deploy

Deploy with estimator sidecar and model server

OPTS="ESTIMATOR SERVER" make deploy

Model Training

• Use Tekton pipeline
• Use Bash script with CPE operator

Local test

Via docker

1. Build image for testing, run

   make build-test

2. Run the test

   Test case	Command
   Training pipeline	make test-pipeline
   Model server	make test-model-server
   Estimator	make test-estimator
   Offline Trainer	make test-offline-trainer

   For more test information, check here.

With native python environment

Compatible version: python 3.10

1. Install hatch

2. Prepare environment

   hatch shell

3. Run the test

Test case	Command
Training pipeline	python -u ./tests/pipeline_test.py
Model server	Terminal 1: export MODEL_PATH=$(pwd)/tests/models;python src/server/model_server.py Terminal 2: python -u tests/estimator_model_request_test.py
Estimator	Terminal 1: python src/estimate/estimator.py Terminal 2: python -u tests/estimator_power_request_test.py
Estimator with Model Server	Terminal 1: export MODEL_PATH=$(pwd)/tests/models;python src/server/model_server.py Terminal 2: export MODEL_SERVER_URL=http://localhost:8100;export MODEL_SERVER_ENABLE=true;python -u src/estimate/estimator.py Terminal 3: python -u tests/estimator_power_request_test.py
Offline Trainer	Terminal 1: python src/train/offline_trainer.py Terminal 2: python -u tests/offline_trainer_test.py

For more test information, check here.

Contributing

Please check the roadmap and guidelines to join us here.