README.md

    Build Releases

    DI Orchestrator

    DI Orchestrator is designed to manage DI (Decision Intelligence) jobs using Kubernetes Custom Resource and Operator.

    Prerequisites

    • A well-prepared kubernetes cluster. Follow the instructions to create a kubernetes cluster, or create a local kubernetes node referring to kind or minikube

    Install DI Orchestrator

    DI Orchestrator consists of two components: di-operator and di-server. Install them with the following command.

    kubectl create -f ./config/di-manager.yaml

    di-operator and di-server will be installed in di-system namespace.

    $ kubectl get pod -n di-system
    NAME                               READY   STATUS    RESTARTS   AGE
    di-operator-57cc65d5c9-5vnvn       1/1     Running   0          59s
    di-server-7b86ff8df4-jfgmp         1/1     Running   0          59s

    Submit DIJob

    # submit DIJob
    $ kubectl create -f config/samples/dijob-gobigger.yaml
    
    # get pod and you will see coordinator is created by di-operator
    # a few seconds later, you will see collectors and learners created by di-server
    $ kubectl get pod
    NAME                READY   STATUS    RESTARTS   AGE
    gobigger-test-0-0   1/1     Running   0          4m17s
    gobigger-test-0-1   1/1     Running   0          4m17s
    
    # get logs of coordinator
    $ kubectl logs -n xlab gobigger-test-0-0
    Bind subprocesses on these addresses: ['tcp://10.148.3.4:22270',
    'tcp://10.148.3.4:22271']
    [Warning] no enough data: 128/0
    ...
    [Warning] no enough data: 128/120
    Current Training: Train Iter(0) Loss(102.256)
    Current Training: Train Iter(0) Loss(103.133)
    Current Training: Train Iter(20)        Loss(28.795)
    Current Training: Train Iter(20)        Loss(32.837)
    ...
    Current Training: Train Iter(360)       Loss(12.850)
    Current Training: Train Iter(340)       Loss(11.812)
    Current Training: Train Iter(380)       Loss(12.892)
    Current Training: Train Iter(360)       Loss(13.621)
    Current Training: Train Iter(400)       Loss(15.183)
    Current Training: Train Iter(380)       Loss(14.187)
    Current Evaluation: Train Iter(404)     Eval Reward(-1788.326)

    User Guide

    Refers to user-guide. For Chinese version, please refer to 中文手册

    Contributing

    Refers to developer-guide.

    Contact us throw opendilab.contact@gmail.com

    项目简介

    OpenDILab RL Kubernetes Custom Resource and Operator Lib

    🚀 Github 镜像仓库 🚀

    源项目地址

    https://github.com/opendilab/DI-orchestrator

    发行版本 5

    v0.2.2

    全部发行版

    贡献者 4

    L liqingping @liqingping
    K Konnase Lee @Konnase Lee
    L lijianwen @lijianwen
    W wubaodong @wubaodong

    开发语言

    • Go 95.4 %
    • Makefile 3.5 %
    • Dockerfile 0.7 %
    • Shell 0.4 %