open source platform for simplifying machine learning deployment

Train, serve and predict in your own built running cluster with 8 commands!


  ...

  Warning - Performing force build of the backend
  Warning - Are you sure about building kaos backend in DOCKER? [y/N]

  ...

  Apply complete! Resources: 21 added, 0 changed, 0 destroyed.

  Info - Endpoint successfully set to http://localhost:80/api/
  Info - Successfully built kaos environment
          

  Info - Successfully set mnist workspace
          

  Info - Successfully loaded mnist template
          

  Info - Submitting source bundle: templates/mnist/model-train
  Compressing source bundle: 100%|███████████████████████████|
  Uploading source bundle: 100%|███████████████████████████|
  ✔ Setting source bundle: /mnist:1909d

  Info - Submitting data bundle: templates/mnist/data
  Compressing data bundle: 100%|███████████████████████████|
  Uploading data bundle: 100%|███████████████████████████|
  ✔ Setting data bundle: /features:5c3ca

  CURRENT TRAINING INPUTS

  +------------+-----------------+-------------+
  |   Image    |       Data      | Hyperparams |
  +------------+-----------------+-------------+
  |     -      |        ✔        |      ✗      |
  | <building> | /features:5c3ca |             |
  +------------+-----------------+-------------+
          

  Info - Retrieving info from eb9617c2032f46e99de48eeac358231b

    Job ID: eb9617c2032f46e99de48eeac358231b
    Process time: 130s
    State: JOB_SUCCESS
    Available metrics: ['accuracy_test', 'accuracy_validation', 'accuracy_train']

    Page count: 1
    Page ID: 0
  +-----+--------------------+-----------------------+--------------------+
  | ind |        Code        |          Data         |      Model ID      |
  +-----+--------------------+-----------------------+--------------------+
  |  0  | Author: jfriedman  |   Author: jfriedman   | 1909d_5c3ca:828c4e |
  |     | Path: /mnist:1909d | Path: /features:5c3ca |                    |
  +-----+--------------------+-----------------------+--------------------+
          

  Submitting source bundle: templates/mnist/model-serve
  Compressing source bundle: 100%|███████████████████████████|
    Uploading source bundle: 100%|███████████████████████████|
  ✔ Adding trained model_id: 1909d_5c3ca:828c4e
  ✔ Setting source bundle: /mnist:e39c0
          

  +----------------------------------------------------------------------+
  |                              running                                 |
  +---------------------+-----------+------------------------------------+
  |      created_at     |    user   |                   url              |
  +---------------------+-----------+------------------------------------+
  | 2019-07-25 19:56:26 | jfriedman | localhost/mnist-53d3da/invocations |
  +---------------------+-----------+------------------------------------+
          

  {"result":[3]}
          
  • 1. Build Cluster

  • 2. Create a Workspace

  • 3. Load the MNIST Template

  • 4. Deploy a Training Job

  • 5. Identify the Trained Model

  • 6. Serve the Trained Model

  • 7. Identify the Endpoint URL

  • 8. Predict

What is kaos?

kaos is the platform for deploying scalable reproducible machine learning workflows in your own private environment.

Why kaos?

The development of kaos was fueled by our ambition to mimic natural incremental model development, simplify model reproducibility and collaboration, and automate ML infrastructure deployment in a flexible language-agnostic environment.

Data Scientist

  • Completely flexible.
    Use any language and any framework.
  • Forget forgetting.
    Automatic reproducibility of all artifacts.
  • Just train.
    Deploy without thinking of any ML infrastructure.
  • Scale when desired.
    Elastic training with a single command.

Team Lead

  • Focus on insights.
    Avoid ML infrastructure hurdles.
  • Knowledge sharing.
    Reproducibility enables any team member to contribute.
  • Deliver confidently.
    Run immediately in own cluster withoutuncertainty.
  • Control resources.
    Balance infrastructure needs with available budget.

Management

  • Keep knowledge in-house.
    Own cloud independent ML platform.
  • Push innovation.
    Develop without focusing on infrastructure assets.
  • Freedom to pivot.
    Migrate ML platform to any cloud.
  • Reduce organizational complexity.
    Avoid knowledge fragmentation between teams.

Features

Simplify deploying, building and serving machine learning models

  • Open Source

    Open Source

    Core built with language-agnostic data pipelines

  • Scalable Workflows

    Scalable Workflows

    Train unlimited models in parallel

  • Incremental Development

    Incremental Development

    Train models with improved code, data and/or params

  • Minimal DevOps

    Minimal DevOps

    Automated infrastructure deployment with IaC

  • Cloud Agnostic

    Cloud Agnostic

    Runs with kubernetes in any cloud

  • Multiple Personas

    Multiple Personas

    Build infrastrcture vs. interact with workflows

  • Fully Versioned

    Fully Versioned

    Records all changes to data, code and pipelines

  • Fully Reproducible

    Fully Reproducible

    Ensures that any input and/or output can be recreated

  • Private

    Private

    Deployed and maintained in own private environment

Getting Started