dstack makes it very easy for ML engineers to run dev environments, pipelines and apps cost-effectively on any cloud.

To use dstack, install it with pip and start the Hub application.

pip install dstack
dstack start

The dstack start command starts the Hub server, and creates the default project to run everything locally.

To enable Hub to run dev environments, pipelines, and apps in your preferred cloud account (AWS, GCP, Azure, etc), log in to Hub, and configure the corresponding project.

A dev environment is a virtual machine that includes the environment and an interactive IDE or notebook setup based on a pre-defined configuration.

Go ahead and define this configuration via YAML (under the .dstack/workflows folder).

workflows:
  - name: code-gpu
    provider: code
    setup:
      - pip install -r dev-environments/requirements.txt
    resources:
      gpu:
        count: 1

The YAML file allows you to configure hardware resources, set up the Python environment, expose ports, configure cache, and many more.

Now, you can start it using the dstack run command:

$ dstack run code-gpu

RUN      WORKFLOW  SUBMITTED  STATUS     TAG
shady-1  code-gpu  now        Submitted  
 
Starting SSH tunnel...

To exit, press Ctrl+C.

Web UI available at http://127.0.0.1:51845/?tkn=4d9cc05958094ed2996b6832f899fda1

If you configure a project to run dev environments in the cloud, dstack will automatically provision the required cloud resources, and forward ports of the dev environment to your local machine.

When you stop the dev environment, dstack will automatically clean up cloud resources.

A pipeline is a set of pre-defined configurations that allow to process data, train or fine-tune models, do batch inference or other tasks.

Go ahead and define such a configuration via YAML (under the .dstack/workflows folder).

workflows:
  - name: train-mnist-gpu
    provider: bash
    commands:
      - pip install -r pipelines/requirements.txt
      - python pipelines/train.py
    artifacts:
      - ./lightning_logs
    resources:
      gpu:
        count: 1

The YAML file allows you to configure hardware resources and output artifacts, set up the Python environment, expose ports, configure cache, and many more.

Now, you can run the pipeline using the dstack run command:

$ dstack run train-mnist-gpu

RUN      WORKFLOW         SUBMITTED  STATUS     TAG
shady-1  train-mnist-gpu  now        Submitted  
 
Provisioning... It may take up to a minute. ✓

GPU available: True, used: True

Epoch 1: [00:03<00:00, 280.17it/s, loss=1.35, v_num=0]

If you configure a project to run pipelines in the cloud, the dstack run command will automatically provision the required cloud resources.

After the pipeline is stopped or finished, dstack will save output artifacts and clean up cloud resources.

An app can be either a web application (such as Streamlit, Gradio, etc.) or an API endpoint (like FastAPI, Flask, etc.) setup based on a pre-defined configuration.

Go ahead and define this configuration via YAML (under the .dstack/workflows folder).

workflows:
  - name: fastapi-gpu
    provider: bash
    ports: 1
    commands:
      - pip install -r apps/requirements.txt
      - uvicorn apps.main:app --port $PORT_0 --host 0.0.0.0
    resources:
      gpu:
        count: 1

The configuration allows you to customize hardware resources, set up the Python environment, configure cache, and more.

Now, you can run the app using the dstack run command:

$ dstack run fastapi-gpu
 RUN           WORKFLOW     SUBMITTED  STATUS     TAG
 silly-dodo-1  fastapi-gpu  now        Submitted     

Starting SSH tunnel...

To interrupt, press Ctrl+C.

INFO:     Started server process [1]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://127.0.0.1:63475 (Press CTRL+C to quit)

If you configure a project to run apps in the cloud, dstack will automatically provision the required cloud resources, and forward ports of the app to your local machine. If you stop the app, it will automatically clean up cloud resources.

For additional information and examples, see the following links:

• Docs
• Examples
• Blog
• Slack

Mozilla Public License 2.0