Skip to content

Get started

Your code.
Your first experiment.

Bring a Docker image, choose the settings to explore, and tell us what to improve. HyperOptimizer runs the search in the cloud.

Find your setup path Open the quickstart

01 Choose your workload

Start with the code
you already use.

Keep your language and libraries. Pick the guide that matches your program.

How Docker integration works

Working with a trading framework?

02 Check one local run

Settings in.
A measured result out.

Before starting a search, run your container locally with one set of settings. It needs to do three things.

Read the full checklist

  1. Accept the parameters

    Each configured flag changes the intended input in your program.

  2. Run once, then exit

    Include the dependencies and data it needs. No interactive prompts.

  3. Print a numeric metric

    Report the actual result as hpo.metrics.<name>=<number>.

One local trialIllustrative command
docker run --rm my-workload:latest \
  --hpo-learning-rate=0.01
Your program prints
hpo.metrics.validation_loss=0.14

Example values only. Use your own image and measured result; the metric name must match your experiment objective.

03 Launch your search

Two ways to launch.
One experiment configuration.

Work from your terminal or use the dashboard. Both run on HyperOptimizer’s cloud and keep your settings, metrics, and logs together.

Keep the experiment with your code.

Install the HPO CLI and open your workload repository. You’ll need an account, a project, and Docker with Buildx for local builds.

  1. 1
    Connect your account

    Log in, then initialize a configuration for your project.

  2. 2
    Edit hpo.toml

    Set your image build, parameter ranges, objective, and run limits.

  3. 3
    Validate and launch

    With build settings, the CLI builds and publishes your image, then starts the cloud experiment.

CLI installation guide

Currently installed from source; requires a compatible API and dashboard. The installation guide covers availability and setup.

TerminalAfter CLI installation
hpo login
hpo init
# Edit hpo.toml for your workload
hpo fmt
hpo validate
hpo optimize
hpo.tomlYour experiment, in a file

Build settings · Parameters · Objectives · Run limits

Explore the configuration

The build is local. The experiment runs in our cloud.

Configure the search in your browser.

Create a project, then choose a published image. Set up your experiment in the dashboard or import an existing hpo.toml.

  1. 1
    Publish your Docker image

    Make it available in your workspace registry before you launch.

  2. 2
    Define the search

    Choose parameter ranges and which metrics to maximize or minimize.

  3. 3
    Review and launch

    Check your machine size, run limits, and available compute.

Open the dashboard Publish an image
Experiment setupWorkflow overview
  1. Docker imageA published image from your registry
  2. ParametersThe ranges you want to explore
  3. ObjectivesYour metric names and directions
  4. Run limitsTrial count, parallel runs, and timeout

Review your configuration, then launch.

Bring your hpo.toml

Import settings, or export them to keep with your code. Local build instructions remain metadata; dashboard launches need a published image.

Import and export guide

Once the first results arrive

Check one result.
Then widen your search.

Open a completed trial. Check its parameters, metric, and logs before increasing the number of runs.

Open your dashboard Troubleshoot a trial