Skip to content

Cli

zenml.cli special

ZenML CLI.

The ZenML CLI tool is usually downloaded and installed via PyPI and a pip install zenml command. Please see the Installation & Setup section above for more information about that process.

How to use the CLI

Our CLI behaves similarly to many other CLIs for basic features. In order to find out which version of ZenML you are running, type:

   zenml version

If you ever need more information on exactly what a certain command will do, use the --help flag attached to the end of your command string.

For example, to get a sense of all the commands available to you while using the zenml command, type:

   zenml --help

If you were instead looking to know more about a specific command, you can type something like this:

   zenml metadata-store register --help

This will give you information about how to register a metadata store. (See below for more on that).

If you want to instead understand what the concept behind a group is, you can use the explain sub-command. For example, to see more details behind what a metadata-store is, you can type:

zenml metadata-store explain

This will give you an explanation of that concept in more detail.

Beginning a Project

In order to start working on your project, initialize a ZenML repository within your current directory with ZenML’s own config and resource management tools:

zenml init

This is all you need to begin using all the MLOps goodness that ZenML provides!

By default, zenml init will install its own hidden .zen folder inside the current directory from which you are running the command. You can also pass in a directory path manually using the --repo_path option:

zenml init --repo_path /path/to/dir

If you wish to specify that you do not want analytics to be transmitted back to ZenML about your usage of the tool, pass in False to the --analytics_opt_in option:

zenml init --analytics_opt_in false

If you wish to delete all data relating to your project from the directory, use the zenml clean command. This will:

  • delete all pipelines
  • delete all artifacts
  • delete all metadata

Note that the clean command is not implemented for the current version.

Loading and using pre-built examples

If you don’t have a project of your own that you’re currently working on, or if you just want to play around a bit and see some functional code, we’ve got your back! You can use the ZenML CLI tool to download some pre-built examples.

We know that working examples are a great way to get to know a tool, so we’ve made some examples for you to use to get started. (This is something that will grow as we add more).

To list all the examples available to you, type:

zenml example list

If you want more detailed information about a specific example, use the info subcommand in combination with the name of the example, like this:

zenml example info quickstart

If you want to pull all the examples into your current working directory (wherever you are executing the zenml command from in your terminal), the CLI will create a zenml_examples folder for you if it doesn’t already exist whenever you use the pull subcommand. The default is to copy all the examples, like this:

zenml example pull

If you’d only like to pull a single example, add the name of that example (for example, quickstart) as an argument to the same command, as follows:

zenml example pull quickstart

If you would like to force-redownload the examples, use the --yes or -y flag as in this example:

zenml example pull --yes

This will redownload all the examples afresh, using the same version of ZenML as you currently have installed. If for some reason you want to download examples corresponding to a previous release of ZenML, use the --version or -v flag to specify, as in the following example:

zenml example pull --yes --version 0.3.8

If you wish to run the example, allowing the ZenML CLI to do the work of setting up whatever dependencies are required, use the run subcommand:

zenml example run quickstart

Using integrations

Integrations are the different pieces of a project stack that enable custom functionality. This ranges from bigger libraries like kubeflow for orchestration down to smaller visualization tools like facets. Our CLI is an easy way to get started with these integrations.

To list all the integrations available to you, type:

zenml integration list

To see the requirements for a specific integration, use the requirements command:

zenml integration requirements INTEGRATION_NAME

If you wish to install the integration, using the requirements listed in the previous command, install allows you to do this for your local environment:

zenml integration install INTEGRATION_NAME

Note that if you don't specify a specific integration to be installed, the ZenML CLI will install all available integrations.

Uninstalling a specific integration is as simple as typing:

zenml integration uninstall INTEGRATION_NAME

Customizing your Metadata Store

The configuration of each pipeline, step, backend, and produced artifacts are all tracked within the metadata store. By default, ZenML initializes your repository with a metadata store kept on your local machine. If you wish to register a new metadata store, do so with the register command:

zenml metadata-store register METADATA_STORE_NAME --flavor=METADATA_STORE_FLAVOR [--OPTIONS]

If you wish to list the metadata stores that have already been registered within your ZenML project / repository, type:

zenml metadata-store list

If you wish to delete a particular metadata store, pass the name of the metadata store into the CLI with the following command:

zenml metadata-store delete METADATA_STORE_NAME

Customizing your Artifact Store

The artifact store is where all the inputs and outputs of your pipeline steps are stored. By default, ZenML initializes your repository with an artifact store with everything kept on your local machine. If you wish to register a new artifact store, do so with the register command:

zenml artifact-store register ARTIFACT_STORE_NAME --flavor=ARTIFACT_STORE_FLAVOR [--OPTIONS]

If you wish to list the artifact stores that have already been registered within your ZenML project / repository, type:

zenml artifact-store list

If you wish to delete a particular artifact store, pass the name of the artifact store into the CLI with the following command:

zenml artifact-store delete ARTIFACT_STORE_NAME

Customizing your Orchestrator

An orchestrator is a special kind of backend that manages the running of each step of the pipeline. Orchestrators administer the actual pipeline runs. By default, ZenML initializes your repository with an orchestrator that runs everything on your local machine.

If you wish to register a new orchestrator, do so with the register command:

zenml orchestrator register ORCHESTRATOR_NAME --flavor=ORCHESTRATOR_FLAVOR [--ORCHESTRATOR_OPTIONS]

If you wish to list the orchestrators that have already been registered within your ZenML project / repository, type:

zenml orchestrator list

If you wish to delete a particular orchestrator, pass the name of the orchestrator into the CLI with the following command:

zenml orchestrator delete ORCHESTRATOR_NAME

Customizing your Container Registry

The container registry is where all the images that are used by a container-based orchestrator are stored. By default, a default ZenML local stack will not register a container registry. If you wish to register a new container registry, do so with the register command:

zenml container-registry register REGISTRY_NAME --flavor=REGISTRY_FLAVOR [--REGISTRY_OPTIONS]

If you want the name of the current container registry, use the get command:

zenml container-registry get

To list all container registries available and registered for use, use the list command:

zenml container-registry list

For details about a particular container registry, use the describe command. By default, (without a specific registry name passed in) it will describe the active or currently used container registry:

zenml container-registry describe [REGISTRY_NAME]

To delete a container registry (and all of its contents), use the delete command:

zenml container-registry delete REGISTRY_NAME

Customizing your Experiment Tracker

Experiment trackers let you track your ML experiments by logging the parameters and allowing you to compare between different runs. If you want to use an experiment tracker in one of your stacks, you need to first register it:

zenml experiment-tracker register EXPERIMENT_TRACKER_NAME     --flavor=EXPERIMENT_TRACKER_FLAVOR [--EXPERIMENT_TRACKER_OPTIONS]

If you want the name of the current experiment tracker, use the get command:

zenml experiment-tracker get

To list all experiment trackers available and registered for use, use the list command:

zenml experiment-tracker list

For details about a particular experiment tracker, use the describe command. By default, (without a specific experiment tracker name passed in) it will describe the active or currently-used experiment tracker:

zenml experiment-tracker describe [EXPERIMENT_TRACKER_NAME]

To delete an experiment tracker, use the delete command:

zenml experiment-tracker delete EXPERIMENT_TRACKER_NAME

Customizing your Step Operator

Step operators allow you to run individual steps in a custom environment different from the default one used by your active orchestrator. One example use-case is to run a training step of your pipeline in an environment with GPUs available. By default, a default ZenML local stack will not register a step operator. If you wish to register a new step operator, do so with the register command:

zenml step-operator register STEP_OPERATOR_NAME --flavor STEP_OPERATOR_FLAVOR [--STEP_OPERATOR_OPTIONS]

If you want the name of the current step operator, use the get command:

zenml step-operator get

To list all step operators available and registered for use, use the list command:

zenml step-operator list

For details about a particular step operator, use the describe command. By default, (without a specific operator name passed in) it will describe the active or currently used step operator:

zenml step-operator describe [STEP_OPERATOR_NAME]

To delete a step operator (and all of its contents), use the delete command:

zenml step-operator delete STEP_OPERATOR_NAME

Setting up a Secrets Manager

ZenML offers a way to securely store secrets associated with your project. To set up a local file-based secrets manager, use the following CLI command:

zenml secrets-manager register SECRETS_MANAGER_NAME --flavor=local

This can then be used as part of your Stack (see below).

Using Secrets

Secrets are administered by the Secrets Manager. You must first register that and then register a stack that includes the secrets manager before you can start to use it. To get a full list of all the possible commands, type zenml secret --help. A ZenML Secret is a collection or grouping of key-value pairs. These Secret groupings come in different types, and certain types have predefined keys that should be used. For example, an AWS secret has predefined keys of aws_access_key_id and aws_secret_access_key (and an optional aws_session_token). If you do not have a specific secret type you wish to use, ZenML will use the arbitrary type to store your key-value pairs.

To register a secret, use the register command and pass the key-value pairs as command line arguments:

zenml secrets-manager secret register SECRET_NAME --key1=value1 --key2=value2 --key3=value3 ...

Note that the keys and values will be preserved in your bash_history file, so you may prefer to use the interactive register command instead:

zenml secrets-manager secret register SECRET_NAME -i

As an alternative to the interactive mode, also useful for values that are long or contain newline or special characters, you can also use the special @ syntax to indicate to ZenML that the value needs to be read from a file:

zenml secrets-manager secret register SECRET_NAME --schema=aws    --aws_access_key_id=1234567890    --aws_secret_access_key=abcdefghij    --aws_session_token=@/path/to/token.txt

To list all the secrets available, use the list command:

zenml secrets-manager secret list

To get the key-value pairs for a particular secret, use the get command:

zenml secrets-manager secret get SECRET_NAME

To update a secret, use the update command:

zenml secrets-manager secret update SECRET_NAME --key1=value1 --key2=value2 --key3=value3 ...

Note that the keys and values will be preserved in your bash_history file, so you may prefer to use the interactive update command instead:

zenml secrets-manager secret update SECRET_NAME -i

Finally, to delete a secret, use the delete command:

zenml secrets-manager secret delete SECRET_NAME

Add a Feature Store to your Stack

ZenML supports connecting to a Redis-backed Feast feature store as a stack component integration. To set up a feature store, use the following CLI command:

zenml feature-store register FEATURE_STORE_NAME --flavor=feast
--feast_repo=REPO_PATH --online_host HOST_NAME --online_port ONLINE_PORT_NUMBER

Once you have registered your feature store as a stack component, you can use it in your ZenML Stack.

Interacting with Deployed Models

If you want to simply see what models have been deployed within your stack, run the following command:

zenml model-deployer models list

This should give you a list of served models containing their uuid, the name of the pipeline that produced them including the run id and the step name as well as the status. This information should help you identify the different models.

If you want further information about a specific model, simply copy the UUID and the following command.

zenml model-deployer models describe <UUID>

If you are only interested in the prediction-url of the specific model you can also run:

zenml model-deployer models get-url <UUID>

Finally, you will also be able to start/stop the services using the following two commands:

zenml model-deployer models start <UUID>
zenml model-deployer models stop <UUID>

If you want to completely remove a served model you can also irreversibly delete it using:

zenml model-deployer models delete <UUID>

Administering the Stack

The stack is a grouping of your artifact store, your metadata store and your orchestrator. With the ZenML tool, switching from a local stack to a distributed cloud environment can be accomplished with just a few CLI commands.

To register a new stack, you must already have registered the individual components of the stack using the commands listed above.

Use the zenml stack register command to register your stack. It takes four arguments as in the following example:

zenml stack register STACK_NAME        -m METADATA_STORE_NAME        -a ARTIFACT_STORE_NAME        -o ORCHESTRATOR_NAME

Each corresponding argument should be the name you passed in as an identifier for the artifact store, metadata store or orchestrator when you originally registered it. (If you want to use your secrets manager, you should pass its name in with the -x option flag.)

Keep in mind that once you register a stack, its artifact store and metadata store will be coupled together, and using these instances in other pairings will not be allowed. This is done to avoid mismatches between these two stores. If you want to override this check when you are declaring/updating a stack, you can use the --decouple_stores flag which will reset any previous couplings and establish a new one with the given artifact- and metadata store.

If you want to immediately set this newly created stack as your active stack, simply pass along the --set flag.

zenml stack register STACK_NAME -m METADATA_STORE_NAME ... --set

To list the stacks that you have registered within your current ZenML project, type:

zenml stack list

To delete a stack that you have previously registered, type:

zenml stack delete STACK_NAME

By default, ZenML uses a local stack whereby all pipelines run on your local computer. If you wish to set a different stack as the current active stack to be used when running your pipeline, type:

zenml stack set STACK_NAME

This changes a configuration property within your local environment.

To see which stack is currently set as the default active stack, type:

zenml stack get

If you want to copy a stack, run the following command:

zenml stack copy SOURCE_STACK_NAME TARGET_STACK_NAME

You can optionally specify profiles from which the stack should be copied to and from:

zenml stack copy SOURCE_STACK_NAME TARGET_STACK_NAME    [--from SOURCE_PROFILE_NAME]    [--to TARGET_PROFILE_NAME]

If you wish to transfer one of your stacks to another machine, you can do so by exporting the stack configuration and then importing it again.

To export a stack to YAML, run the following command:

zenml stack export STACK_NAME FILENAME.yaml

This will create a FILENAME.yaml containing the config of your stack and all of its components, which you can then import again like this:

zenml stack import STACK_NAME FILENAME.yaml

If you wish to update a stack that you have already registered, first make sure you have registered whatever components you want to use, then use the following command:

# assuming that you have already registered a new orchestrator
# with NEW_ORCHESTRATOR_NAME
zenml stack update STACK_NAME -o NEW_ORCHESTRATOR_NAME

You can update one or many stack components at the same time out of the ones that ZenML supports. To see the full list of options for updating a stack, use the following command:

zenml stack update --help

To remove a stack component from a stack, use the following command:

# assuming you want to remove the secrets-manager and the feature-store
# from your stack
zenml stack remove-component -x -f

If you wish to rename your stack, use the following command:

zenml stack rename STACK_NAME NEW_STACK_NAME

If you want to copy a stack component, run the following command:

zenml STACK_COMPONENT copy SOURCE_COMPONENT_NAME TARGET_COMPONENT_NAME

You can optionally specify profiles from which the component should be copied to and from:

zenml STACK_COMPONENT copy SOURCE_COMPONENT_NAME TARGET_COMPONENT_NAME    [--from SOURCE_PROFILE_NAME]    [--to TARGET_PROFILE_NAME]

If you wish to update a specific stack component, use the following command, switching out "STACK_COMPONENT" for the component you wish to update (i.e. 'orchestrator' or 'artifact-store' etc):

zenml STACK_COMPONENT update --some_property=NEW_VALUE

Note that you are not permitted to update the stack name or UUID in this way. To change the name of your stack component, use the following command:

zenml STACK_COMPONENT rename STACK_COMPONENT_NAME NEW_STACK_COMPONENT_NAME

If you wish to remove an attribute (or multiple attributes) from a stack component, use the following command:

zenml STACK_COMPONENT remove-attribute STACK_COMPONENT_NAME --ATTRIBUTE_NAME [--OTHER_ATTRIBUTE_NAME]

Note that you can only remove optional attributes.

If you want to register secrets for all secret references in a stack, use the following command:

zenml stack register-secrets [<STACK_NAME>]

Managing users, teams, projects and roles

When using the ZenML service, you can manage permissions by managing users, teams, projects and roles using the CLI. If you want to create a new user or delete an existing one, run either

zenml user create USER_NAME

or

zenml user delete USER_NAME

To see a list of all users, run:

zenml user list

A team is a grouping of many users that allows you to quickly assign and revoke roles. If you want to create a new team, run:

zenml team create TEAM_NAME

To add one or more users to a team, run:

zenml team add TEAM_NAME --user USER_NAME [--user USER_NAME ...]

Similarly, to remove users from a team run:

zenml team remove TEAM_NAME --user USER_NAME [--user USER_NAME ...]

To delete a team (keep in mind this will revoke any roles assigned to this team from the team members), run:

zenml team delete TEAM_NAME

To see a list of all teams, run:

zenml team list

A role groups permissions and can be assigned to users or teams. To create or delete a role, run one of the following commands:

zenml role create ROLE_NAME
zenml role delete ROLE_NAME

To see a list of all roles, run:

zenml role list

If you want to assign or revoke a role from users or teams, you can run

zenml role assign ROLE_NAME --user USER_NAME [--user USER_NAME ...]
zenml role assign ROLE_NAME --team TEAM_NAME [--team TEAM_NAME ...]

or

zenml role revoke ROLE_NAME --user USER_NAME [--user USER_NAME ...]
zenml role revoke ROLE_NAME --team TEAM_NAME [--team TEAM_NAME ...]

You can see a list of all current role assignments by running:

zenml role assignment list

Interacting with Model Deployers

Model deployers are stack components responsible for online model serving. They are responsible for deploying models to a remote server. Model deployers also act as a registry for models that are served with ZenML.

If you wish to register a new model deployer, do so with the register command:

zenml model-deployer register MODEL_DEPLOYER_NAME --flavor=MODEL_DEPLOYER_FLAVOR [--OPTIONS]

If you wish to list the model-deployers that have already been registered within your ZenML project / repository, type:

zenml model-deployer list

If you wish to get more detailed information about a particular model deployer within your ZenML project / repository, type:

zenml model-deployer describe MODEL_DEPLOYER_NAME

If you wish to delete a particular model deployer, pass the name of the model deployers into the CLI with the following command:

zenml model-deployer delete MODEL_DEPLOYER_NAME

If you wish to retrieve logs corresponding to a particular model deployer, pass the name of the model deployer into the CLI with the following command:

zenml model-deployer logs MODEL_DEPLOYER_NAME

Deploying cloud resources using Stack Recipes

Stack Recipes allow you to quickly deploy fully-fledged MLOps stacks with just a few commands. Each recipe uses Terraform modules under the hood and once executed can set up a ZenML stack, ready to run your pipelines!

A number of stack recipes are already available at the mlops-stacks repository. List them using the following command:

zenml stack recipes list

If you want to pull any specific recipe to your local system, use the pull command:

zenml stack recipe pull <stack-recipe-name>

If you don't specify a name, zenml stack recipe pull will pull all the recipes.

If you notice any inconsistency with the locally-pulled version and the GitHub repository, run the pull command with the -y flag to download any recent changes.

zenml stack recipe pull <stack-recipe-name> -y

Optionally, you can specify the relative path at which you want to install the stack recipe(s). Use the -p or --path flag.

zenml stack recipe pull <stack-recipe-name> --path=<PATH>

By default, all recipes get downloaded under a directory called zenml_stack_recipes.

To deploy a recipe, use the deploy command. Before running deploy, review the zenml_stack_recipes/<stack-recipe-name>/locals.tf file for configuring non-sensitive variables and the zenml_stack_recipes/<stack-recipe-name>/values.tfvars file to add sensitive information like access keys and passwords.

zenml stack recipe deploy <stack-recipe-name>

Running deploy without any options will create a new ZenML stack with the same name as the stack recipe name. Use the --stack-name option to specify your own name.

zenml stack recipe deploy <stack-recipe-name> --stack-name=my_stack

If you wish to review the stack information from the newly-generated resources before importing, you can run deploy with the --no-import flag.

zenml stack recipe deploy <stack-recipe-name> --no-import

This will still create a stack YAML configuration file but will not auto-import it. You can make any changes you want to the configuration and then run zenml stack import manually.

To remove all resources created as part of the recipe, run the destroy command.

zenml stack recipe destroy <stack-recipe-name>

To delete all the recipe files from your system, you can use the clean command.

zenml stack recipe clean

This deletes all the recipes from the default path where they were downloaded.