Flyte CLI

This is the command line interface for Flyte.

flyte

flyte [OPTIONS] COMMAND [ARGS]...

The Flyte CLI is the command line interface for working with the Flyte SDK and backend.

It follows a simple verb/noun structure, where the top-level commands are verbs that describe the action to be taken, and the subcommands are nouns that describe the object of the action.

The root command can be used to configure the CLI for persistent settings, such as the endpoint, organization, and verbosity level.

Set endpoint and organization:

$ flyte --endpoint <endpoint> --org <org> get project <project_name>

Increase verbosity level (This is useful for debugging, this will show more logs and exception traces):

$ flyte -vvv get logs <run-name>

Override the default config file:

$ flyte --config /path/to/config.yaml run ...

• Documentation
• GitHub: Please leave a star if you like Flyte!
• Slack: Join the community and ask questions.
• Issues

flyte abort

flyte abort COMMAND [ARGS]...

Abort an ongoing process.

flyte abort action

flyte abort action [OPTIONS] RUN_NAME ACTION_NAME

Abort an action associated with a run.

flyte abort run

flyte abort run [OPTIONS] RUN_NAME

Abort a run.

flyte build

flyte build [OPTIONS] COMMAND [ARGS]...

Build the environments defined in a python file or directory. This will build the images associated with the environments.

flyte create

flyte create COMMAND [ARGS]...

Create resources in a Flyte deployment.

# Create an API key named "ci-pipeline"
$ flyte create api-key --name ci-pipeline

# The output will include an export command like:
# export FLYTE_API_KEY="<base64-encoded-credentials>"

$ flyte --org my-org create assignment --user-subject user-123 --policy admin
$ flyte --org my-org create assignment --creds-subject app-456 --policy admin
$ flyte --org my-org create assignment --email [email protected] --policy admin

flyte create config

flyte create config [OPTIONS]

Creates a configuration file for Flyte CLI. If the --output option is not specified, it will create a file named config.yaml in the current directory. If the file already exists, it will raise an error unless the --force option is used.

$ flyte --org my-org create policy my-policy --edit
$ flyte --org my-org create policy my-policy --file policy.yaml

flyte create project

flyte create project [OPTIONS]

Create a new project.

 Example usage:

flyte create project --id my_project_id --name "My Project"
flyte create project --id my_project_id --name "My Project" --description "My project" -l team=ml -l env=prod

$ flyte --org my-org create role my-role --edit
$ flyte --org my-org create role my-role --file role.yaml

flyte create secret

flyte create secret [OPTIONS] NAME

Create a new secret. The name of the secret is required. For example:

$ flyte create secret my_secret --value my_value

If you don’t provide a --value flag, you will be prompted to enter the secret value in the terminal.

$ flyte create secret my_secret
Enter secret value:

If --from-file is specified, the value will be read from the file instead of being provided directly:

$ flyte create secret my_secret --from-file /path/to/secret_file

The --type option can be used to create specific types of secrets. Either regular or image_pull can be specified. Secrets intended to access container images should be specified as image_pull. Other secrets should be specified as regular. If no type is specified, regular is assumed.

For image pull secrets, you have several options:

1. Interactive mode (prompts for registry, username, password):

$ flyte create secret my_secret --type image_pull

2. With explicit credentials:

$ flyte create secret my_secret --type image_pull --registry ghcr.io --username myuser

3. Lastly, you can create a secret from your existing Docker installation (i.e., you’ve run docker login in the past) and you just want to pull from those credentials. Since you may have logged in to multiple registries, you can specify which registries to include. If no registries are specified, all registries are added.

$ flyte create secret my_secret --type image_pull --from-docker-config --registries ghcr.io,docker.io

flyte create trigger

flyte create trigger [OPTIONS] TASK_NAME NAME

Create a new trigger for a task. The task name and trigger name are required.

Example:

$ flyte create trigger my_task my_trigger --schedule "0 0 * * *"

This will create a trigger that runs every day at midnight.

$ flyte --org my-org create user --first-name Jane --last-name Doe --email [email protected]
$ flyte --org my-org create user --first-name Jane --last-name Doe --email [email protected] --policy admin

flyte delete

flyte delete COMMAND [ARGS]...

Remove resources from a Flyte deployment.

# Delete an API key (with confirmation)
$ flyte delete api-key my-client-id

# Delete without confirmation
$ flyte delete api-key my-client-id --yes

flyte delete app

flyte delete app [OPTIONS] NAME

Delete apps from a Flyte deployment.

$ flyte --org my-org delete assignment --user-subject user-123 --policy admin
$ flyte --org my-org delete assignment --creds-subject app-456 --policy admin

$ flyte --org my-org delete policy my-policy
$ flyte --org my-org delete policy my-policy --yes

$ flyte --org my-org delete role my-role
$ flyte --org my-org delete role my-role --yes

flyte delete secret

flyte delete secret [OPTIONS] NAME

Delete a secret. The name of the secret is required.

flyte delete trigger

flyte delete trigger [OPTIONS] NAME TASK_NAME

Delete a trigger. The name of the trigger is required.

$ flyte --org my-org delete user user-subject-id
$ flyte --org my-org delete user user-subject-id --yes

flyte deploy

flyte deploy [OPTIONS] COMMAND [ARGS]...

Deploy one or more environments from a python file.

This command will create or update environments in the Flyte system, registering all tasks and their dependencies.

Example usage:

flyte deploy hello.py my_env

Arguments to the deploy command are provided right after the deploy command and before the file name.

To deploy all environments in a file, use the --all flag:

flyte deploy --all hello.py

To recursively deploy all environments in a directory and its subdirectories, use the --recursive flag:

flyte deploy --recursive ./src

You can combine --all and --recursive to deploy everything:

flyte deploy --all --recursive ./src

You can provide image mappings with --image flag. This allows you to specify the image URI for the task environment during CLI execution without changing the code. Any images defined with Image.from_ref_name("name") will resolve to the corresponding URIs you specify here.

flyte deploy --image my_image=ghcr.io/myorg/my-image:v1.0 hello.py my_env

If the image name is not provided, it is regarded as a default image and will be used when no image is specified in TaskEnvironment:

flyte deploy --image ghcr.io/myorg/default-image:latest hello.py my_env

You can specify multiple image arguments:

flyte deploy --image ghcr.io/org/default:latest --image gpu=ghcr.io/org/gpu:v2.0 hello.py my_env

To deploy a specific version, use the --version flag:

flyte deploy --version v1.0.0 hello.py my_env

To preview what would be deployed without actually deploying, use the --dry-run flag:

flyte deploy --dry-run hello.py my_env

You can specify the --config flag to point to a specific Flyte cluster:

flyte --config my-config.yaml deploy hello.py my_env

You can override the default configured project and domain:

flyte deploy --project my-project --domain development hello.py my_env

If loading some files fails during recursive deployment, you can use the --ignore-load-errors flag to continue deploying the environments that loaded successfully:

flyte deploy --recursive --ignore-load-errors ./src

Other arguments to the deploy command are listed below.

To see the environments available in a file, use --help after the file name:

flyte deploy hello.py --help

flyte gen

flyte gen COMMAND [ARGS]...

Generate documentation.

flyte gen docs

flyte gen docs [OPTIONS]

Generate documentation.

flyte get

flyte get COMMAND [ARGS]...

Retrieve resources from a Flyte deployment.

You can get information about projects, runs, tasks, actions, secrets, logs and input/output values.

Each command supports optional parameters to filter or specify the resource you want to retrieve.

Using a get subcommand without any arguments will retrieve a list of available resources to get. For example:

• get project (without specifying a project), will list all projects.
• get project my_project will return the details of the project named my_project.

In some cases, a partially specified command will act as a filter and return available further parameters. For example:

• get action my_run will return all actions for the run named my_run.
• get action my_run my_action will return the details of the action named my_action for the run my_run.

flyte get action

flyte get action [OPTIONS] RUN_NAME [ACTION_NAME]

Get all actions for a run or details for a specific action.

# List all API keys
$ flyte get api-key

# List with a limit
$ flyte get api-key --limit 10

# Get a specific API key
$ flyte get api-key my-client-id

flyte get app

flyte get app [OPTIONS] [NAME]

Get a list of all apps, or details of a specific app by name.

Apps are long-running services deployed on the Flyte platform.

$ flyte --org my-org get assignment
$ flyte --org my-org get assignment --user-subject user-123
$ flyte --org my-org get assignment --creds-subject app-456

flyte get config

flyte get config

Shows the automatically detected configuration to connect with the remote backend.

The configuration will include the endpoint, organization, and other settings that are used by the CLI.

flyte get io

flyte get io [OPTIONS] RUN_NAME [ACTION_NAME]

Get the inputs and outputs of a run or action. If only the run name is provided, it will show the inputs and outputs of the root action of that run. If an action name is provided, it will show the inputs and outputs for that action. If --inputs-only or --outputs-only is specified, it will only show the inputs or outputs respectively.

Examples:

$ flyte get io my_run

$ flyte get io my_run my_action

flyte get logs

flyte get logs [OPTIONS] RUN_NAME [ACTION_NAME]

Stream logs for the provided run or action. If only the run is provided, only the logs for the parent action will be streamed:

$ flyte get logs my_run

If you want to see the logs for a specific action, you can provide the action name as well:

$ flyte get logs my_run my_action

By default, logs will be shown in the raw format and will scroll the terminal. If automatic scrolling and only tailing --lines number of lines is desired, use the --pretty flag:

$ flyte get logs my_run my_action --pretty --lines 50

$ flyte --org my-org get member

$ flyte --org my-org get policy
$ flyte --org my-org get policy --limit 10
$ flyte --org my-org get policy my-policy

flyte get project

flyte get project [OPTIONS] [NAME]

Get a list of all projects, or details of a specific project by name.

By default, only active (unarchived) projects are shown. Use --archived to show archived projects instead.

$ flyte --org my-org get role
$ flyte --org my-org get role --limit 10
$ flyte --org my-org get role my-role

flyte get run

flyte get run [OPTIONS] [NAME]

Get a list of all runs, or details of a specific run by name.

The run details will include information about the run, its status, but only the root action will be shown.

If you want to see the actions for a run, use get action <run_name>.

You can filter runs by task name and optionally task version:

$ flyte get run --task-name my_task
$ flyte get run --task-name my_task --task-version v1.0

flyte get secret

flyte get secret [OPTIONS] [NAME]

Get a list of all secrets, or details of a specific secret by name.

flyte get task

flyte get task [OPTIONS] [NAME] [VERSION]

Retrieve a list of all tasks, or details of a specific task by name and version.

Currently, both name and version are required to get a specific task.

flyte get trigger

flyte get trigger [OPTIONS] [TASK_NAME] [NAME]

Get a list of all triggers, or details of a specific trigger by name.

$ flyte --org my-org get user
$ flyte --org my-org get user --limit 10
$ flyte --org my-org get user user-subject-id
$ flyte --org my-org get user --email [email protected]

flyte prefetch

flyte prefetch COMMAND [ARGS]...

Prefetch artifacts from remote registries.

These commands help you download and prefetch artifacts like HuggingFace models to your Flyte storage for faster access during task execution.

flyte prefetch hf-model

flyte prefetch hf-model [OPTIONS] REPO

Prefetch a HuggingFace model to Flyte storage.

Downloads a model from the HuggingFace Hub and prefetches it to your configured Flyte storage backend. This is useful for:

• Pre-fetching large models before running inference tasks
• Sharding models for tensor-parallel inference
• Avoiding repeated downloads during development

Basic Usage:

$ flyte prefetch hf-model meta-llama/Llama-2-7b-hf --hf-token-key HF_TOKEN

With Sharding:

Create a shard config file (shard_config.yaml):

engine: vllm
args:
  tensor_parallel_size: 8
  dtype: auto
  trust_remote_code: true

Then run:

$ flyte prefetch hf-model meta-llama/Llama-2-70b-hf \
    --shard-config shard_config.yaml \
    --accelerator A100:8 \
    --hf-token-key HF_TOKEN

Wait for Completion:

$ flyte prefetch hf-model meta-llama/Llama-2-7b-hf --wait