DeployApi¶

class DeployApi(api)[source]¶

Bases: object

API for deploying models and controlling serving apps.

This class is used internally by NeuralNetworkApi, but it can also be used directly for advanced deployment workflows.

Key capabilities:

deploy a pretrained model into a new serving task,
deploy a custom checkpoint (from team files or experiment artifacts),
load/replace model inside an existing serving task.

Parameters:

api¶: Api object to use for API connection.

Methods

`deploy_custom_model_by_artifacts_dir`	Deploy a custom model based on the artifacts directory.
`deploy_custom_model_by_checkpoint`	Deploy a custom model based on the checkpoint path.
`deploy_custom_model_from_experiment_info`	Deploy a custom model based on the training session.
`deploy_pretrained_model`	Deploy a pretrained model.
`find_serving_app_by_framework`	Find the serving app by framework.
`find_serving_app_by_slug`	Find the serving app by slug.
`get_deploy_info`	Get deploy info of a serving task.
`get_serving_app_by_train_app`	Get the serving app by train app.
`load_custom_model`	Load a custom model in running serving App.
`load_custom_model_from_experiment_info`	Load a custom model in running serving App based on the training session.
`load_pretrained_model`	Load a pretrained model in running serving App.
`start_serve_app`	Run a serving app.
`wait`	Wait for the model to be started or deployed.

deploy_custom_model_by_artifacts_dir(artifacts_dir, checkpoint_name=None, device=None, runtime=None, timeout=100, team_id=None, workspace_id=None, agent_id=None, **kwargs)[source]¶

Deploy a custom model based on the artifacts directory.

Parameters:

artifacts_dir : str¶: Path to the artifacts directory in the team fies.
checkpoint_name : Optional[str]¶: Checkpoint name (with file extension) to deploy, e.g. “best.pt”. If not provided, checkpoint will be chosen automatically, depending on the app version.
device : Optional[str]¶: Device string. If not provided, will be chosen automatically.
runtime : Optional[str]¶: Runtime string, if not present will be defined automatically.
timeout : Optional[int]¶: Timeout in seconds (default is 100). The maximum time to wait for the serving app to be ready.
team_id : Optional[int]¶: Team ID where the artifacts are stored. If not provided, will be taken from the current context.
workspace_id : Optional[int]¶: Workspace ID where the app will be deployed. If not provided, will be taken from the current context.
agent_id : Optional[int]¶: Agent ID. If not provided, will be found automatically.
kwargs : Dict[str, Any]¶: Additional parameters to start the task. See start() for more details.

Returns:

Task Info

Return type:

Dict[str, Any]

Raises:

ValueError – if validations fail.

deploy_custom_model_by_checkpoint(checkpoint, device=None, runtime=None, timeout=100, team_id=None, workspace_id=None, agent_id=None, **kwargs)[source]¶

Deploy a custom model based on the checkpoint path.

Parameters:

checkpoint : str¶: Path to the checkpoint in Team Files.
device : Optional[str]¶: Device string. If not provided, will be chosen automatically.
runtime : Optional[str]¶: Runtime string, if not present will be defined automatically.
timeout : Optional[int]¶: Timeout in seconds (default is 100). The maximum time to wait for the serving app to be ready.
team_id : Optional[int]¶: Team ID where the artifacts are stored. If not provided, will be taken from the current context.
workspace_id : Optional[int]¶: Workspace ID where the app will be deployed. If not provided, will be taken from the current context.
agent_id : Optional[int]¶: Agent ID. If not provided, will be found automatically.
kwargs : Dict[str, Any]¶: Additional parameters to start the task. See start() for more details.

Returns:

Task Info

Return type:

Dict[str, Any]

Raises:

ValueError – if validations fail.

deploy_custom_model_from_experiment_info(agent_id, experiment_info, checkpoint_name=None, device=None, runtime=None, timeout=100, **kwargs)[source]¶

Deploy a custom model based on the training session.

Parameters:

experiment_info¶: Experiment info from TrainApp.
checkpoint_name : Optional[str]¶: Checkpoint name (with file extension) to deploy, e.g. “best.pt”. If not provided, the best checkpoint will be chosen.
device : Optional[str]¶: Device string. If not provided, will be chosen automatically.
timeout : Optional[int]¶: Timeout in seconds (default is 100). The maximum time to wait for the serving app to be ready.
kwargs : Dict[str, Any]¶: Additional parameters to start the task. See start() for more details.

Returns:

Task Info

Return type:

Dict[str, Any]

Raises:

ValueError – if validations fail.

deploy_pretrained_model(framework, model_name, device=None, runtime=None, workspace_id=None, agent_id=None, app=None, **kwargs)[source]¶

Deploy a pretrained model.

Parameters:

framework : Union[str, int]¶: Framework name or Framework ID in Supervisely.
model_name : str¶: Model name to deploy.
device : Optional[str]¶: Device string. If not provided, will be chosen automatically.
runtime : Optional[str]¶: Runtime string, if not present will be defined automatically.
workspace_id : Optional[int]¶: Workspace ID where the app will be deployed. If not provided, will be taken from the current context.
agent_id : Optional[int]¶: Agent ID. If not provided, will be found automatically.
app : Union[str, int]¶: App name or App module ID in Supervisely.
kwargs : Dict[str, Any]¶: Additional parameters to start the task. See start() for more details.

Returns:

Task Info

Return type:

Dict[str, Any]

Raises:

ValueError – if no serving apps found for the app name or multiple serving apps found for the app name.

find_serving_app_by_framework(framework)[source]¶

Find the serving app by framework.

Parameters:

framework : str¶: Framework name.

Returns:

Serving app info

Return type:

Dict[str, Any]

find_serving_app_by_slug(slug)[source]¶

Find the serving app by slug.

Parameters:

slug : str¶: Slug of the serving app.

Returns:

Serving app ID

Return type:

int

get_deploy_info(task_id)[source]¶

Get deploy info of a serving task.

Parameters:

task_id : int¶: Task ID of the serving App.

Returns:

Deploy Info

Return type:

Dict[str, Any]

get_serving_app_by_train_app(app_name=None, module_id=None)[source]¶

Get the serving app by train app.

Parameters:

app_name : Optional[str]¶: App name in Supervisely.
module_id : int¶: Module ID in Supervisely.

Returns:

Serving app info

Return type:

Dict[str, Any]

load_custom_model(session_id, team_id, artifacts_dir, checkpoint_name=None, device=None, runtime=None)[source]¶

Load a custom model in running serving App.

Parameters:

session_id : int¶: Task ID of the serving App.
team_id : int¶: Team ID in Supervisely.
artifacts_dir : str¶: Path to the artifacts directory in the team fies.
checkpoint_name : Optional[str]¶: Checkpoint name (with file extension) to deploy, e.g. “best.pt”. If not provided, checkpoint will be chosen automatically, depending on the app version.
device : Optional[str]¶: Device string. If not provided, will be chosen automatically.
runtime : Optional[str]¶: Runtime string, if not present will be defined automatically.

load_custom_model_from_experiment_info(session_id, experiment_info, checkpoint_name=None, device=None, runtime=None)[source]¶

Load a custom model in running serving App based on the training session.

Parameters:

session_id : int¶: Task ID of the serving App.
experiment_info¶: Experiment info from TrainApp.
checkpoint_name : Optional[str]¶: Checkpoint name (with file extension) to deploy, e.g. “best.pt”. If not provided, checkpoint will be chosen automatically, depending on the app version.
device : Optional[str]¶: Device string. If not provided, will be chosen automatically.
runtime : Optional[str]¶: Runtime string, if not present will be defined automatically.

load_pretrained_model(session_id, model_name, device=None, runtime=None)[source]¶

Load a pretrained model in running serving App.

Parameters:

session_id : int¶: Task ID of the serving App.
model_name : str¶: Model name to deploy.
device : Optional[str]¶: Device string. If not provided, will be chosen automatically.
runtime : Optional[str]¶: Runtime string, if not present will be defined automatically.

start_serve_app(agent_id, app_name=None, module_id=None, **kwargs)[source]¶

Run a serving app. Either app_name or module_id must be provided.

Parameters:

app_name : Optional[str]¶: App name in Supervisely.
module_id : Optional[int]¶: Module ID in Supervisely.
kwargs : Dict[str, Any]¶: Additional parameters to start the task. See start() for more details.

Returns:

Task Info

Return type:

Dict[str, Any]

wait(model_id, target='started', timeout=300)[source]¶

Wait for the model to be started or deployed.

Parameters:

model_id : int¶: Model ID.
target : Literal["started", "deployed"]¶: Target status.
timeout : int¶: Timeout in seconds.