Skip to main content


The following page provides an overview of the basic Pythonic interface to ClearML Tasks.

Task Creation#

Task.init is the main method used to create tasks in ClearML. It will create a task, and populate it with:

  • A link to the running git repository (including commit ID and local uncommitted changes)
  • Python packages used (i.e. directly imported Python packages, and the versions available on the machine)
  • Argparse arguments (default and specific to the current execution)
  • Reports to Tensorboard & Matplotlib and model checkpoints.

ClearML object (e.g. task, project) names are required to be at least 3 characters long

from clearml import Task
task = Task.init(
project_name='example', # project name of at least 3 characters
task_name='task template', # task name of at least 3 characters

Once a task is created, the task object can be accessed from anywhere in the code by calling Task.current_task.

If multiple tasks need to be created in the same process (for example, for logging multiple manual runs), make sure to close a task, before initializing a new one. To close a task simply call Task.close (see example here).

When initializing a task, its project needs to be specified. If the project entered does not exist, it will be created on-the-fly. Projects can be divided into sub-projects, just like folders are broken into sub-folders.

For example:

Task.init(project_name='main_project/sub_project', task_name='test')

Nesting projects works on multiple levels. For example: project_name=main_project/sub_project/sub_sub_project

Automatic Logging#

After invoking Task.init in a script, ClearML starts its automagical logging, which includes the following elements:

  • Hyperparameters - ClearML logs the following types of hyperparameters:
    • Command Line Parsing - ClearML captures any command line parameters passed when invoking code that uses standard python packages, including:
    • TensorFlow Definitions (absl-py)
    • Hydra - the Omegaconf which holds all the configuration files, as well as overridden values.
  • Models - ClearML automatically logs and updates the models and all snapshot paths saved with the following frameworks:
  • Metrics, scalars, plots, debug images reported through supported frameworks, including:
  • Execution details including:
    • Git information
    • Uncommitted code modifications - In cases where no git repository is detected (e.g. when a single python script is executed outside a git repository, or when running from a Jupyter Notebook), ClearML logs the contents of the executed script
    • Python environment
    • Execution configuration

To control a task's framework logging, use the auto_connect_framworks parameter of the Task.init method. Turn off all automatic logging by setting the parameter to False. For finer grained control of logged frameworks, input a dictionary, with framework-boolean pairs.

For example:

'matplotlib': True, 'tensorflow': False, 'tensorboard': False, 'pytorch': True,
'xgboost': False, 'scikit': True, 'fastai': True, 'lightgbm': False,
'hydra': True, 'detect_repository': True, 'tfdefines': True, 'joblib': True,
'megengine': True, 'jsonargparse': True, 'catboost': True

Task Reuse#

Every Task.init call will create a new task for the current execution. In order to mitigate the clutter that a multitude of debugging tasks might create, a task will be reused if:

  • The last time it was executed (on this machine) was under 72 hours ago (configurable, see sdk.development.task_reuse_time_window_in_hours in the sdk.development section of the ClearML configuration reference)
  • The previous task execution did not have any artifacts / models

It's possible to always create a new task by passing reuse_last_task_id=False.

See full Task.init reference here.

Empty Task Creation#

A task can also be created without the need to execute the code itself. Unlike the runtime detections, all the environment and configuration details need to be provided explicitly.

For example:

task = Task.create(
task_name='task template',

See full Task.create reference here.

Tracking Task Progress#

Track a task’s progress by setting the task progress property using the Task.set_progress method. Set a task’s progress to a numeric value between 0 - 100. Access the task’s current progress, using the Task.get_progress method.

task = Task.init(project_name="examples", task_name="Track experiment progress")
# task doing stuff
# task doing more stuff

While the task is running, the WebApp will show the task’s progress indication in the experiment table, next to the task’s status. If a task failed or was aborted, you can view how much progress it had made.

Experiment table progress indication

Additionally, you can view a task’s progress in its INFO tab in the WebApp.

Accessing Tasks#

A task can be identified by its project and name, and by a unique identifier (UUID string). The name and project of a task can be changed after an experiment has been executed, but its ID can't be changed.

Programmatically, task objects can be retrieved by querying the system based on either the task ID or a project and name combination using the Task.get_task class method. If a project / name combination is used, and multiple tasks have the exact same name, the function will return the last modified task.

For example:

  • Accessing a task object with a task ID:

    a_task = Task.get_task(task_id='123456deadbeef')
  • Accessing a task with a project and name:

    a_task = Task.get_task(project_name='examples', task_name='artifacts')

Once a task object is obtained, it's possible to query the state of the task, reported scalars, etc. The task's outputs, such as artifacts and models, can also be retrieved.

Querying / Searching Tasks#

Searching and filtering tasks can be done via the web UI and programmatically. Input search parameters into the Task.get_tasks method, which returns a list of task objects that match the search.

For example:

task_list = Task.get_tasks(
task_ids=None, # type Optional[Sequence[str]]
project_name=None, # Optional[str]
task_name=None, # Optional[str]
task_filter=None # Optional[Dict]

It's possible to also filter tasks by passing filtering rules to task_filter.

For example:

# only tasks with tag `included_tag` and without tag `excluded_tag`
'tags': ['included_tag', '-excluded_tag'],
# filter out archived tasks
'system_tags': ['-archived'],
# only completed & published tasks
'status': ['completed', 'published'],
# only training type tasks
'type': ['training'],
# match text in task comment or task name
'search_text': 'reg_exp_text'

Cloning & Executing Tasks#

Once a task object is created, it can be copied (cloned). Task.clone returns a copy of the original task (source_task). By default, the cloned task is added to the same project as the original, and it's called "Clone Of ORIGINAL_NAME", but the name / project / comment (description) of the cloned task can be directly overridden.

task = Task.init(project_name='examples', task_name='original task',)
cloned = Task.clone(
source_task=task, # type: Optional[Union[Task, str]]
# override default name
name='newly created task', # type: Optional[str]
comment=None, # type: Optional[str]
# insert cloned task into a different project
project=None, # type: Optional[str]

A newly cloned task has a draft status, so it's modifiable.

Once a task is modified, launch it by pushing it into an execution queue with the Task.enqueue class method. Then a ClearML Agent assigned to the queue will pull the task from the queue and execute it.

task=task, # type: Union[Task, str]
queue_name='default', # type: Optional[str]
queue_id=None # type: Optional[str]

See enqueue example.

Advanced Flows#

Remote Execution#

A compelling workflow is:

  1. Run code on a development machine for a few iterations, or just set up the environment.
  2. Move the execution to a beefier remote machine for the actual training.

Use the Task.execute_remotely method to implement this workflow. This method stops the current manual execution, and then re-runs it on a remote machine.

For example:

queue_name='default', # type: Optional[str]
clone=False, # type: bool
exit_process=True # type: bool

Once the method is called on the machine, it stops the local process and enqueues the current task into the default queue. From there, an agent can pull and launch it.

See the Remote Execution example.

Remote Function Execution#

A specific function can also be launched on a remote machine with the Task.create_function_task method.

For example:

def run_me_remotely(some_argument):
a_func_task = task.create_function_task(
func=run_me_remotely, # type: Callable
func_name='func_id_run_me_remotely', # type:Optional[str]
task_name='a func task', # type:Optional[str]
# everything below will be passed directly to our function as arguments

Arguments passed to the function will be automatically logged under the Function section in the Hyperparameters tab. Like any other arguments, they can be changed from the UI or programmatically.

Function Task Creation

Function tasks must be created from within a regular task, created by calling Task.init

Offline Mode#

You can work with tasks in Offline Mode, in which all the data and logs that the Task captures are stored in a local folder, which can later be uploaded to the ClearML Server.

Before initializing a Task, use the Task.set_offline class method and set the offline_mode argument to True. The method returns the Task ID and a path to the session folder.

from clearml import Task
# Use the set_offline class method before initializing a Task
# Initialize a Task
task = Task.init(project_name="examples", task_name="my_task")
# Rest of code is executed. All data is logged locally and not onto the server

All the information captured by the Task is saved locally. Once the task script finishes execution, it's zipped.

Upload the execution data that the Task captured offline to the ClearML Server using one of the following:

  • clearml-task CLI

    clearml-task --import-offline-session "path/to/session/.clearml/cache/offline/"

    Pass the path to the zip folder containing the session with the --import-offline-session parameter

  • Task.import_offline_session class method

    from clearml import Task

    In the session_folder_zip argument, insert the path to the zip folder containing the session.

Both options will upload the Task's full execution details and outputs and return a link to the Task's results page on the ClearML Server.


Artifacts are the output files created by a task. ClearML uploads and logs these products so they can later be easily accessed, modified, and used.

Logging Artifacts#

To log an artifact in a task, use the upload_artifact method.

For example:

  • Upload a local file containing the preprocessing results of the data:

    task.upload_artifact(name='data', artifact_object='/path/to/preprocess_data.csv')
  • Upload an entire folder with all its content by passing the folder, which will be zipped and uploaded as a single zip file:

    task.upload_artifact(name='folder', artifact_object='/path/to/folder')
  • Serialize and upload a Python object. ClearML automatically chooses the file format based on the object’s type, or you can explicitly specify the format as follows:

    • dict - .json (default), .yaml
    • pandas.DataFrame - .csv.gz (default), .parquet, .feather, .pickle
    • numpy.ndarray - .npz (default), .csv.gz
    • PIL.Image - Any PIL-supported extensions (default .png)

    For example:

    person_dict = {'name': 'Erik', 'age': 30}
    # upload as JSON artifact
    task.upload_artifact(name='person dictionary json', artifact_object=person_dict)
    # upload as YAML artifact
    name='person dictionary yaml',

See more details in the Artifacts Reporting example and in the SDK reference.

Using Artifacts#

A task's artifacts are accessed through the task’s artifact property which lists the artifacts’ locations.

The artifacts can subsequently be retrieved from their respective locations by using:

  • get_local_copy()- Downloads the artifact and caches it for later use, returning the path to the cached copy.
  • get() - Returns a Python object constructed from the downloaded artifact file.

The code below demonstrates how to access a file artifact using the previously generated preprocessed data:

# get instance of task that created artifact, using task ID
preprocess_task = Task.get_task(task_id='the_preprocessing_task_id')
# access artifact
local_csv = preprocess_task.artifacts['data'].get_local_copy()

See more details in the Using Artifacts example.


The following is an overview of working with models through a Task object. It is also possible to work directly with model objects (see Models (SDK)).

Logging Models Manually#

To manually log a model in a task, create an instance of the OutputModel class. An OutputModel object is always registered as an output model of the task it is constructed from.

For example:

from clearml import OutputModel, Task
# Instantiate a Task
task = Task.init(project_name="myProject", task_name="myTask")
# Instantiate an OutputModel with a task object argument
output_model = OutputModel(task=task, framework="PyTorch")

Updating Models Manually#

The snapshots of manually uploaded models aren't automatically captured. To update a task's model, use the Task.update_output_model method:


It's possible to modify the following parameters:

  • Model location
  • Model name
  • Model description
  • Iteration number
  • Model tags

Models can also be manually updated independently, without any task. See OutputModel.update_weights.

Using Models#

Accessing a task’s previously trained model is quite similar to accessing task artifacts. A task's models are accessed through the task’s models property which lists the input models and output model snapshots’ locations.

The models can subsequently be retrieved from their respective locations by using get_local_copy() which downloads the model and caches it for later use, returning the path to the cached copy (if using Tensorflow, the snapshots are stored in a folder, so the local_weights_path will point to a folder containing the requested snapshot).

prev_task = Task.get_task(task_id='the_training_task')
last_snapshot = prev_task.models['output'][-1]
local_weights_path = last_snapshot.get_local_copy()

Notice that if one of the frameworks loads an existing weights file, the running task will automatically update its "Input Model", pointing directly to the original training task's model. This makes it easy to get the full lineage of every trained and used model in our system!

Models loaded by the ML framework appear under the "Input Models" section, under the Artifacts tab in the ClearML UI.

Setting Upload Destination#

ClearML automatically captures the storage location of Models created by frameworks such as TF, Pytorch, and scikit-learn. By default, it stores the local path they are saved at.

To automatically store all created models by a specific experiment, modify the Task.init function as such:

task = Task.init(
task_name='storing model',

To automatically store all models created by any experiment at a specific location, edit the clearml.conf (see ClearML Configuration Reference) and set sdk.developmenmt.default_output_uri to the desired storage (see Storage). This is especially helpful when using clearml-agent to execute code.


Manual Hyperparameter Logging#

Setting Parameters#

To define parameters manually use the Task.set_parameters method to specify name-value pairs in a parameter dictionary.

Parameters can be designated into sections: specify a parameter’s section by prefixing its name, delimited with a slash (i.e. section_name/parameter_name’:value). General is the default section.

Call the set_parameter method to set a single parameter.

task = Task.init(project_name='examples', task_name='parameters')
# override parameters with provided dictionary
task.set_parameters({'Args/epochs':7, 'lr': 0.5})
# setting a single parameter
Overwriting Parameters

The set_parameters method replaces any existing hyperparameters in the task.

Adding Parameters#

To update the parameters in a task, use the Task.set_parameters_as_dict method. Arguments and values are input as a dictionary. Like in set_parameters above, the parameter's section can be specified.

task = Task.task_get(task_id='123456789')
# add parameters
task.set_parameters_as_dict({'my_args/lr':0.3, 'epochs':10})

Accessing Parameters#

To access all task parameters, use the Task.get_parameters method. This method returns a flattened dictionary of the 'section/parameter': 'value' pairs.

task = Task.get_task(project_name='examples', task_name='parameters')
# will print a flattened dictionary of the 'section/parameter': 'value' pairs

Access a specific parameter with the Task.get_parameter method specifying the parameter name.

Tracking Python Objects#

ClearML can track Python objects (such as dictionaries and custom classes) as they evolve in your code, and log them to your task’s configuration using the Task.connect method. Once objects are connected to a task, ClearML automatically logs all object elements (e.g. class members, dictionary key-values pairs).

class person:
def __init__(self, name, age): = name
self.age = age
me = person('Erik', 5)
params_dictionary = {'epochs': 3, 'lr': 0.4}
task = Task.init(project_name='examples',task_name='argparser')

Task parameters

Configuration Objects#

Configuration objects more elaborate than a key-value dictionary (such as nested dictionaries or configuration files), can be logged to a task using the Task.connect_configuration method. This method saves configuration objects as blobs (i.e. ClearML is not aware of their internal structure).

# connect a configuration dictionary
model_config_dict = {
'value': 13.37, 'dict': {'sub_value': 'string'}, 'list_of_ints': [1, 2, 3, 4],
model_config_dict = task.connect_configuration(
name='dictionary', configuration=model_config_dict
# connect a configuration file
config_file_yaml = task.connect_configuration(
name="yaml file", configuration='path/to/configuration/file.yaml'

Task configuration objects

User Properties#

A task’s user properties do not impact task execution so you can add / modify the properties at any stage. Add user properties to a task with the Task.set_user_properties method.

{"name": "backbone", "description": "network type", "value": "great"}

The above example sets the "backbone" property in a task.

Task user properties

SDK Reference#

For detailed information, see the complete Task SDK reference page