Creating a Training Job

project_id

Yes

String

Definition: Project ID. For details, see Obtaining a Project ID and Name.

Constraints: N/A

Range: The value can contain 1 to 64 characters. Only letters, digits, and hyphens (-) are allowed.

Default Value: N/A

kind

Yes

String

Definition: Type of a training job.

Constraints: N/A

Range

• job: common job

• federated_pool_job: resource pool federated job

• edge_job: edge job

• hetero_job: heterogeneous job

• mrs_job: MRS job

• autosearch_job: auto search job

• diag_job: diagnosis job

• visualization_job: visualization job

Default Value: job

metadata

Yes

JobMetadata object

Definition: Training job metadata.

Constraints: N/A

algorithm

No

JobAlgorithm object

Definition: Training job algorithm.

Constraints: This parameter is mandatory when tasks is left blank. If this parameter is specified, you do not need to specify tasks. The options are as follows:

• id: Only the algorithm ID is used.

• subscription_id+item_version_id: The subscription ID and version ID of the algorithm are used.

• code_dir+boot_file: The code directory and boot file of the training job are used.

tasks

No

Array of Task objects

Definition: Task list.

Constraints: This parameter is mandatory when algorithm is left blank. If this parameter is specified, you do not need to specify algorithm. This function is unavailable currently.

spec

No

Spec object

Definition: Training job specifications.

Constraints: This parameter is mandatory when tasks is left blank. If this parameter is specified, you do not need to specify tasks.

endpoints

No

JobEndpointsReq object

Definition: Configurations required for remotely accessing a training job.

Constraints: N/A

train_type

No

String

Definition: Type of a training job.

Constraints: N/A

Range: SFT (full fine-tuning), PRETRAIN (pre-training), LORA (LoRA fine-tuning), DPO (Direct Preference Optimization), or RFT (reinforcement fine-tuning)

Default Value: N/A

ftjob_config

No

MasJobConfig object

Definition: Fine-tuning training job parameters.

name

Yes

String

Definition: Training job name.

Constraints: N/A

Range: The value must contain 1 to 128 characters consisting of only digits, letters, underscores (_), and hyphens (-).

Default Value: N/A

workspace_id

No

String

Definition: Workspace where a specified job is located.

Constraints: N/A

Range: N/A

Default Value: 0

description

No

String

Definition: Training job description.

Constraints: The value cannot contain angle brackets (<>).

Range: The value can contain 0 to 512 characters.

Default Value: NULL

annotations

No

Map<String,String>

Definition: Advanced functions of a training job. Options:

• "job_template": "Template RL" (heterogeneous job)

• "fault-tolerance/job-retry-num": "3" (number of automatic restarts upon faults)

• "fault-tolerance/job-unconditional-retry": "true" (unconditional restart)

• "fault-tolerance/hang-retry": "true" (restart upon suspension)

• "jupyter-lab/enable": "true" (JupyterLab training application)

• "tensorboard/enable": "true" (TensorBoard training application)

• "mindstudio-insight/enable": "true" (MindStudio Insight training application)

• "fault-tolerance/hccl_op_retry": "true" (operator re-execution)

• "performance_diagnosis_enabled": "true" (performance diagnosis)

• "record/code-dir": "OBS directory" (code backup)

Constraints: N/A

training_experiment_reference

No

TrainingExperimentRequest object

Definition: Training experiment parameters.

Constraints: N/A

id

No

String

Definition: Experiment ID. If you enter an experiment ID, the training job will be added to the existing experiment group.

Constraints: N/A

Range: N/A

Default Value: N/A

name

No

String

Definition: Experiment name. If you enter only the experiment name, the experiment group will be created and the training job will be added to the group.

Constraints: The value can contain only letters, digits, underscores (_), and hyphens (-).

Range: a string of up to 128 characters

Default Value: N/A

description

No

String

Definition: Description.

Constraints: The value cannot contain angle brackets (< or >).

Range: a string of up to 512 characters

Default Value: N/A

id

No

String

Definition: Algorithm ID in algorithm management.

Constraints: N/A

Range: N/A

Default Value: N/A

name

No

String

Definition: Algorithm name.

Constraints: This parameter does not need to be specified.

Range: N/A

Default Value: N/A

subscription_id

No

String

Definition: Subscription ID of a subscription algorithm.

Constraints: This parameter must be used with item_version_id.

Range: N/A

Default Value: N/A

item_version_id

No

String

Definition: Version of a subscription algorithm.

Constraints: This parameter must be used with subscription_id.

Range: N/A

Default Value: N/A

code_dir

No

String

Definition: Code directory of a training job, for example, /usr/app/.

Constraints: This parameter must be used with boot_file. Leave this parameter blank if id, or subscription_id and item_version_id are specified.

Range: N/A

Default Value: N/A

boot_file

No

String

Definition: Boot file of a training job, which must be stored in the code directory, for example, /usr/app/boot.py.

Constraints: This parameter must be used with code_dir. Leave this parameter blank if id, or subscription_id and item_version_id are specified.

Range: N/A

Default Value: N/A

autosearch_config_path

No

String

Definition: YAML configuration path of an auto search job. An OBS URL is required. For example, obs://bucket/file.yaml.

Constraints: N/A

Range: N/A

Default Value: N/A

autosearch_framework_path

No

String

Definition: Framework code directory of an auto search job. An OBS URL is required. For example, obs://bucket/files/.

Constraints: N/A

Range: N/A

Default Value: N/A

command

No

String

Definition: Command for starting the custom image container of a training job. For example, python train.py.

Constraints: N/A

Range: N/A

Default Value: N/A

parameters

No

Array of Parameters objects

Definition: Running parameters of the training job.

Constraints: N/A

policies

No

JobPolicies object

Definition: Policies supported by jobs, which are used for hyperparameter search.

Constraints: N/A

inputs

No

Array of Input objects

Definition: Data input of a training job.

Constraints: N/A

outputs

No

Array of Output objects

Definition: Output of the training job.

Constraints: N/A

engine

No

JobEngine object

Definition: Engine of a training job.

Constraints: Leave this parameter blank if the job is created using id of the algorithm in algorithm management, or subscription_id+item_version_id of the subscribed algorithm.

local_code_dir

No

String

Definition: Local directory of the training container to which the algorithm code directory is downloaded.

Constraints

• The directory must be under /home.

• In v1 compatibility mode, the current field does not take effect.

• When code_dir is prefixed with file://, the current field does not take effect.

• The directory cannot be set to /home/ma-user/modelarts, /home/ma-user/modelarts-dev, /home/ma-user/infer, or their subdirectories, and cannot be set to /home/ma-user.

Range: N/A

Default Value: N/A

working_dir

No

String

Definition: Work directory where an algorithm is executed.

Constraints: In v1 compatibility mode, the current field does not take effect.

Range: N/A

Default Value: N/A

environments

No

Map<String,String>

Definition: Environment variables of a training job. Format: "key":"value" The key can contain a maximum of 8,192 characters, and the value can contain a maximum of 4,096 characters. A maximum of 100 key-value pairs are allowed. The variable name can contain only letters, digits, and underscores (), and must start with a letter or underscore ().

Constraint: The symbol $ cannot be used to reference variables.

summary

No

Summary object

Definition: Visualization log summary.

Constraints: N/A

name

No

String

Definition: Parameter name.

Constraints: N/A

Range: N/A

Default Value: N/A

value

No

String

Definition: Parameter value.

Constraints: N/A

Range: N/A

Default Value: N/A

description

No

String

Definition: Parameter description.

Constraints: N/A

Range: N/A

Default Value: N/A

constraint

No

ParametersConstraint object

Definition: Parameter attribute.

Constraints: N/A

i18n_description

No

I18nDescription object

Definition: Internationalization description.

Constraints: N/A

type

No

String

Definition: Parameter type.

Constraints: N/A.

Range:

• Integer: Integer

• Float: Floating point number

• String: String

• Boolean: Boolean value

Default Value: N/A.

editable

No

Boolean

Definition: Whether the parameter can be edited.

Constraints: N/A

Range:

• true: editable

• false: Not uneditable

Default Value: N/A

required

No

Boolean

Definition: Whether the parameter is mandatory.

Constraints: N/A

Range:

• true: mandatory

• false: optional

Default Value: N/A

sensitive

No

Boolean

Definition: Whether the parameter is sensitive.

Constraints: This function is unavailable currently.

Range

• true: sensitive

• false: insensitive

Default Value: N/A

valid_type

No

String

Definition: Valid type.

Constraints: N/A.

Range:

• Choice: Enumerated values

• Range: Range values

• None: None

Default Value: N/A.

valid_range

No

Array of strings

Definition: Valid range.

Constraints: N/A

language

No

String

Definition: Internationalization language.

Constraints: N/A

Range: N/A

Default Value: N/A

description

No

String

Definition: Description.

Constraints: N/A

Range: N/A

Default Value: N/A

auto_search

No

AutoSearch object

Definition: Hyperparameter search configuration.

Constraints: N/A

skip_search_params

No

String

Definition: Hyperparameter parameters that need to be skipped.

Constraints: N/A

Range: N/A

Default Value: N/A

reward_attrs

No

Array of RewardAttrs objects

Definition: Search metrics.

Constraints: N/A

search_params

No

Array of SearchParams objects

Definition: Search parameters.

Constraints: N/A

algo_configs

No

Array of AlgoConfigs objects

Definition: Search algorithm configurations.

Constraints: N/A

name

No

String

Definition: Metric name.

Constraints: N/A

Range: N/A

Default Value: N/A

mode

No

String

Definition: Search mode.

Constraints: N/A

Range:

• max: A larger metric value is preferred.

• min: A smaller metric value is preferred.

Default Value: N/A

regex

No

String

Definition: Regular expression of a metric.

Constraints: N/A

Range: N/A

Default Value: N/A

name

No

String

Definition: Hyperparameter name.

Constraints: N/A

Range: N/A

Default Value: N/A

param_type

No

String

Definition: Parameter type.

Constraints: N/A

Range:

• continuous: The hyperparameter is of the continuous type. When an algorithm is used in a training job, continuous hyperparameters are displayed as text boxes on the console.

• discrete: The hyperparameter is of the discrete type. When an algorithm is used in a training job, discrete hyperparameters are displayed as drop-down lists on the console.

Default Value: N/A

lower_bound

No

String

Definition: Lower bound of the hyperparameter.

Constraints: N/A

Range: N/A

Default Value: N/A

upper_bound

No

String

Definition: Upper bound of the hyperparameter.

Constraints: N/A

Range: N/A

Default Value: N/A

discrete_points_num

No

String

Definition: Number of discrete points of a hyperparameter with continuous values.

Constraints: N/A

Range: N/A

Default Value: N/A

discrete_values

No

Array of strings

Definition: Discrete hyperparameter values.

Constraints: N/A

name

No

String

Definition: Search algorithm name.

Constraints: N/A

Range: N/A

Default Value: N/A

params

No

Array of AutoSearchAlgoConfigParameter objects

Definition: Search algorithm parameters.

Constraints: N/A

key

No

String

Definition: Parameter key.

Constraints: N/A

Range: N/A

Default Value: N/A

value

No

String

Definition: Parameter value.

Constraints: N/A

Range: N/A

Default Value: N/A

type

No

String

Definition: Parameter type.

Constraints: N/A.

Range:

• Float: Floating point number

• Integer: Integer

• String: String

Default Value: N/A.

engine_id

No

String

Definition: Engine ID selected for a training job.

Constraints: The value can be engine_id, engine_name + engine_version, or image_url.

Range: N/A

Default Value: N/A

engine_name

No

String

Definition: Engine name selected for a training job.

Constraints: If engine_id has been set, you do not need to set this parameter. If you use a preset framework and custom image to create a training job, you must set both this parameter and image_url.

Range: N/A

Default Value: N/A

engine_version

No

String

Definition: Engine version selected for a training job.

Constraints: If engine_id has been set, you do not need to set this parameter.

Range: N/A

Default Value: N/A

image_url

No

String

Definition: Custom image URL selected for a training job. The URL is obtained from SWR.

Constraints: N/A

Range: The format is organization_name/image_name:tag.

Default Value: N/A

install_sys_packages

No

Boolean

Definition: Specifies whether to install the MoXing version specified by the training platform.

Constraints: This parameter is available only when engine_name, engine_version, and image_url are set.

Range:

• true: yes

• false: no

Default Value: N/A

image_repo_id

No

String

Description: SWR enterprise repository instance ID. This parameter is required when a custom image from the enterprise repository is used.

Constraints

• This parameter is mandatory when a custom image from the enterprise repository is used (image_url is set to an enterprise repository image).

• If this parameter is not specified, image_url is a common custom image (shared SWR).

• After this parameter is specified, the system checks whether the user has the permission to access the enterprise repository image and whether the image exists.

Range: Valid SWR enterprise repository instance ID.

Default Value: N/A

log_type

No

String

Definition: Visualization log type of a training job. After this parameter is configured, the training job can be used as the data source of a visualization job.

Constraints: N/A.

Range:

• tensorboard: Logs of the TensorBoard visualization tool type.

• mindstudio-insight: Logs of the MindStudio Insight visualization tool type.

Default Value: N/A.

log_dir

No

LogDir object

Definition: Visualization log output of a training job.

Constraints: This parameter is mandatory when log_type is not left empty.

data_sources

No

Array of DataSource objects

Definition: Visualization log input of the visualization job or training job debugging mode.

Constraints: This parameter is mandatory when the advanced function "tensorboard/enable": "true" or "mindstudio-insight/enable": "true" is enabled for the training job.

pfs

Yes

PFSSummary object

Definition: Visualized log output of an OBS parallel file system.

Constraints: N/A

pfs_path

Yes

String

Definition: URL of the OBS parallel file system, for example, /example/path.

Constraints: N/A

Range: N/A

Default Value: N/A

job

Yes

JobSummary object

Definition: Job data source.

Constraints: Visualization log input of the visualization job or training job debug mode. Before setting this parameter, ensure that "tensorboard/enable": "true" or "mindstudio-insight/enable": "true" is enabled.

nfs

No

NFSSummary object

Definition: SFS Turbo file system data source.

Constraints: Before setting this parameter, ensure that the advanced function mindstudio-insight/enable is set to true or tensorboard/enable is set to true.

job_id

Yes

String

Definition: ID of a training job.

Constraints: N/A

Range: N/A

Default Value: N/A

nfs_path

Yes

String

Definition: URL of the SFS Turbo file system, for example, 140f956f-c31-*.sfsturbo.internal:/.

Constraints: N/A

Range: N/A

Default Value: N/A

role

No

String

Definition: Task role.

Constraints: This function is not supported currently.

Range: N/A

Default Value: N/A

algorithm

No

algorithm object

Definition: Algorithm configurations for algorithm management.

Constraints: N/A

task_resource

No

task_resource object

Definition: Resource flavor of a training job.

Constraints: N/A

log_export_path

No

log_export_path object

Definition: Saved information about training job logs.

Constraints: N/A

job_config

No

job_config object

Definition: Algorithm configuration, such as the boot file.

Constraints: N/A

code_dir

No

String

Definition: Algorithm code directory, for example, /usr/app/.

Constraints: This parameter must be used with boot_file.

Range: N/A

Default Value: N/A

boot_file

No

String

Definition: Code boot file of the algorithm, which must be stored in the code directory, for example, /usr/app/boot.py.

Constraints: This parameter must be used with code_dir.

Range: N/A

Default Value: N/A

engine

No

engine object

Definition: Algorithm engine of a heterogeneous job.

Constraints: N/A

inputs

No

Array of inputs objects

Definition: Data input of an algorithm.

Constraints: N/A

outputs

No

Array of outputs objects

Definition: Data output of an algorithm.

Constraints: N/A

local_code_dir

No

String

Definition: Local directory of the training container to which the algorithm code directory is downloaded.

Constraints:

• The directory must be under /home.

• In v1 compatibility mode, the current field does not take effect.

• When code_dir is prefixed with file://, the current field does not take effect.

Range: N/A

Default Value: N/A

working_dir

No

String

Definition: Work directory where an algorithm is executed.

Constraints: In v1 compatibility mode, the current field does not take effect.

Range: N/A

Default Value: N/A

environments

No

Map<String,String>

Definition: Environment variables of a training job.

Constraints: N/A

Range: N/A

Default Value: N/A

parameters

No

Array of Parameter objects

Definition: Running parameters of an algorithm.

Constraints: N/A

inputs

No

Array of Input objects

Definition: Data input of an algorithm.

Constraints: N/A

outputs

No

Array of Output objects

Definition: Data output of an algorithm.

Constraints: N/A

engine

No

engine object

Definition: Algorithm engine.

Constraints: N/A

name

No

String

Definition: Parameter name.

Constraints: N/A

Range: N/A

Default Value: N/A

value

No

String

Definition: Parameter value.

Constraints: N/A

Range: N/A

Default Value: N/A

description

No

String

Definition: Parameter description.

Constraints: N/A

Range: N/A

Default Value: N/A

constraint

No

constraint object

Definition: Parameter attribute.

Constraints: N/A

i18n_description

No

i18n_description object

Definition: Internationalization description.

Constraints: N/A

type

No

String

Definition: Parameter type.

Constraints: N/A

Range: N/A

Default Value: N/A

editable

No

Boolean

Definition: Whether the parameter can be edited.

Constraints: N/A

Range:

• true: editable

• false: Not uneditable

Default Value: N/A

required

No

Boolean

Definition: Whether the parameter is mandatory.

Constraints: N/A

Range:

• true: mandatory

• false: optional

Default Value: N/A

sensitive

No

Boolean

Definition: Whether the parameter is sensitive.

Constraints: This function is unavailable currently.

Range:

• true: sensitive

• false: insensitive

Default Value: N/A

valid_type

No

String

Definition: Valid type.

Constraints: N/A

Range: N/A

Default Value: N/A

valid_range

No

Array of strings

Definition: Valid range.

Constraints: N/A

language

No

String

Definition: Internationalization language.

Constraints: N/A

Range: N/A

Default Value: N/A

description

No

String

Definition: Internationalization language description.

Constraints: N/A

Range: N/A

Default Value: N/A

name

Yes

String

Definition: Name of the data input channel.

Constraints: N/A

Range: N/A

Default Value: N/A

description

No

String

Definition: Description of the data input channel.

Constraints: N/A

Range: N/A

Default Value: N/A

local_dir

No

String

Definition: Local path of the container to which the data input channels are mapped. Example: /home/ma-user/modelarts/inputs/data_url_0

Constraints: N/A

Range: N/A

Default Value: N/A

access_method

No

String

Definition: Access method of the input data channel path (local_dir).

Constraints: N/A

Range:

• parameter: hyperparameters

• env: environment variables

Default Value: parameter

remote

Yes

InputDataInfo object

Description: actual data input. Options:

• dataset: The data input is a dataset.

• obs: The data input is an OBS path.

Constraints: N/A

remote_constraint

No

Array of remote_constraint objects

Definition: Data input constraint.

Constraints: N/A

data_type

No

String

Definition: data input type, which can be data storage location or dataset. The dataset function will be brought offline soon.

Constraints: N/A

Range: N/A

Default Value: N/A

attributes

No

String

Definition: Related attributes.

Constraints: N/A

Range:

If the input is a dataset:

• data_format: data format

• data_segmentation: data segmentation method

• dataset_type: data labeling type

Default Value: N/A

name

Yes

String

Definition: Name of the data output channel.

Constraints: N/A

Range: N/A

Default Value: N/A

description

No

String

Definition: Description of the data output channel.

Constraints: N/A

Range: N/A

Default Value: N/A

local_dir

No

String

Definition: Local path of the container to which the data output channels are mapped.

Constraints: N/A

Range: N/A

Default Value: N/A

access_method

No

String

Definition: Access method of the output data channel path (local_dir).

Constraints: N/A

Range:

• parameter: hyperparameters

• env: environment variables

Default Value: parameter

remote

Yes

Remote object

Definition: Description of the actual data output.

Constraints: N/A

obs

Yes

RemoteObs object

Definition: Data actually output to OBS.

Constraints: N/A

obs_url

Yes

String

Definition: Data is actually output to the OBS path, for example, obs://example/path.

Constraints: N/A

Range: N/A

Default Value: N/A

engine_id

No

String

Definition: Engine ID selected for an algorithm.

Constraints: N/A

Range: N/A

Default Value: N/A

engine_name

No

String

Definition: Engine name selected for an algorithm.

Constraints: If engine_id is specified, leave this parameter blank.

Range: N/A

Default Value: N/A

engine_version

No

String

Definition: Engine version selected for an algorithm.

Constraints: If engine_id is specified, leave this parameter blank.

Range: N/A

Default Value: N/A

image_url

No

String

Definition: Custom image URL selected for an algorithm, for example, train-image/pytorch_2_1_xxx:1.0.1. The URL is obtained from SWR.

Constraints: N/A

Range: N/A

Default Value: N/A

run_user

No

String

Definition: user for starting the container image.

Constraints: This parameter can only be configured in the custom image scenario.

Range: N/A

Default Value: 1000

engine_id

No

String

Definition: ID of the engine flavor of a heterogeneous job, for example, caffe-1.0.0-python2.7.

Constraints: N/A

Range: N/A

Default Value: N/A

engine_name

No

String

Definition: Name of the engine flavor of a heterogeneous job, for example, Caffe.

Constraints: N/A

Range: N/A

Default Value: N/A

engine_version

No

String

Definition: Version of the engine flavor of a heterogeneous job.

Constraints: N/A

Range: N/A

Default Value: N/A

image_url

No

String

Definition: Custom image URL selected for an algorithm.

Constraints: N/A

Range: N/A

Default Value: N/A

run_user

No

String

Definition: user for starting the container image.

Constraints: This parameter can only be configured in the custom image scenario.

Range: N/A

Default Value: 1000

name

Yes

String

Definition: Name of the data input channel.

Constraints: N/A

Range: N/A

Default Value: N/A

description

No

String

Definition: Description of the data input channel.

Constraints: N/A

Range: N/A

Default Value: N/A

local_dir

No

String

Definition: Local path of the container to which the data input channels are mapped.

Constraints: N/A

Range: N/A

Default Value: N/A

remote

Yes

InputDataInfo object

Description: actual data input. Options:

• dataset: The data input is a dataset.

• obs: The data input is an OBS path.

Constraints: N/A

dataset

No

dataset object

Definition: The input is a dataset. Both the new and old dataset functions are supported. The old dataset function will be brought offline. You are advised to use the new dataset function.

Constraints: N/A

obs

No

obs object

Definition: OBS in which data input and output are stored.

Constraints: N/A

id

Yes

String

Definition: Dataset ID of a training job.

Constraints: N/A

Range: N/A

Default Value: N/A

version_id

No

String

Definition: Dataset version ID of a training job.

Constraints: This parameter is mandatory when an old-version dataset is used, that is, service_type is not V3.

Range: N/A

Default Value: N/A

obs_url

No

String

Definition: OBS URL of the dataset for a training job. It is automatically parsed by ModelArts based on the dataset ID and dataset version ID. For example, /usr/data/.

Constraints: N/A

Range: N/A

Default Value: N/A

service_type

No

String

Definition: Dataset service type.

Constraints: N/A

Range: The value V4 indicates that a new-version dataset asset is used. The value V3 indicates that an old-version v3 dataset is used. Other values indicate that an old-version v2 dataset is used. Use V4 unless otherwise specified.

Default Value: N/A

name

No

String

Definition: Dataset name of a training job.

Constraints: This parameter is mandatory when a new-version dataset is used.

Range: N/A

Default Value: N/A

dataset_proportion

No

Integer

Definition: Dataset ratio for fine-tuning a training job, indicating the percentage of the dataset used for training.

Constraints: Only datasets of the new version are supported.

Range: 0 to 100.

Default Value: N/A

obs_url

Yes

String

Definition: OBS URL of the dataset for a training job, Example: obs://example/path.

Constraints: N/A

Range: N/A

Default Value: N/A

name

Yes

String

Definition: Name of the data output channel.

Constraints: N/A

Range: N/A

Default Value: N/A

description

No

String

Definition: Description of the data output channel.

Constraints: N/A

Range: N/A

Default Value: N/A

local_dir

No

String

Definition: Local path of the container to which the data output channels are mapped.

Constraints: N/A

Range: N/A

Default Value: N/A

remote

Yes

remote object

Definition: Description of the actual data output.

Constraints: N/A

obs

Yes

obs object

Definition: Data actually output to OBS.

Constraints: N/A

obs_url

Yes

String

Definition: Path of the data output to OBS.

Constraints: N/A

Range: N/A

Default Value: N/A

flavor_id

No

String

Definition: ID of the resource flavor selected for a training job.

Constraints: N/A

Range: N/A

Default Value: N/A

node_count

Yes

Integer

Definition: Number of resource replicas selected for a training job.

Constraints: N/A

Range: greater than or equal to 1

Default Value: N/A

pool_id

No

String

Definition: ID of the resource pool selected for a training task.

Constraints: N/A

Range: N/A

Default Value: N/A

obs_url

No

String

Definition: OBS path for storing training job logs, for example, obs://example/path.

Constraints: N/A

Range: N/A

Default Value: N/A

resource

No

SpecResource object

Definition: Resource flavor of a training job.

Constraints: Select either flavor_id or pool_id or flavor_id.

• If you select a public resource pool, only flavor_id is needed. Select the number of PUs and memory your training job needs. If the public resource pool has enough idle resources, your job will be scheduled.

• If you select a dedicated resource pool, both pool_id and flavor_id are needed. Select the smallest number of PUs that meet your training needs to save resources and boost efficiency.

volumes

No

Array of SpecVolumes objects

Definition: Mounting volume information of a training job.

Constraints: N/A

log_export_path

No

LogExportPath object

Definition: Log output of a training job.

Constraints: N/A

auto_stop

No

AutoStop object

Definition: Auto stop configuration of a training job.

Constraints: N/A

retention

No

RetentionPolicy object

Definition: Automatic aging configuration for job-level library records.

Constraints: This parameter is specified only during job creation and cannot be modified after the job is created.

Range: N/A

Default Value: If this parameter is not passed, it indicates that no separate setting is made.

schedule_policy

No

SchedulePolicy object

Definition: Scheduling policy of a training job.

Constraints: N/A

log_export_config

No

LogExportConfig object

Definition: Log export configuration.

Constraints: N/A.

notification

No

Notification object

Definition: Message notification of a training event.

Constraints: N/A

custom_metrics

No

Array of CustomMetrics objects

Definition: Metric collection configuration.

output_model

No

OutputModel object

Definition: Output information of a custom training job.

asset_model

No

AssetModel object

Definition: Information about the model created using a custom training job.

asset_id

No

String

Definition: Model ID of the fine-tuning training job.

reserved_time

No

ReserveTime object

Definition: Retention duration configuration of a training job.

Constraints: N/A

flavor_id

No

String

Definition: ID of the resource flavor of a training job.

Constraints: The flavor_id parameter cannot be specified for a dedicated resource pool of CPU specifications.

Range: The options for GPU- or Ascend-based dedicated resource pools are as follows:

• modelarts.pool.visual.xlarge (1 PU)

• modelarts.pool.visual.2xlarge (2 PUs)

• modelarts.pool.visual.4xlarge (4 PUs)

• modelarts.pool.visual.8xlarge (8 PUs)

• modelarts.pool.visual.16xlarge (16 cards, only for the Snt9b23 supernode resource pool)

Default Value: N/A

node_count

Yes

Integer

Definition: Number of nodes used to create a training job in a resource pool.

Constraints: N/A

Range: greater than or equal to 1

Default Value: N/A

pool_id

No

String

Definition: Dedicated resource pool ID.

Constraints: N/A

Range: N/A

Default Value: N/A

pool_group_id

No

String

Definition: Resource pool federation ID.

Constraints: This parameter is mandatory when kind is set to federated_pool_job.

Range: N/A

Default Value: N/A

main_container_customized_flavor

No

MainContainerCustomizedFlavor object

Definition: Custom flavor.

Constraints: The number of CPU cores and memory size must be greater than 0, and the number of accelerator cards must be greater than or equal to 0.

cpu_core_num

No

Float

Definition: Number of CPU cores.

Range: greater than 0

Constraints: N/A

Default Value: N/A

mem_size

No

Float

Definition: Memory size, in GB.

Range: greater than 0

Constraints: N/A

Default Value: N/A

accelerator_num

No

Float

Definition: Number of accelerator cards.

Range: greater than or equal to 0

Constraints: N/A

Default Value: N/A

nfs

No

Nfs object

Definition: NFS mounting volume information of a training job.

Constraints: N/A

pfs

No

Pfs object

Definition: obsfs mounting volume information of a training job.

Constraints: N/A

obs

No

Obs object

Definition: OBS mounting volume information of a training job.

Constraints: N/A

nfs_server_path

No

String

Definition: NFS server path, for example, 10.10.10.10:/example/path.

Constraints: N/A

Range: N/A

Default Value: N/A

local_path

No

String

Definition: Path for attaching volumes to the training container, for example, /example/path.

Constraints: N/A

Range: N/A

Default Value: N/A

read_only

No

Boolean

Definition: Specifies whether the disks attached to the container in NFS mode are read-only.

Constraints: N/A

Range:

• true: read only

• false: non-read-only

Default Value: N/A

pfs_path

No

String

Definition: Address of obsfs. For example, /test-bucket/path.

Constraints: N/A

Range: N/A

Default Value: N/A

local_path

No

String

Definition: Path for attaching volumes to the training container, for example, /example/path.

Constraints: N/A

Range: N/A

Default Value: N/A

obs_path

No

String

Definition: OBS path to be mounted. For example, /test-bucket/path.

Constraints: N/A

Range: N/A

Default Value: N/A

local_path

No

String

Definition: Path for attaching volumes to the training container, for example, /example/path.

Constraints: N/A

Range: N/A

Default Value: N/A

obs_url

No

String

Definition: OBS path for storing training job logs, for example, obs://example/path.

Constraints: N/A

Range: N/A

Default Value: N/A

host_path

No

String

Definition: Path of the host where training job logs are stored, for example, /example/path.

Constraints: N/A

Range: N/A

Default Value: N/A

time_unit

Yes

String

Definition: Time unit.

Constraints: N/A

Range

• HOURS: hour

• MINUTES minute

Default Value: N/A

duration

Yes

Integer

Definition: Runtime.

Constraints: N/A

Range: The minimum value is 1. If time_unit is set to HOURS, the minimum value is 1 hour.

Default Value: N/A

policy

No

String

Definition: Automatic aging policy for job-level library records.

Constraints

• This parameter takes effect only when the job aging capability is enabled on the platform and the job type is custom training job (kind=job).

• This parameter is associated with the user-level automatic job aging switch:

  • If the user-level switch is enabled, all jobs of the user are aged (jobs with policy=disabled cannot be exempted separately).

  • If the user-level switch is disabled, only jobs with policy=enabled are aged. Jobs with no policy set or policy=disabled are not aged.

  Range:

• enabled: Enable job aging.

• disabled: Disable job aging (only valid when the user-level switch is disabled).

Default Value: If this parameter is not passed, it is not set separately and follows the user-level switch policy.

required_affinity

No

RequiredAffinity object

Definition: Mandatory affinity of the training job.

Constraints: N/A.

preferred_affinity

No

PreferredAffinity object

Definition: Preferred affinity of the training job.

Constraints: N/A.

priority

No

Integer

Definition: Priority of a training job.

Constraints:

• The priority can be set for a training job only when a dedicated resource pool is used.

• The value ranges from 1 to 3. The default priority is 1, and the highest priority is 3.

By default, the job priority can be set to 1 or 2. After the permission to set the highest job priority is configured, the priority can be set to 1 to 3.

Range: 0 to 3

Default Value: N/A

preemptible

No

Boolean

Definition: Whether the resource can be preempted.

Constraints: N/A

Range:

• true: The resource can be preempted.

• false: The resource cannot be preempted.

Default Value: N/A

affinity_type

No

String

Definition: Affinity scheduling policy.

Constraints: If the affinity scheduling policy is hyperinstance, the resource pool must be a supernode resource pool. Otherwise, scheduling cannot be performed.

Range

• cabinet: strong cabinet scheduling

• hyperinstance: supernode affinity scheduling

• networkTopology: network topology affinity scheduling

Default Value: N/A

job_level

No

String

Definition: Overall network topology constraints of a job.

Constraints: This parameter is valid only when affinity_type is set to networkTopology. The system schedules all tasks of a job to a node group whose level is no higher than job_level.

When a user delivers a training job to the supernode resource pool, if the overall network topology constraints of the job are not set, the system sets the value to cluster by default.

Range

• cluster: resource pool

• hyperinstanceGroup: supernode

Default Value: cluster

affinity_group_size

No

Integer

Definition: Affinity group size.

Constraints: This parameter is mandatory when affinity_type is set to hyperinstance or networkTopology. The system schedules affinity_group_size tasks to a supernode to form an affinity group. The value of affinity_group_size cannot exceed the step of the supernode.

When a user delivers a training job to the supernode resource pool, if the affinity group size is not set, the system sets the value to 1 by default.

Range: N/A

Default Value: 1

affinity_group_level

No

String

Definition: Network topology constraints of an affinity group.

Constraints: This parameter is valid only when affinity_type is set to networkTopology. The system schedules the affinity group consisting of affinity_group_size tasks to a node group whose level is not higher than affinity_group_level.

When a user delivers a training job to the supernode resource pool, if the network topology constraints of the affinity group are not set, the system sets the value to hyperinstanceGroup by default.

Range

• hyperinstance: supernode

• slice: cabinet

Default Value: hyperinstanceGroup

node_affinity

No

NodeSelector object

Definition: Node affinity.

Constraints: N/A

nodeSelectorTerms

Yes

Array of NodeSelectorTerm objects

Definition: List of node selector items. These items are in the OR relationship.

Constraints: N/A

matchExpressions

No

Array of NodeSelectorRequirement objects

Definition: Node selector requirements listed by node tag.

Constraints: N/A.

Range: N/A.

Default Value: N/A.

matchFields

No

Array of NodeSelectorRequirement objects

Definition: Node selector requirements listed by node field.

Constraints: N/A.

Range: N/A.

Default Value: N/A.

node_affinity

No

Array of PreferredSchedulingTerm objects

Definition: The scheduler prioritizes scheduling Pods onto nodes that satisfy the affinity expressions specified in this field; however, it may still select a node that violates one or more of these expressions. The most preferred nodes are those with the highest total weight. For each node that meets all scheduling requirements (such as resource requests and required affinity expressions), the scheduler calculates a sum by iterating through the elements in this field. If a node satisfies a specific match expression, its corresponding "weight" is added to the total. The node with the highest cumulative weight is given the highest priority for selection.

Constraints: N/A

weight

No

Integer

Definition: Weight associated with the matched nodeSelectorTerm.

Constraints: N/A

Range: 1 to 100.

Default Value: N/A

preference

No

NodeSelectorTerm object

Definition: Node selector item, which is associated with the corresponding weight.

Constraints: N/A

key

Yes

String

Definition: Label key to which the selector is applied.

Constraints: N/A.

Range: N/A.

Default Value: N/A.

operator

Yes

String

Definition: Relationship between the key and a group of values.

Constraints: N/A

Range

• In: The value of the key must be in the provided list of values. For example, if the key is color and the value list is ["red", "blue"], then color In ["red", "blue"] means the color must be red or blue.

• NotIn: The value of the key must not be in the provided list of values. For example, color NotIn ["red", "blue"] means the color cannot be red or blue.

• Exists: The key must exist, but there is no requirement on the value. For example, color Exists means that the color key must exist regardless of the value.

• DoesNotExist: The key cannot exist. For example, color DoesNotExist means that the color key cannot exist.

• Gt: The value of the key must be greater than the given value. For example, if the key is age, age Gt 18 indicates that the value of age must be greater than 18.

• Lt: The value of the key must be less than the specified value. For example, age Lt 18 means the value of age must be less than 18.

Default Value: N/A

values

No

Array of strings

Definition: An array of string values.

Constraints: If the operator is In or NotIn, the value array cannot be empty. If the operator is Exists or DoesNotExist, the value array cannot be empty. If the operator is Gt or Lt, the value array must contain one element, which will be interpreted as an integer.

version

No

String

Definition: Log version.

Constraints: N/A

Range: v0 or v1

Default Value: v0.

rotation_enabled

No

Boolean

Definition: Specifies whether to enable the download of logs by time segment.

Constraints: N/A.

Range:

• true: Enable the download of logs by time segment.

• false: Disable the download of logs by time segment.

Default Value: false.

topic_urn

No

String

Definition: URN of the selected topic in SMN.

Constraints: N/A

Range: N/A

Default Value: N/A

events

No

Array of strings

Definition: Training event that triggers a notification.

Constraints: The options are as follows:

• JobStarted: The job is started.

• JobCompleted: The job is completed.

• JobFailed: The job is failed.

• JobTerminated: The job is terminated.

• JobRestarted: The job is restarted.

• JobHanged: The job is suspended.

• JobPreempted: The job is preempted.

exec

No

Exec object

Definition: Metrics are collected in CLI mode.

http_get

No

HttpGet object

Definition: Metrics are collected in HTTP mode.

command

No

Array of strings

Definition: Metrics are collected in CLI mode.

path

No

String

Definition: URL for obtaining metrics through HTTP.

Constraints: This parameter and the following port parameter must be specified or left blank at the same time.

Range: N/A

Default Value: N/A

port

No

Integer

Definition: Port for obtaining metrics through HTTP.

Constraints: Both the URL and the port must either be configured together or remain empty.

Range: N/A

Default Value: N/A

obs

No

ObsModel object

Definition: OBS output information stored by a custom training job.

Constraints: N/A

obs_path

Yes

String

Definition: OBS path for storing custom training jobs, for example, obs://example/path.

Range: N/A

local_path

No

String

Definition: Path of the host where custom training jobs are stored, for example, /example/path.

Range: N/A

name

Yes

String

Definition: Model name.

Constraints: N/A

Range: N/A

Default Value: N/A

code

No

String

Definition: model code.

Constraints: N/A

Range: N/A

Default Value: N/A

version

Yes

String

Definition: Model version.

Constraints: N/A

Range: N/A

Default Value: N/A

desc

No

String

Definition: model version description.

Constraints: N/A

Range: N/A

Default Value: N/A

series

No

String

Definition: Model brand.

Constraints: N/A

Range: N/A

Default Value: N/A

type

Yes

String

Definition: Model type.

Constraints: N/A

Range: N/A

Default Value: N/A

model_desc

No

String

Definition: model asset description.

Constraints: N/A

Range: N/A

Default Value: N/A

parent_asset_id

No

String

Definition: (Optional) parent asset ID, which is transferred when an existing model is selected.

Constraints: N/A

Range: N/A

Default Value: N/A

time_unit

Yes

String

Definition: Time unit.

Constraints: N/A

Range:

• HOURS (hour)

Default Value: N/A

duration

Yes

Integer

Definition: Retention duration.

Constraints: N/A

Range: The minimum value is 1.

Default Value: N/A

ssh

No

SSHReq object

Definition: SSH connection information.

Constraints: N/A

key_pair_names

No

Array of strings

Definition: Name of the SSH key pair, which can be created and viewed on the Key Pair page of the Elastic Cloud Server (ECS) console.

Constraints: N/A

ft_job_uuid

No

String

Definition: Fine-tuning job UUID, which can be obtained by calling the API for creating a fine-tuning job.

Constraints: N/A

Range: N/A

Default Value: N/A

ft_train_type

No

String

Definition: Training type of a fine-tuning job. The supported training types include pre-training, full fine-tuning, and LoRA fine-tuning.

Constraints: N/A

Range: SFT (full fine-tuning), PRETRAIN (pre-training), and LORA (LoRA fine-tuning)

Default Value: N/A

model_type

No

String

Definition: Type of the asset model selected for the training job. Supported model types include text generation and image understanding.

Constraints: N/A

Range: TextGeneration (text generation) and ImageUnderstanding (image understanding)

Default Value: N/A

train_output_path

No

String

Definition: Path for storing the checkpoints or final outputs generated during fine-tuning. This path is an OBS path and is configured by you during training job creation, for example, obs://yyy/test/.

Constraints: N/A

Range: N/A

Default Value: N/A

train_process

No

Double

Definition: progress of fine-tuning a training job.

Constraints: N/A

Range: N/A

Default Value: N/A

checkpoint_id

No

String

Definition: Breakpoint ID. When a training job is resumed from a checkpoint, this field records the unique identifier (UUID) of the checkpoint.

Constraints: N/A

Range: UUID.

Default Value: N/A

task_env

No

TaskEnv object

Definition: Fine-tuning job training parameters.

checkpoint_config

No

CheckpointConf object

Definition: Breakpoint configuration.

envs

No

Array of EnvVar objects

Definition: Fine-tuning environment variable information.

Constraints: N/A

label

No

String

Definition: Tag.

Constraints: N/A

Range: N/A

Default Value: N/A

des

No

String

Definition: Description.

Constraints: N/A

Range: N/A

Default Value: N/A

env_name

No

String

Definition: Environment variable name.

Constraints: N/A

Range: N/A

Default Value: N/A

env_type

No

String

Definition: environment variable type.

Constraints: N/A

Range: N/A

Default Value: N/A

value

No

String

Definition: environment variable value.

Constraints: N/A

Range: N/A

Default Value: N/A

modifiable

No

Boolean

Definition: Specifies whether the value can be modified.

Constraints: N/A

Range

• true: yes

• false: no

Default Value: N/A

displayable

No

Boolean

Definition: Specifies whether to display.

Constraints: N/A

Range

• true: displayed.

• false: not displayed.

Default Value: N/A

used_steps

No

Array of strings

Definition: phase when the environment variable is used.

Constraints: N/A

checkpoint_id

No

String

Definition: breakpoint ID.

Range: UUID.

save_checkpoints_max

No

Integer

Definition: Specifies the number of steps to be saved for a training resumption task.

Range

• 0: disabled.

• 1: automatic and unlimited.

skipped_steps

No

Integer

Definition: Specifies whether to skip steps during resumable training. If steps are not skipped, training interruptions from hardware or network issues are fixed. If steps are skipped, some data after the checkpoint (set by users) is ignored to prevent poor data quality from causing the loss to not converge.

Range

• 0: not skipped.

• 1: skipped.

restore_training

No

Integer

Definition: Specifies whether to resume a training task.

Range

• 0: No.

• 1: Yes.

kind

String

Definition: Type of a training job.

Range

• job: common job

• federated_pool_job: resource pool federated job

• edge_job: edge job

• hetero_job: heterogeneous job

• mrs_job: MRS job

• autosearch_job: auto search job

• diag_job: diagnosis job

• visualization_job: visualization job

metadata

JobMetadataResponse object

Definition: Training job metadata.

status

Status object

Definition: Training job status information.

algorithm

JobAlgorithmResponse object

Definition: Training job algorithm.

tasks

Array of TaskResponse objects

Definition: Heterogeneous training tasks.

spec

SpecResponse object

Definition: Training job specifications.

endpoints

JobEndpointsResp object

Definition: Configurations required for remotely accessing a training job.

ftjob_config

MasJobConfig object

Definition: Fine-tuning training job parameters.

id

String

Definition: Training job ID, which is generated and returned by ModelArts after a training job is created.

Range: N/A

name

String

Definition: Name of a training job.

Range: The value must contain 1 to 64 characters consisting of only digits, letters, underscores (_), and hyphens (-).

workspace_id

String

Definition: Workspace where a specified job is located.

Range: N/A

description

String

Definition: Definition of a training job.

Range: N/A

create_time

Long

Definition: Time when a training job was created, in milliseconds. The value is generated and returned by ModelArts after a training job is created.

Range: N/A

user_name

String

Definition: Username for creating a training job. The username is generated and returned by ModelArts after a training job is created.

Range: N/A

annotations

Map<String,String>

Definition: advanced function configuration of a training job. The key is the function switch/configuration name, and the value is a string. (Even if the value is a number or Boolean value, it is represented by a string, for example, "true" or "3".)

training_experiment_reference

TrainingExperimentResp object

Definition: Training experiment parameters.

name

String

Definition: Experiment name.

Range: The value can contain a maximum of 128 characters. Special characters are not allowed.

id

String

Parameter Explanation: Experiment ID.

Value Range: N/A.

serial_number

String

Definition: sequence number of the current training job in the training experiment to which the job belongs. The default value is 0.

phase

String

Definition: Level-1 status of a training job.

Range:

• Creating: The job is being created.

• Pending: The job is pending.

• Running: The job is running.

• Failed: The job failed to run.

• Completed: The job is complete.

• Terminating: The job is being stopped.

• Terminated: The job has been stopped.

• Abnormal: The job is abnormal.

secondary_phase

String

Definition: Level-2 status of a training job. The values are internal detailed statuses and may be added, changed, or deleted. Dependency on the status is not recommended.

Range:

• Creating: The job is being created.

• Queuing: The job is queuing.

• Running: The job is running.

• Failed: The job failed to run.

• Completed: The job is complete.

• Terminating: The job is being stopped.

• Terminated: The job has been stopped.

• CreateFailed: The job fails to be created.

• TerminatedFailed: The job fails to be stopped.

• Unknown: The job is in an unknown state.

• Lost: The job is abnormal.

duration

Long

Definition: Running duration of a training job, in ms.

Range: N/A

node_count_metrics

Array<Array<Integer>>

Definition: number of nodes during training job running. Each inner array indicates a [time point, number of running nodes] 2-tuple, which records the number of nodes running at a specific time point.

tasks

Array of strings

Definition: Training job subtask name.

start_time

Long

Definition: training job start time. The value is a Unix timestamp, in milliseconds.

Range: N/A

task_statuses

Array of TaskStatuses objects

Definition: Status of the first failed subtask of a training job.

running_records

Array of RunningRecord objects

Definition: Running and fault recovery records of a training job.

retention_time

Integer

Definition: Retention period of a job.

reserved_time is set during training job creation.

Range: N/A

Default value: N/A

task_ips

Array of TaskIP objects

Definition: IP addresses of tasks in a training job.

host_ips, and only the records that match the specified IP addresses are returned.

Range: N/A

host_ips is not passed, no value is returned.

task

String

Definition: Training job subtask name.

Range: N/A

exit_code

Integer

Definition: Exit code of a training job subtask.

Range: N/A

message

String

Definition: Error message of a training job subtask.

Range: N/A

start_at

Long

Definition: Unix timestamp of the start time in the current running record, in seconds.

Range: N/A

end_at

Long

Definition: Unix timestamp of the end time in the current running record, in seconds.

Range: N/A

xpu_start_at

Long

Definition: Unix timestamp of the accelerator card startup time in the current running record, in seconds.

Range: N/A

start_type

String

Definition: Startup mode of the current execution.

Range

• init_or_rescheduled: This startup is the first running after scheduling, including the first startup and the running after scheduling recovery.

• restarted: This startup is not the first running after scheduling but the running after a process restart.

end_reason

String

Definition: Reason why the running ends.

Range: N/A

end_related_task

String

Definition: ID of the task worker (for example, worker-0) that ends the running.

Range: N/A

end_recover

String

Definition: Fault tolerance policy adopted when the execution ends abnormally.

Range

• npu_proc_restart: NPU in-place hot recovery

• proc_restart: in-place process recovery

• npu_step_retry: step recomputation

• pod_reschedule: pod-level rescheduling

• job_reschedule: job-level rescheduling

• job_reschedule_with_taint: isolated job-level rescheduling

end_recover_before_downgrade

String

Definition: There is a downgrade relationship between policies. If a policy fails to be executed, it will be downgraded to another specified policy. end_recover_before_downgrade indicates the tolerance policy used before end_recover is downgraded.

Range: same as that of end_recover.

recover_records

Array of RecoverRecord objects

Definition: details about all fault tolerance policies adopted when the execution ends abnormally.

recover_start_at

Long

Definition: Unix timestamp of the start time of the fault tolerance policy, in seconds. The timestamp is also the fault occurrence time.

Range: N/A

recover_end_at

Long

Definition: Unix timestamp of the end time of the fault tolerance policy, in seconds.

Range: N/A

recover

String

Definition: Fault tolerance policy.

Range

• npu_step_retry: step recomputation

• npu_proc_restart: NPU in-place hot recovery

• proc_restart: in-place process recovery

• pod_reschedule: pod-level rescheduling

• job_reschedule: job-level rescheduling

• job_reschedule_with_taint: isolated job-level rescheduling

fault_scenario

String

Definition: Fault scenario.

Range

• chip_fault: chip fault

• node_fault: node fault

• job_failed: job exit upon a failure

• job_hanged: job suspension

• job_subhealth: job subhealth

• error_in_log: log exception

reason

String

Definition: Fault cause.

Range: N/A

related_task

String

Definition: ID of the task worker (for example, worker-0) that ends the running.

Range: N/A

recover_result

String

Definition: Fault recovery result.

Range

• recovering

• success

• failed

• downgrade: policy downgrade

• terminated: The policy is terminated.

• quotaExceeded: The number of policy executions exceeds the limit.

task

String

Task name, for example, worker-0.

ip

String

Task or pod IP address.

host_ip

String

Host IP address.

schedule_count

Integer

Number of times the current task is scheduled. The default value is 1.

The value increases in scenarios such as rescheduling and preemption.

id

String

Definition: algorithm ID of a training job.

name

String

Definition: Algorithm name.

Range: N/A

subscription_id

String

Definition: Subscription ID of a subscription algorithm, which must be used with item_version_id.

Range: N/A

item_version_id

String

Definition: Version of a subscription algorithm, which must be used with subscription_id.

Range: N/A

code_dir

String

Definition: Code directory of a training job, for example, /usr/app/. This parameter must be used with boot_file. Leave this parameter blank if id, or subscription_id and item_version_id are specified.

Range: N/A

boot_file

String

Definition: Boot file of a training job, which must be stored in the code directory, for example, /usr/app/boot.py. This parameter must be used with code_dir. Leave this parameter blank if id, or subscription_id and item_version_id are specified.

Range: N/A

autosearch_config_path

String

Definition: YAML configuration path of an auto search job. An OBS URL is required. For example, obs://bucket/file.yaml.

Range: N/A

autosearch_framework_path

String

Definition: Framework code directory of an auto search job. An OBS URL is required. For example, obs://bucket/files/.

Range: N/A

command

String

Definition: Boot command for starting the container of a custom image for a training job. For example, python train.py.

Range: N/A

parameters

Array of ParameterResp objects

Definition: Running parameters of the training job.

policies

policies object

Definition: Policy supported by a job.

inputs

Array of InputResp objects

Definition: Data input of a training job.

outputs

Array of OutputResp objects

Definition: Output of the training job.

engine

JobEngineResp object

Definition: Engine of a training job. Leave this parameter blank if the job is created using id of the algorithm in algorithm management, or subscription_id+item_version_id of the subscribed algorithm.

local_code_dir

String

Definition: Local directory of the training container to which the algorithm code directory is downloaded. The rules are as follows:

• The directory must be under /home.

• In v1 compatibility mode, the current field does not take effect.

• When code_dir is prefixed with file://, the current field does not take effect.

Range: N/A

working_dir

String

Definition: Work directory where an algorithm is executed. Rules:

In v1 compatibility mode, this parameter does not take effect.

Range: N/A

environments

Array of Map<String,String> objects

Definition: Environment variables of a training job. The format is key:value. Leave this parameter blank.

summary

SummaryResp object

Definition: Visualization log summary.

name

String

Definition: Parameter name.

Range: N/A

value

String

Definition: Parameter value.

Range: N/A

description

String

Definition: Parameter description.

Range: N/A

constraint

constraint object

Definition: Parameter attribute.

i18n_description

i18n_description object

Definition: Internationalization description.

type

String

Definition: Parameter type.

Range:

• Integer: Integer

• Float: Floating point number

• String: String

• Boolean: Boolean value

editable

Boolean

Definition: Whether the parameter can be edited.

Range:

• true: editable

• false: Not uneditable

required

Boolean

Definition: Whether the parameter is mandatory.

Range:

• true: mandatory

• false: optional

sensitive

Boolean

Definition: Whether the parameter is sensitive. This function is unavailable currently.

Range:

• true: sensitive

• false: insensitive

valid_type

String

Definition: Valid type.

Range:

- Choice: Enumerated values

- Range: Range values

- None: None

valid_range

Array of strings

Definition: Valid range.

language

String

Definition: Internationalization language. The options are as follows:

• zh-cn: Chinese

• en-us: English](tag:hc,hk)

Range: N/A

description

String

Definition: Internationalization language description.

Range: N/A

auto_search

auto_search object

Definition: Hyperparameter search configuration.

skip_search_params

String

Definition: Hyperparameter parameters that need to be skipped.

Range: N/A

reward_attrs

Array of reward_attrs objects

Definition: Search metrics.

search_params

Array of search_params objects

Definition: Search parameters.

algo_configs

Array of algo_configs objects

Definition: Search algorithm configurations.

name

String

Definition: Metric name.

Range: N/A

mode

String

Definition: Search mode.

Range:

• max: A larger metric value is preferred.

• min: A smaller metric value is preferred.

regex

String

Definition: Regular expression of a metric.

Range: N/A

name

String

Definition: Hyperparameter name.

Range: N/A

param_type

String

Definition: Parameter type.

Range:

• continuous: The hyperparameter is of the continuous type. When an algorithm is used in a training job, continuous hyperparameters are displayed as text boxes on the console.

• discrete: The hyperparameter is of the discrete type. When an algorithm is used in a training job, discrete hyperparameters are displayed as drop-down lists on the console.

lower_bound

String

Definition: Lower bound of the hyperparameter.

Range: N/A

upper_bound

String

Definition: Upper bound of the hyperparameter.

Range: N/A

discrete_points_num

String

Definition: Number of discrete points of a hyperparameter with continuous values.

Range: N/A

discrete_values

Array of strings

Definition: Discrete hyperparameter values.

name

String

Definition: Search algorithm name.

Range: N/A

params

Array of AutoSearchAlgoConfigParameterResp objects

Definition: Search algorithm parameters.

key

String

Definition: Parameter key.

Range: N/A

value

String

Definition: Parameter value.

Range: N/A

type

String

Definition: Parameter type.

Range:

• obs: Data storage location (OBS)

• modelarts_dataset: ModelArts dataset

name

String

Definition: Name of the data input channel.

Range: N/A

description

String

Definition: Description of the data input channel.

Range: N/A

local_dir

String

Definition: Local path of the container to which the data input channels are mapped. Example: /home/ma-user/modelarts/inputs/data_url_0

Range: N/A

access_method

String

Definition: Access method of the input data channel path (local_dir).

Range:

• parameter: hyperparameters

• env: environment variables