Help Center/ ModelArts/ ModelArts User Guide (Standard)/ Using ModelArts Standard to Train Models/ Creating a Production Training Job

Updated on 2025-08-18 GMT+08:00

View PDF

Creating a Production Training Job

Developing models involves optimizing their performance effectively. Traditional methods require repeatedly testing various model designs, datasets, and hyperparameters, which takes significant time and effort but may still fail to deliver good results. ModelArts simplifies this process by offering tools for creating training jobs, tracking progress in real time, and managing versions. With ModelArts, users can test different configurations easily and identify the best-performing setup faster.

Create a production training job in either of the following ways:

Use the ModelArts Standard console. For details, see the following sections.
Use the ModelArts API to create a production training job. For details, see Using PyTorch to Create a Training Job (New-Version Training).

Notes and Constraints

By default, up to 10,000 training jobs can be created. You can view the remaining quota on the training job list page.

Figure 1 Viewing the remaining quota of a training job

Prerequisites

Data for training uploaded to an OBS directory.
At least one empty folder in OBS for storing training output.

ModelArts does not support encrypted OBS buckets. When creating an OBS bucket, do not enable bucket encryption.
Account not in arrears (paid resources required for training jobs).
OBS directory and ModelArts in the same region.
Access authorization configured. If it is not configured, configure it by referring to Configuring Agency Authorization for ModelArts with One Click.
Training algorithm. For details, see Creating an Algorithm.

Billing

Model training in ModelArts uses compute and storage resources, which are billed. Compute resources are billed for running training jobs. Storage resources are billed for storing data in OBS or SFS. For details, see Model Training Billing Items.

Procedure

To create a training job, follow these steps:

Follow the steps in Accessing the Page for Creating a Training Job.
Follow the steps in Configuring Basic Information.
Select an algorithm type.
- Use an existing algorithm to create a training job by referring to Choosing an Algorithm Type (My Algorithm).
- Use a preset image to create a training job by referring to Choosing an Algorithm Type (Custom Algorithm).
- Use a custom image to create a training job by referring to Choosing a Boot Mode (Custom Image).
Configure training parameters, including the input, output, hyperparameters, and environment variables. For details, see Configuring Training Parameters.
Select a resource pool as needed. A dedicated resource pool is recommended for optimal performance. For details about the differences between dedicated and public resource pools, see Differences Between Dedicated Resource Pools and Public Resource Pools.
- Configuring a Public Resource Pool
- Configuring a Dedicated Resource Pool
Select a training mode when you use a preset MindSpore engine and Ascend resources. For details, see (Optional) Selecting a Training Mode.
Add tags if you want to manage training jobs by group. For details, see (Optional) Adding Tags.
Perform follow-up procedure. For details, see Follow-Up Operations.

Accessing the Page for Creating a Training Job

Log in to the ModelArts console.
In the navigation pane, choose Model Training > Training Jobs.
Click Create Training Job and click Old Version in the upper right corner. The old version of the Create Training Job page is displayed.

Configuring Basic Information

On the Create Training Job page, configure parameters.

**Table 1** Basic information
Parameter	Description
Name	Job name, which is mandatory. The system automatically generates a name, which you can then rename according to the following rules. The name contains 1 to 64 characters. Letters, digits, hyphens (-), and underscores (_) are allowed.
Description	Job description, which helps you learn about the job information in the training job list.
Experiment	Experiment for classifying and managing the job. If you select Create new, enter the experiment name and description. If you select Use existing, select an experiment name. If you select Not required, this job will not be managed in any experiment.

Choosing an Algorithm Type (My Algorithm)

Set Algorithm Type to My algorithm and select an algorithm from the algorithm list. If no algorithm meets the requirements, you can create an algorithm. For details, see Creating an Algorithm.

Choosing an Algorithm Type (Custom Algorithm)

If an algorithm is available in algorithm management, choose My algorithm. If no algorithm is available, choose Custom algorithm. If you use a custom algorithm to create a training job, select a boot mode by referring to Table 2.

**Table 2** Creating a training job using a custom algorithm
Parameter	Description
Algorithm Type	Select Custom algorithm. This parameter is mandatory.
Boot Mode	Select Preset image and select the preset image engine and engine version to be used by the training job. If you select Customize for the engine version, select a custom image from Image.
Image	This parameter is displayed and mandatory only when the preset image version is set to Customize. You can set the container image path in either of the following ways: To select your image or an image shared by others, click Select on the right and select a container image for training. The required image must be uploaded to SWR beforehand. To select a public image, enter the address of the public image in SWR. Enter the image path in the format of "Organization name/Image name:Version name". Do not contain the domain name in the path because the system will automatically add the domain name to the path.
Code Source	Select a training code source. OBS: Select OBS if the training code is stored in an OBS bucket. SFS: Select SFS if the training code is stored in an SFS file system.
Code Directory	This parameter is available only when Code Source is set to OBS. Select the OBS directory where the training code file is stored. This parameter is mandatory. Upload code to the OBS bucket beforehand. The total size of files in the directory cannot exceed 5 GB, the number of files cannot exceed 1000, and the folder depth cannot exceed 32. The training code file is automatically downloaded to the ${MA_JOB_DIR}/demo-code directory of the training container when the training job is started. demo-code is the last-level OBS directory for storing the code. For example, if Code Directory is set to /test/code, the training code file is downloaded to the ${MA_JOB_DIR}/code directory of the training container.
Boot File	Select the Python boot script of the training job in the code directory. This parameter is mandatory. ModelArts supports only the boot file written in Python. Therefore, the boot file must end with .py.
Local Code Directory	This parameter is available only when Code Source is set to OBS. Specify the local directory of a training container. When a training starts, the system automatically downloads the code directory to this directory. The default local code directory is /home/ma-user/modelarts/user-job-dir. This parameter is optional.
Work Directory	During training, the system automatically runs the cd command to execute the boot file in this directory.

Selecting a preset image with customization results in the same background behavior as running a training job directly with that image. For example:

The system automatically injects environment variables.

PATH=${MA_HOME}/anaconda/bin:${PATH}
LD_LIBRARY_PATH=${MA_HOME}/anaconda/lib:${LD_LIBRARY_PATH}
PYTHONPATH=${MA_JOB_DIR}:${PYTHONPATH}

The selected boot file will be automatically started using Python commands. Ensure that the Python environment is correct. The PATH environment variable is automatically injected. Run the following commands to check the Python version for the training job:
```
export MA_HOME=/home/ma-user; docker run --rm {image} ${MA_HOME}/anaconda/bin/python -V
docker run --rm {image} $(which python) -V
```
The system automatically adds hyperparameters associated with the preset image.

Choosing a Boot Mode (Custom Image)

If you use a custom image to create a training job, select a boot mode by referring to Table 3.

**Table 3** Creating a training job using a custom image
Parameter	Description
Algorithm Type	Select Custom algorithm. This parameter is mandatory.
Boot Mode	Select Custom image. This parameter is mandatory.
Image	Container image path. This parameter is mandatory. You can set the container image path in either of the following ways: To select your image or an image shared by others, click Select on the right and select a container image for training. The required image must be uploaded to SWR beforehand. To select a public image, enter the address of the public image in SWR. Enter the image path in the format of "Organization name/Image name:Version name". Do not contain the domain name in the path because the system will automatically add the domain name to the path.
Code Directory	OBS directory where the training code file is stored. Configure this parameter only if your custom image does not contain training code. Upload code to the OBS bucket beforehand. The total size of files in the directory cannot exceed 5 GB, the number of files cannot exceed 1000, and the folder depth cannot exceed 32. The training code file is automatically downloaded to the ${MA_JOB_DIR}/demo-code directory of the training container when the training job is started. demo-code is the last-level OBS directory for storing the code. For example, if Code Directory is set to /test/code, the training code file is downloaded to the ${MA_JOB_DIR}/code directory of the training container.
User ID	User ID for running the container. The default value 1000 is recommended. If the UID needs to be specified, its value must be within the specified range. The UID ranges of different resource pools are as follows: Public resource pool: 1000 to 65535 Dedicated resource pool: 0 to 65535 If the user ID is set to 0, the user in the training container is root.
Boot Command	Command for booting an image. This parameter is mandatory. When a training job is running, the boot command is automatically executed after the code directory is downloaded. If the training boot script is a .py file, train.py for example, the boot command is as follows. python ${MA_JOB_DIR}/demo-code/train.py If the training boot script is a .sh file, main.sh for example, the boot command is as follows: bash ${MA_JOB_DIR}/demo-code/main.sh You can use semicolons (;) and ampersands (&&) to combine multiple commands. demo-code in the command is the last-level OBS directory where the code is stored. Replace it with the actual one. If there are input pipes, output pipes, or hyperparameters, ensure that the last command of the boot command runs the training script. Reason: The system appends input pipes, output pipes, and hyperparameters to the end of the boot command. If the last command is not the training script, an error will occur. Example: If the last line of the boot command is python train.py and the --data_url hyperparameter exists, the system executes python train.py --data_url=/input when running properly. However, if the boot command ends with another command, such as: python train.py pwd # The last command is pwd instead of the training script. The system will execute python train.py pwd --data_url=/input, leading to an error. NOTE: To ensure data security, do not enter sensitive information, such as plaintext passwords.
Local Code Directory	This parameter is available only when Code Source is set to OBS. Specify the local directory of a training container. When a training starts, the system automatically downloads the code directory to this directory. The default local code directory is /home/ma-user/modelarts/user-job-dir. This parameter is optional.
Work Directory	During training, the system automatically runs the cd command to execute the boot file in this directory.

For details about how to use custom images supported by training, see Boot Command Specifications for Custom Images.

Configuring Training Parameters

Data is obtained from an OBS bucket or dataset for model training. The training output can also be stored in an OBS bucket. When creating a training job, you can configure parameters such as input, output, hyperparameters, and environment variables by referring to Table 4.

The input, output, and hyperparameter parameters of a training job vary depending on the algorithm type selected during training job creation. If a parameter value is dimmed, the parameter has been configured in the algorithm code and cannot be modified.

**Table 4** Configuring training parameters
Parameter	Sub-Parameter	Description
Input	Parameter name	The algorithm code reads the training input data based on the input parameter name. The recommended value is data_url. The training input parameters must match the input parameters of the selected algorithm. For details, see Table 4.
	Dataset	Click Dataset and select the target dataset and its version in the ModelArts dataset list. When the training job is started, ModelArts automatically downloads the data in the input path to the training container.
	Data path	Click Data path and select the storage path to the training input data from an OBS bucket. Files must not exceed 10 GB in total size, 1,000 in number, or 1 GB per file. When the training job is started, ModelArts automatically downloads the data in the input path to the training container.
	Obtained from	The following uses training input data_path as an example. If you select Hyperparameters, use this code to obtain the data: import argparse parser = argparse.ArgumentParser() parser.add_argument('--data_path') args, unknown = parser.parse_known_args() data_path = args.data_path If you select Environment variables, use this code to obtain the data: import os data_path = os.getenv("data_path", "")
Output	Parameter name	The algorithm code reads the training output data based on the output parameter name. The recommended value is train_url. The training output parameters must match the output parameters of the selected algorithm. For details, see Table 5.
	Data path	Click Data path and select the storage path for the training output data from an OBS bucket. Files must not exceed 1 GB in total size, 128 in number, or 128 MB per file. During training, the system automatically synchronizes files from the local code directory of the training container to the data path. NOTE: The data path can only be an OBS path. To prevent any issues with data storage, choose an empty directory as the data path.
	Obtained from	The following uses the training output train_url as an example. If you select Hyperparameters, use this code to obtain the data: import argparse parser = argparse.ArgumentParser() parser.add_argument('--train_url') args, unknown = parser.parse_known_args() train_url = args.train_url If you select Environment variables, use this code to obtain the data: import os train_url = os.getenv("train_url", "")
	Predownload	Indicates whether to pre-download the files in the output directory to a local directory. If you set Predownload to No, the system does not download the files in the training output data path to a local directory of the training container when the training job is started. If you set Predownload to Yes, the system automatically downloads the files in the training output data path to a local directory of the training container when the training job is started. The larger the file size, the longer the download time. To avoid excessive training time, remove any unneeded files from the local code directory of the training container as soon as possible. To use Resumable Training, select Yes.
Hyperparameter	N/A	Used for training tuning. This parameter is determined by the selected algorithm. If hyperparameters have been defined in the algorithm, all hyperparameters in the algorithm are displayed. Hyperparameters can be modified and deleted. The status depends on the hyperparameter constraint settings in the algorithm. For details, see Table 6. To import hyperparameters in batches, click Upload. You will need to fill in the hyperparameters based on the provided template. The total number of hyperparameters should not exceed 100, or the import will fail. NOTE: To ensure data security, do not enter sensitive information, such as plaintext passwords.
Environment Variable	N/A	Add environment variables based on service requirements. For details about the environment variables preset in the training container, see Managing Environment Variables of a Training Container. To import environment variables in batches, click Upload. You will need to fill in the environment variables based on the provided template. The total number of environment variables should not exceed 100, or the import will fail. NOTE: To ensure data security, do not enter sensitive information, such as plaintext passwords.
Auto Restart	N/A	Once this feature is enabled, you can set the number of restarts and whether to enable Unconditional auto restart. After you enable auto restart, ModelArts will handle any exceptions caused by environmental issues during a training job. It will either automatically handle the exception or isolate the faulty node and then restart the job, which helps to increase the success rate of the training. To avoid losing training progress and make full use of compute, ensure that your code logic supports resumable training before enabling this function. For details, see Resumable Training. The value ranges from 1 to 128. The default value is 3. The value cannot be changed once the training job is created. Set this parameter based on your needs. If Unconditional auto restart is selected, the training job will be restarted unconditionally once the system detects a training exception. To prevent invalid restarts, it supports a maximum of three consecutive unconditional restarts. ModelArts continuously monitors job processes to detect suspension and optimize resource usage. When Restart Upon Suspension is enabled, suspended jobs can be automatically restarted at the process level. However, ModelArts does not verify code logic, and suspension detection is periodic, which may result in false reports. By enabling this feature, you acknowledge the possibility of false positives. To prevent unnecessary restarts, ModelArts limits consecutive restarts to three. If auto restart is triggered during training, the system records the restart information. You can check the fault recovery details on the training job details page. For details, see Training Job Fault Tolerance Check.

Configuring a Public Resource Pool

To configure a public resource pool, refer to Table 5.

**Table 5** Configuring a public resource pool for a training job
Parameter	Description
Resource Pool	Select Public resource pool.
Resource Type	Select the resource type required for training. This parameter is mandatory. If a resource type has been defined in the training code, select a proper resource type based on algorithm constraints. For example, if the resource type defined in the training code is CPU and you select other types, the training fails. If some resource types are invisible or unavailable for selection, they are not supported.
Specifications	Select the required resource specifications based on the resource type. If Data path is selected for Input, you can click Check Input Size on the right to ensure the storage is larger than the input data size. NOTICE: The resource flavor GPU:n*tnt004 (n indicates a specific number) does not support multi-process training.
Compute Nodes	Select the number of instances as required. The default value is 1. If only one instance is used, a single-node training job is created. ModelArts starts one training container on this node. The training container exclusively uses the compute resources of the selected flavor. If more than one instance is used, a distributed training job is created. For more information about distributed training configurations, see Overview. Before creating distributed training jobs, pre-install all required pip dependencies (see Installing pip Dependencies in an Image). If there are more than 10 nodes, the system automatically deletes the pip source configuration. Executing pip install commands during training may cause training failures.
Persistent Log Saving	If you select CPU or GPU flavors, Persistent Log Saving is available for you to configure. After this feature is enabled (default), configure Job Log Path. The system permanently stores training logs to the specified OBS path. After this feature is disabled, ModelArts automatically stores the logs for 30 days. You can download all logs on the job details page to a local path.
Job Log Path	When enabling Persistent Log Saving or selecting Ascend resources, select an empty OBS directory for Job Log Path to store log files generated by the training job. Ensure that you have read and write permissions to the selected OBS directory.
Event Notification	Indicates whether to enable event notification. This feature is disabled by default, which means SMN is disabled. After this feature is enabled, you will be notified of specific events, such as job status changes or suspected suspensions, via an SMS or email. Notifications will be billed based on SMN pricing. In this case, you must configure the topic name and events. Topic: topic of event notifications. Click Create Topic to create a topic on the SMN console. Event: events you want to subscribe to. Examples: JobStarted, JobCompleted, JobFailed, JobTerminated, and JobHanged. NOTE: After you create a topic on the SMN console, add a subscription to the topic, and confirm the subscription. Then, you will be notified of events. For details, see Adding a Subscription. SMN charges you for the number of notification messages. For details, see Billing. Only training jobs using GPUs or NPUs support JobHanged events.
Auto Stop	When using paid resources, you can determine whether to enable auto stop. This function is disabled by default, the training job keeps running until the training is completed. If this function is enabled, configure the auto stop time. The value can be 1 hour, 2 hours, 4 hours, 6 hours, or Customize. The customized time must range from 1 hour to 720 hours. When you enable this feature, the training stops automatically when the time limit is reached. The time limit does not count down when the training is paused.
training_ssh_configure_nodes	Whether to enable password-free SSH mutual trust between nodes. This feature is disabled by default. Enabling this feature requires you to configure the SSH key directory. This is where the automatically generated SSH key file will be stored within the training container. The default value is /home/ma-user/.ssh.

Configuring a Dedicated Resource Pool

To configure a dedicated resource pool, refer to Table 6.

**Table 6** Configuring a dedicated resource pool for a training job
Parameter	Description
Resource Pool	Select a dedicated resource pool. If you select a dedicated resource pool, you can view the name, status, node specifications, number of idle/fragmented nodes, number of available/total nodes, and number of cards of the resource pool. Hover over View in the Idle/Fragmented Nodes column to check fragment details and check whether the resource pool meets the training requirements.
Specifications	Select the required resource specifications based on the resource type. If Data path is selected for Input, you can click Check Input Size on the right to ensure the storage is larger than the input data size. NOTE: The resource specifications GPU:n*tnt004 (n indicates a specific number) do not support multi-process training.
Compute Nodes	Select the number of instances as required. The default value is 1. If only one instance is used, a single-node training job is created. ModelArts starts one training container on this node. The training container exclusively uses the compute resources of the selected flavor. If more than one instance is used, a distributed training job is created. For more information about distributed training configurations, see Overview. Before creating distributed training jobs, pre-install all required pip dependencies (see Installing pip Dependencies in an Image). If there are more than 10 nodes, the system automatically deletes the pip source configuration. Executing pip install commands during training may cause training failures.
Job Priority	When using a dedicated resource pool, you can set the priority of the training job. 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. If a training job is in the Pending state for a long time, you can change the job priority to reduce the queuing duration. For details, see Priority of a Training Job.
Preemption	When enabled, jobs that allow preemption may be terminated and re-queued if resource pool capacity is insufficient. To avoid losing training progress, configure resumable training before enabling this function. For details, see Resumable Training.
SFS Turbo	When ModelArts and SFS Turbo are directly connected, multiple SFS Turbo file systems can be mounted to a training job to store training data. Click Add Mount Configuration and set the following parameters: Name: Select an SFS Turbo file system. Mount Path: Enter the SFS Turbo mounting path in the training container. Directory: Specify the SFS Turbo storage location. If you have configured the folder control permission, select a storage location. If you have not configured the folder control permission, retain the default value / or customize a location. Mounting Mode: Permission on the mounted SFS Turbo file system. This parameter is displayed as Read/Write or Read-only based on the permission of the SFS Turbo storage location. If you have not configured the folder control permission, this parameter is unavailable. Mount Options: Configure SFS mount parameters to accelerate and optimize training. For details about the parameters, see Configuring SFS Turbo Mount Options. Alternatively, retain the default settings below: mountOptions: - vers=3 - timeo=600 - nolock - hard NOTE: You can mount a file system multiple times, but each mount path must be distinct. A maximum of five disks can be mounted to a training job. The mounting path cannot be a / directory or a default mounting path, such as /cache and /home/ma-user/modelarts. For details about how to set permissions for SFS Turbo folders, see Permissions Management.
Persistent Log Saving	If you select CPU or GPU flavors, Persistent Log Saving is available for you to configure. After this feature is enabled (default), configure Job Log Path. The system permanently stores training logs to the specified OBS path. After this feature is disabled, ModelArts automatically stores the logs for 30 days. You can download all logs on the job details page to a local path.
Job Log Path	When enabling Persistent Log Saving or selecting Ascend resources, select an empty OBS directory for Job Log Path to store log files generated by the training job. Ensure that you have read and write permissions to the selected OBS directory.
Event Notification	Indicates whether to enable event notification. This feature is disabled by default, which means SMN is disabled. After this feature is enabled, you will be notified of specific events, such as job status changes or suspected suspensions, via an SMS or email. Notifications will be billed based on SMN pricing. In this case, you must configure the topic name and events. Topic: topic of event notifications. Click Create Topic to create a topic on the SMN console. Event: events you want to subscribe to. Examples: JobStarted, JobCompleted, JobFailed, JobTerminated, and JobHanged. NOTE: After you create a topic on the SMN console, add a subscription to the topic, and confirm the subscription. Then, you will be notified of events. For details, see Adding a Subscription. SMN charges you for the number of notification messages. For details, see Billing. Only training jobs using GPUs or NPUs support JobHanged events.
Auto Stop	When using paid resources, you can determine whether to enable auto stop. This feature is disabled by default, the training job keeps running until the training is completed. If this feature is enabled, configure the auto stop time. The value can be 1 hour, 2 hours, 4 hours, 6 hours, or Customize. The customized time must range from 1 hour to 720 hours. When you enable this feature, the training stops automatically when the time limit is reached. The time limit does not count down when the training is paused.
training_ssh_configure_nodes	Whether to enable password-free SSH mutual trust between nodes. This feature is disabled by default. Enabling this feature requires you to configure the SSH key directory. This is where the automatically generated SSH key file will be stored within the training container. The default value is /home/ma-user/.ssh.

(Optional) Selecting a Training Mode

Select a training mode when you use a preset MindSpore engine and Ascend resources. ModelArts provides three training modes for you to select. You can obtain different diagnosis information based on the actual scenario.

Common mode: It is the default training scenario.
High performance mode: In this mode, certain O&M functions will be adjusted or even disabled to accelerate the running speed, but this will deteriorate fault locating. This mode is suitable for stable networks requiring high performance.
Fault diagnosis mode: In this mode, certain O&M functions will be enabled or adjusted to collect more information for locating faults. This mode provides fault diagnosis. You can select a diagnosis type as required.

(Optional) Adding Tags

If you want to manage training jobs by group using tags, select Configure Now for Advanced Configuration to set tags for training jobs. For details about how to use tags, see Using TMS Tags to Manage Resources by Group.

Follow-Up Operations

After parameter setting for creating a training job, click Submit. On the Confirm dialog box, click OK.

A training job runs for a period of time. You can go to the training job list to view the basic information about the training job.

In the training job list, Status of a newly created training job is Pending.
When the status of a training job changes to Completed, the training job is finished, and the generated model is stored in the corresponding output path.
If the status is Failed or Abnormal, click the job name to go to the job details page and view logs for troubleshooting.