ML Studio Interface

This section describes the screens and forms of the ML Studio module, the available actions, and the purpose of each field. Backend configuration is described in SM-ML Settings.

Definitions of the terms algorithm, model, job, runtime environment, and deployment are provided in the module overview.

Location and Structure

To open the module, go to Main Menu - ML Studio. The following screens are available:

Menu Item	Description
Algorithms	Algorithm registry
Deployed Images	Algorithm deployments in runtime environments
Models	Trained models
Running Jobs	History and statuses of training and application operations

Algorithms Screen

This screen displays a list of registered algorithms. The heading includes the number of records, for example, Algorithms (5).

Toolbar above the table:

• Search - filters the list by name
• Create - opens the algorithm creation form
• Import - uploads algorithms from a JSON file
• Export - saves the selected algorithms to a JSON file. For details, see Import and Export. The button appears after you select rows using the checkboxes

Table columns:

Column	Description
Name	Algorithm name and a link to its details page
Image Used	Base container image, for example, sm-ml-runtime:6.1.0-amd64
Deployed Image	Name of the associated deployment, for example, Logistic Regression:local
Deployed Image Status	Deployment state, for example, Deployed

Row actions (⋮ menu on the right):

• Edit - edits the algorithm
• Delete - deletes the algorithm

To select records, select the checkboxes to the left of the corresponding rows. To select all records, select the checkbox in the table header. After you select at least one record, the Export button appears next to Import. For details, see Import and Export.

Create New Algorithm Form

Click Create to open the dialog.

Field	Type	Description
Name	Text	Human-readable algorithm name
Image	Drop-down list	Container image in which the algorithm will run. The list contains images available in the runtime environment, for example, sm-ml-runtime:*

Buttons: Cancel, Save.

Import and Export

Use JSON export and import to transfer algorithms between Search Anywhere Framework instances.

Export. Select the required algorithms using the checkboxes, and click Export. The selected algorithms are saved to a JSON file.

Import. Click Import and upload a previously exported file in the Select JSON to Import area. After you select a file, the Import button becomes available. To close the dialog without importing, click Close.

Algorithm Details Page

Click an algorithm name to open its details page. Click ← Back to return to the list.

The page contains the following elements:

• Image Used - the current base image with an edit icon (✎)
• Deployed Image - a panel with the status (Deployed) and the ⋮ actions menu:
  • deployment Name, for example, Binary Neural Network Classifier:local
  • Environment, for example, local
• Tabs:
  • Models - models trained using this algorithm. The columns are described in the Models section
  • Running Jobs - training and application jobs for this algorithm

The ⋮ menu contains the same actions as the row menu on the Deployed Images screen: open Jupyter, MLflow, or TensorBoard; synchronize the algorithm; update the status; or delete the deployment.

Deployed Images Screen

This screen displays a list of deployments, which are algorithm instances running in runtime environments. The heading includes the number of records, for example, Deployed Images (7).

Toolbar:

• Search - filters by name
• Algorithm - filters by algorithm
• Environment - filters by runtime environment
• Create - opens the deployment creation form

Table columns:

Column	Description
Name	Deployment name, for example, TiDE:local. Click the name to open the update form
Algorithm	Algorithm associated with the deployment
Environment	Runtime environment, for example, local
Status	State, for example, Deployed

Row actions (⋮ menu):

• Go to Jupyter - opens the deployment's Jupyter environment
• Go to MLflow - opens MLflow
• Go to TensorBoard - opens TensorBoard
• Synchronize Algorithm - synchronizes the algorithm
• Update Status - retrieves the current deployment state
• Delete - deletes the deployment

note

An action is disabled if the corresponding service is unavailable for the deployment. In the screenshot, MLflow and TensorBoard are unavailable.

Create Deployment Form

Click Create to open the dialog.

Field	Type	Description
Algorithm	Drop-down list	Algorithm to deploy
Environment	Drop-down list	Runtime environment to which the image is deployed
CPU Count	Number	Number of allocated CPU cores, for example, 2
Memory, MB	Number	Memory limit in megabytes, for example, 4096
GPU Request	Value	Number of requested GPUs or all for all available GPUs

Buttons: Cancel, Save.

Update Deployment Form

Click a deployment name to open the form. It contains the same fields as the creation form. The Algorithm and Environment fields cannot be changed. You can change the CPU count, memory size, GPU parameters, and model storage.

Models Screen

This screen displays the latest ready versions of trained models. The heading includes the number of records, for example, Models (19).

Toolbar:

• Search - filters by name or ID
• Algorithm - filters by algorithm

Table columns:

Column	Description
ID	Unique model version identifier, for example, binary-nn-smoke-direct-001138
Name	Model name
Algorithm	Algorithm used to train the model
Status	State, for example, Ready
Artifact URL	Link to the model artifact, if available
Last Running Job ID	Job that created or updated the model (job_…)
Last Updated	Date and time when the model information was last updated

Row actions (⋮ menu):

• Delete - deletes the model

Running Jobs Screen

This screen displays the history of training and application jobs. The heading includes the number of records, for example, Running Jobs (48).

Toolbar:

• Search - filters by ID or name
• Model Name - filters by model

Table columns:

Column	Description
ID	Job identifier (job_…)
Model Name	Model on which the job ran
Status	Job state, for example, Error or Ready
Operation	Operation type: fit (training) or apply (application)
Runtime URL	Address of the environment where the job ran, for example, http://…:30011
Created At	Date and time when the job was created
Last Updated	Date and time when the status was last changed

Row actions (⋮ menu):

• Delete - deletes the job record

Page navigation is available at the bottom of the table.

Common Interface Elements

• Search and filters are available in all lists. Results are immediately displayed in the table
• Pagination - the bottom of each table contains a Rows per Page field (default: 15) and page navigation controls
• Row menu ⋮ - contextual actions for a specific record
• Statuses are displayed as colored labels: green indicates a successful or ready state, and red indicates an error
• The heading of each list includes the number of records in parentheses

Related Resources

For a step-by-step workflow using these screens, see the ML Studio Step-by-Step Guide.