Training Queue

The Training Queue lets you pre-configure training preparations that will automatically merge with models when they are processed by the ETL pipeline. Use it to set up annotation splits and training configurations before training runs begin.

Page Overview

The page displays:

• Statistics cards - Total preparations, pending, completed, and failed counts
• Status filter - Filter by status (All, Pending, Processing, Completed, Failed)
• Queue table - All preparations with tag, model ID, status, train/test splits, and actions

Understanding Statuses

Status	Meaning
Pending	Waiting to be processed. Will auto-merge when a matching model is processed by ETL.
Processing	Currently being processed by the ETL pipeline.
Completed	Successfully merged to a model. Cannot be edited.
Failed	Processing failed. Can be reset to pending to retry.

Creating a Preparation

Click New Preparation to create a training preparation.

Field	Description
Tag	Select the tag this preparation is for (required)
Model ID	Optional. If provided, data will auto-merge when this model is processed by ETL.
Annotation Process Split	Configure which annotation processes go to training vs testing
Training Configuration	JSON configuration for training parameters (optional)
Notes	Additional notes about this preparation

Annotation Process Split

The annotation process split determines which data goes into training and which goes into testing.

Features:

• Training List - Annotation processes used for training the model
• Test List - Annotation processes used for evaluating the model
• Add - Add a single process from the tag's available processes
• Bulk Add - Add multiple processes at once
• Edit - Modify process settings (sample ratio, etc.)
• Remove - Remove a process from the split

Training Configuration

View or edit the training configuration by clicking View Config in the actions menu.

The configuration dialog provides:

• Structured View - Organized display of configuration parameters
• JSON View - Raw JSON editor for advanced editing
• Copy - Copy configuration to clipboard
• Download - Download as JSON file
• Save - Save changes (enabled when modifications are made)

Available Actions

Click the three-dot menu on any row to access actions:

Action	Description
Edit	Modify the preparation (disabled for completed items)
View Config	View or edit training configuration
Merge to Model	Manually merge to a specific model ID
Duplicate	Create a copy of the preparation
Change Status	Change status (e.g., reset failed to pending)
Delete	Delete the preparation

Merging to a Model

Normally, preparations auto-merge when a model with a matching Model ID is processed. To manually merge, use the Merge to Model action.

Enter the Model ID you want to merge to. The annotation splits from the preparation will be copied to that model.

Changing Status

Use Change Status to manually transition between statuses.

Available transitions:

• Pending → Failed - Mark as failed with an optional reason
• Processing → Failed - Mark as failed if stuck
• Failed → Pending - Reset to pending to retry processing

Note: Completed items cannot have their status changed.

Typical Workflow

1. Create a preparation for a tag with the desired annotation split
2. Optionally specify a Model ID if you know which model will use this configuration
3. Set up training configuration parameters
4. When the ETL processes a matching model, the preparation auto-merges
5. Status changes to "Completed" when successful

Tips

• Use Duplicate to quickly create similar preparations for different models
• Leave Model ID empty if you want to manually merge later
• Use the status filter to focus on pending or failed items
• Check failed items for error messages to diagnose issues