Use this article to configure and manage Transcription models throughout their lifecycle.
Before you begin
To understand how Transcription models use confidence, target accuracy, and QA data to balance automation and human review, see Transcription Models Overview.
View model details and versions
Go to Models > Transcription to view the Transcription models available in your instance. In the Transcription Models table, select a model to open its details page.
The details page shows:
The model name.
The flows using the model. If the complete list is not visible, hover over the partial list.
The model’s supported text types appear as tabs. Tabs for text types that are not included appear in light gray.
Select a model tab to view:
The model status and the total number of eligible QA records.
Before the model is trained, a yellow progress bar shows how many more QA records are required.
After the model is trained, the progress bar is green and full.
Available model versions.
A new version is added after each completed training run.
The Available Models table shows the version name and UTC training timestamp, the training date, whether the version is live or archived, and the available download, deployment, or deactivation actions.
A Transcription model is configured at the flow level
This determines how extracted values are evaluated and handled within that flow.
Assign a transcription model to a flow
To use a transcription model, assign it to your Document Processing flow.
Export your flow by following the steps in the interactive walkthrough below:
Open the Document Processing Flow on your machine.
Find the
transcription_modelproperty in the JSON file.
Set the name of your model and save the changes on your machine.
Import the JSON file into the system and find your Transcription model in Models > Transcription. See the interactive walkthrough below for guidance:
Train an individual model
When enough eligible QA records are available, you can start training manually regardless of whether daily autotraining is enabled.
Go to Models > Transcription.
Select the Transcription model you want to train.
Select the tab for the applicable text type.
Select Run Training.
After training finishes, a new version appears in the Available Models table.
Monitor model behavior
Monitor both model-level projections and production outcomes. Neither one is sufficient by itself:
Use the
ProjectedTranscriptionAutomationCSV file in the Keyer Projection Report to review when the model was last trained, projected automation at different target accuracies, and the projection’s margin of error.Use Reporting > Automation to monitor actual Field Transcription Automation and Table Transcription Automation. Filter by date range, flow, and layout variation to isolate where a change occurred. To learn more, see Automation.
Use the
HourlyReportingTaskOverviewdata in the Keyer Projection Report to monitor the number of Transcription tasks entering and remaining in the work queue. A decrease in automation can create additional manual work and processing delays even when model accuracy has not decreased.
Deploy a different model version
If a model version causes an unexpected loss of automation, you can deploy an older compatible version while you investigate. Only one model version can be live for each text type, such as Latin Semi-structured, Checkbox, or Signature.
Go to Models > Transcription.
Select the Transcription model that contains the model version you want to deploy.
Select the tab for the applicable text type.
In the Available Models table, find the version and select Deploy Model (
).
The newly deployed version becomes live, and the previously deployed version is archived automatically.
Deactivate a model
You can deactivate a model without deploying another version. After deactivation, no model version is active for that text type.
Go to Models > Transcription.
Select the Transcription model.
Select the tab for the applicable text type.
In the Available Models table, open the menu for the live version and select Deactivate.
The previously deployed version is archived automatically.
Download Transcription models
You can download the complete Transcription model package or an individual model version.
Download all models
Go to Models > Transcription.
In the Transcription Models table, select the model you want to download.
Select Download all models.
The models are downloaded in a ZIP file named confidence_thresholding_model_export.zip.
Download an individual model version
Go to Models > Transcription.
In the Transcription Models table, select the model that contains the version you want to download.
Select the tab for the applicable text type.
In the Available Models table, find the version and select Download Model (
).
The model is downloaded in a ZIP file named confidence_thresholding_model_export.zip.
Import Transcription models
You can import models trained in a development environment into a production environment without retraining them after import.
Before importing a model
Make sure the model's flows and releases are available in the destination environment. Importing a model before its flows and releases produces errors. See Upload an existing release and Import a flow.
Go to Models > Transcription.
In the Transcription Models table, select the destination Transcription model.
Select Upload model. You do not need to select a text type first; the system detects it during import.
Drag the model ZIP file into the Import dialog, or select Choose Files and open the file.
Select Import.
The imported model appears under the tab for its text type.
Import models from other versions
A Transcription model trained on a different version can be imported successfully, but it does not appear in the model list unless its version matches the current instance. See Compatibility Across Application, Flow, and Model Versions.