Managing Transcription Models

• Published on Sep 17, 2024
• 7 minute(s) read

The features available in the Model Library ( Library > Models) allow you to manage your finetuning models. From the Model Library, you can:

• deploy a more performant version of a model,
• check the progress of a finetuning model’s training,
• export your models from your development environment so you can import them to production,
• and more.

Viewing a list of transcription models

In the application, transcription models are collections of fine-tuning models. To view a list of transcription models, go to Library > Models, and then select Transcription Models in the drop-down list at the top of the page.

The Transcription Models table lists the sets of finetuning models that you can manage.

For each transcription model, the table provides the following information:

• The name of the transcription model
• The number of flows using the model. Clicking this number reveals a list of the flows.

  Each flow can have only one set of finetuning models assigned to it.

• The types of finetuning models that have at least one QA record available to them (e.g., Latin Semi-structured, Checkbox, Signature)

Currently, it is not possible to create new sets of finetuning models in the application. If the transcription models listed in the Model Library do not meet your needs, contact your Hyperscience representative.

Enabling or disabling daily training

Daily auto-training of fine-tuning models is enabled by default. If it is enabled, the training occurs for all finetuning models in the early morning, UTC time.

If you are not obtaining high-quality results from Transcription Supervision and QA tasks, you may want to disable daily training to prevent these results from affecting your models’ performance. Likewise, after the quality of the results increases, you may choose to re-enable the automatic daily training of your models.

To enable or disable the daily training of your v37 transcription models:

1. Go to Library > Models, and then select Transcription Models in the drop-down list at the top of the page.
2. Select or deselect the Daily autotraining enabled option above the Transcription Models table.

Viewing information about a specific transcription model

1. Go to Library > Models, and then select Transcription Models in the drop-down list at the top of the page.
2. In the Transcription Models table, find the model you want to know more about, and click on its name.

The transcription model details page contains the following information about the model:

• The model's name
• A list of the flows using the model. If several flows use the model, the full list may not appear at the top of the page. To view all the flows in the list, hover over the partial list.
• The kinds of finetuning models included in the model, are shown as tabs at the top of the page. The tabs for text types that aren’t included in the model appear in light gray.

Clicking on one of these tabs reveals the following additional details:

• The status of the model, along with the total number of QA records available to train the model.
  • If the model hasn't been trained yet, this section shows how many more QA records are needed to train the model. The progress bar is yellow and increases in length with each QA record that becomes available for the model’s training.
  • If the model has been trained, the progress bar is green and completely full.
• The available versions of the model. A new version is added to the table each time the model’s training is completed. For each version, the Available Models table contains the following information:
  • The name of the version
    • The name includes the timestamp of when the model’s training was completed in UTC time.
  • The date the version was trained
  • Whether the version is live or archived
  • Links to deploy or download the model, along with an option to deactivate the model if it is currently live

Downloading models

You can download all finetuning models in a transcription model, or you can download individual finetuning models.

Download all finetuning models in a transcription model

To download all finetuning models in a transcription model:

1. Go to Library > Models, and then select Transcription Models in the drop-down list at the top of the page.
2. In the Transcription Models table, click on the name of the transcription model whose finetuning models you want to download.
3. Click Download all models.

The models are downloaded as a ZIP file with the name c onfidence_thresholding_model_export.zip.

Download an individual finetuning model

To download a specific finetuning model:

1. Go to Library > Models, and then select Transcription Models in the drop-down list at the top of the page.
2. In the Transcription Models table, click on the name of the transcription model that contains the finetuning model you want to download.
3. Click on the tab that matches the type of finetuning model you want to download.
4. In the Available Models table, find the version of the model you want to download, and click Download Model ( .

The model is downloaded as a ZIP file with the name c onfidence_thresholding_model_export.zip.

Importing models

If you’ve trained models in a development environment, you can import them to your production environment without having to retrain them after import.

Before you import models to an environment, make sure that the models’ flows and releases are available in the new environment. If you try to import models before importing their flows and releases, the model-import process will end with a series of errors.

• To learn more about importing releases, see Upload an existing release.
• More information about importing flows can be found in Import a flow.

To import a model:

1. Go to Library > Models, and then select Transcription Models in the drop-down list at the top of the page.
2. In the Transcription Models table, find the transcription model you want to import a finetuning model to and click on its name.
3. Click Upload model.

• You do not need to click the tab for the type of finetuning model you want to import. The system automatically detects the type of model during the import process.

Importing models from other versions

Note that you can only import models created in v37; you cannot import models from older versions. If you have a model from a previous version and want to deploy it in v37, you first need to upgrade the model to v37 and retrain it with the v37 trainer.

Training individual finetuning models

If you have enough QA records to train a particular finetuning model, you can start training for it at any time, regardless of whether daily auto training is enabled.

To manually start the training of a finetuning model:

1. Go to Library > Models, and then select Transcription Models in the drop-down list at the top of the page.
2. In the Transcription Models table, click on the name of the transcription model that contains the finetuning model you want to train.
3. Click on the tab that matches the type of finetuning model you want to train.
4. Click Run Training.

The system begins training the model. After training is completed, a new version of the model is added to the Available Models table.

Deploying a different version of a finetuning model

If you see a loss of automation in one of your finetuning models, you may want to deploy an older version of it, if one is available, while you investigate the cause of the automation loss.

Only one finetuning model can be deployed for each text type (e.g., Latin Semi-structured, Checkbox, Signature).

To deploy a different version of a finetuning model:

1. Go to Library > Models, and then select Transcription Models in the drop-down list at the top of the page.
2. In the Transcription Models table, click on the name of the transcription model that contains the finetuning model you want to deploy a different version of.
3. Click on the tab that matches the type of finetuning model you want to deploy a different version of.
4. In the Available Models table, find the version of the model you want to deploy, and click Deploy Model ( .

The model that was previously deployed is now automatically archived, and the version you deployed is now live.

Deactivate a finetuning model

You can deactivate a finetuning model without deploying another. If you do so, the system does not perform finetuning on the type of text the model is trained on.

To deactivate a finetuning model:

1. Go to Library > Models, and then select Transcription Models in the drop-down list at the top of the page.
2. In the Transcription Models table, click on the name of the transcription model that contains the finetuning model you want to deploy a different version of.
3. Click on the tab that matches the type of finetuning model you want to deploy a different version of.
4. In the Available Models table, find the live model, click on the three dots in its row, and then click Deactivate.

The model that was previously deployed is now automatically archived.