View system test results

When your training is successful, review system tests to analyze your training results. If you're satisfied with the training results, place a deployment request for the trained model.

System test results page

Select a project, then select the models tab of that project, locate the model you want to use and finally select the test tab.

The test tab shows you:

  1. System Test Results: The result of the test process in the trainings. The test process produces the BLEU score.

    Sentence Count: How many parallel sentences were used in the test set.

    BLEU Score: BLEU score generated for a model after training completion.

    Status: Indicates if the test process is complete or in progress.

    System test results

  2. Select the System test results, and that will take you to test result details page. This page shows the machine translation of sentences that were part of the test dataset.

  3. The table on the test result details page has two columns - one for each language in the pair. The column for the source language shows the sentence to be translated. The column for the target language contains two sentences in each row.

    Ref: This sentence is the reference translation of the source sentence as given in the test dataset.

    MT: This sentence is the automatic translation of the source sentence done by the model built after the training was conducted.

    System test results compare

Download test

Select the Download Translations link to download a zip file. The zip contains the machine translations of source sentences in the test data set.

Download test

This downloaded zip archive contains three files.

  1. custom.mt.txt: This file contains machine translations of source language sentences in the target language done by the model trained with user's data.

  2. ref.txt: This file contains user provided translations of source language sentences in the target language.

  3. source.txt: This file contains sentences in the source language.

    Downloaded system test results

Deploy a model

To request a deployment:

  1. Select a project, go to Models tab.

  2. For a successfully trained model, it shows "Deploy" button, if not deployed.

    Screenshot that highlights the Deploy button for deploying a model.

  3. Select Deploy.

  4. Select Deployed for the region(s) where you want your model to be deployed, and select Save. You can select Deployed for multiple regions.

    Screenshot that shows where you can deploy or undeploy a model.

  5. You can view the status of your model in the "Status" column.

Note

Custom Translator supports 10 deployed models within a workspace at any point in time.

Update deployment settings

To update deployment settings:

  1. Select a project, and go to the Models tab.

  2. For a successfully deployed model, it shows an Update button.

    Screenshot that highlights the Update button for updating deployment settings.

  3. Select Update.

  4. Select Deployed or Undeployed for the region(s) where you want your model deployed or undeployed, then select Save.

    Deploy model

Note

If you select Undeployed for all regions, the model is undeployed from all regions, and put into an undeployed state. It's now unavailable for use.

Next steps