Custom AI

MT Engine Model Training

To train an engine, follow these steps:

  1. From the MT models page, click Train MT model.

    The New MT model window opens.

  2. Provide a name for the model.

  3. Select a Dataset from the dropdown list.

    The source and target languages are presented along with the total number of segments.

    In order to be usable for engine training, a dataset requires a minimum of 5,000 segments but ideally, more than 10,000.

  4. Click Create and train.

    The name of the model is added to the list along with the state. Training can take several hours depending on number of segments.

The More menu ellipses.png can also be used to open the Overview.

In the case a training fails and has an error status, click Retry Retryfrom the More menu to make another attempt.

Using a Model

Once an engine is trained, the model can be used in Phrase Language AI.

To use the model in Phrase Language AI, follow these steps:

  1. Select Deploy to Language AI (TMS) from the ellipses menu ellipses.png of any model in the list or click Deploy to Language AI (TMS) on the model details page.

    The Deploy to Language AI (TMS) window opens.

  2. Select any specific locale combination belonging to the model language pair from the Source language and Target language dropdown lists.

  3. Select which MT profiles the model should be deployed to from the dropdown list.

    Click the x beside the profile name to remove it.

  4. Click Deploy.

    The model will be marked both in the model list and the details page as being in a state of Deployed along with how many profiles it is deployed to. Hover over the number of deployments to see the profile names.

The model is now visible in Phrase NextMT configuration.

To use the custom model for translation, create a dedicated MT profile using only the customer model and then apply that model in the machine translation settings of a project.

If deployed models are removed or undeployed, quota is freed up. Assuming the limit hasn’t changed, freed quota can then be reused to deploy new models.

To move an already-deployed model to a different MT profile, or to free its deployment quota for reuse, remove the model from its current MT profile first. This control is not available on the Custom AI MT Models page. It is located inside the Phrase Language AI MT profile that currently holds the model.

  1. On the Custom AI MT Models page, copy the Custom Model ID of the model to remove.

  2. In Phrase Language AI, open the MT profile that currently holds the model.

  3. Click Edit next to Phrase Custom NextMT.

  4. Locate the model in the list using the Custom Model ID copied earlier.

  5. Click the trash icon next to the model, then save the profile.

After removal, the Custom AI MT Models page may take a few minutes to sync the updated status. Once synced, the model status changes to Ready to deploy, and the Deploy to Language AI (TMS) button becomes enabled again. The model can then be deployed to a different MT profile without deleting or retraining it and without using additional deployment quota, since the model is only moved between profiles.

Deleting a Model

If a model is no longer required, it can be deleted from either the More menu More Menu on the MT model page or from the three-dot menu More Menu on the specific model page.

Deleting a model will remove it from the Phrase platform and undeploy it automatically.

Was this article helpful?

Sorry about that! In what way was it not helpful?

The article didn’t address my problem.
I couldn’t understand the article.
The feature doesn’t do what I need.
Other reason.

Note that feedback is provided anonymously so we aren't able to reply to questions.
If you'd like to ask a question, submit a request to our Support team.
Thank you for your feedback.