Possible to upload more than one dataset at time to azure custom translator

Hello Klaus Bergh,

In Azure Custom Translator, each project is tied to a single language pair, so it is not possible to train multiple language pairs (e.g., English–Danish and English–German) within the same project. If you require multiple translation directions, you will need to create a separate project for each pair. Within a project, you can upload multiple datasets (parallel, monolingual, or dictionary files) as long as they match the project’s language pair, and you can retrain your model at any time.

To simplify uploads, you can also group multiple files into a ZIP archive, provided they are structured correctly and follow the required naming conventions (e.g., data_en for English, data_da for Danish). This makes it easier to upload and process several datasets in one step.

Additionally, Custom Translator offers a REST API that allows you to automate tasks such as uploading data, training models, checking training status, and deploying models. Once trained, your custom models can be accessed through the standard Azure Translator Text API by using the category parameter to specify your custom model.

For more details, please refer How to upload Azure AI Foundry custom translation model datasets,

Build and manage Azure AI Custom Translator training documents.

I Hope this helps. Do let me know if you have any further queries.

If this answers your query, please do click Accept Answer and Yes for was this answer helpful.

Thank you!

Hi,

to my knowledge:

You can not train one “multi-language” custom model. In Custom Translator, each project is for one language pair (e.g., EN-DA or EN-DE). You should create one project per pair and train them separately. You can kick off several trainings in parallel (separate projects), and you can reuse the same workspace and datasets across those projects.

And yes - there is an API. Custom Translator has a management API for creating projects, uploading data, training, and publishing. Once a model is published, you use the regular Translator Text API to translate, passing your models "category" (an ID).

Even though training is per pair, the Text Translation API supports multi-target in one request (e.g. translating EN-DA & EN-DE).

Best regards :)