BERT Model

Options
vinhdiesal
vinhdiesal Registered Posts: 11 ✭✭✭✭

Dataiku published a natural language processing article which states different models that can be used including BERT. I really would like to use BERT inside the DSS to train on the datasets we have.

Can you provide some guidance on how to load the BERT model inside DSS? Perhaps we can start with the models from TF Hub.

Thanks,

Vinh

Best Answer

  • CoreyS
    CoreyS Dataiker Alumni, Dataiku DSS Core Designer, Dataiku DSS Core Concepts, Registered Posts: 1,150 ✭✭✭✭✭✭✭✭✭
    Answer ✓
    Options

    As an update, while you can always integrate it via code, additionally, you can now use BERT & any other model on HuggingFace in the Visual ML design interface, using the text preprocessing method called Sentence Embedding.

Answers

  • Alex_Combessie
    Alex_Combessie Alpha Tester, Dataiker Alumni Posts: 539 ✭✭✭✭✭✭✭✭✭
    Options

    Hi,

    As of today, using BERT in DSS can be achieved with python code. You can use any library such as tensorflow-hub or huggingface, as you would do outside of DSS.

    Optionally, you can use the dataiku API to store model files in a DSS managed folder.

    Note that we plan to add transformers models such as BERT to the NLP Embedding plugin.

    Best regards,

    Alex

Setup Info
    Tags
      Help me…