Beginner questions regarding custom Intent/Slot classification models


To start, I’m a software engineer with little experience in the world of machine learning, so i apologize for some questions that may be elementary.

I’m trying to use the RIVA API for Natural language processing and I want to custom train the model on different domains (airline, medical, finance, etc).

I’ve gathered that I need NeMo/TAO to automate most of the training, deployment etc but I can’t find any docs on how to actually add custom intents, what format the training data should be, and what training data even looks like.

So my questions are:

  1. Is there a central hub for free-to-use models?
  2. When i find a model, where can i find the encryption key that accompanies the model?
  3. When I get my hands on a model, how can i find the spec/format that was used to train it?
  4. Once I have the training data format, how can I go about adding domains/intents/slots to the existing model? Do i keep my training data in a specific location?
  5. Finally, once I have my trained custom model, I noticed there is a “domain” or “model” param on a lot of RIVA APIs, they seem to be string values. Is there a specific location RIVA expects the models to live so they can be accessed?

Thank you.


We recommend you to please go through Riva user guide, where you can also find inputs/links for pretrained models/custom training and deployement related

Thank you.