Is it possible to train a NLU model only and have the trained model saved in the cloud? Then reference the trained model to parse text at a later time?
yes, if you setup all the API correctly
and can the model be trained using a project directory that is in a s3 bucket (i.e. training data and config file are in a known s3 bucket)?
please check s3 bucket documentation, in general Rasa NLU model can be trained on was