Can we train a yolo model for deepstream on Azure ML or are we obligated to use TAO?

I am using tao to train a licence plate detector with a yolov4 tiny model, can I use Azure ML without tao do train or to be compatible with deepstream ?

You can train without tao. But for deepstream, I am not sure if there is an available example to show it can run smoothly against the model from Azure ML. You can ask question to Deepstream forum to confirm.