from experience, what kind of ec2 instances are suited best to train rasa models, our training data is not a lot, mostly around 3-4k examples. Only training models, for inference we will have a separate instance
will medium sized instances do?
Hi there and thanks for your question! We do not have training-only recommendations and it highly depends on your model configuration. This is something that I think your could empirically validate on a local setup by analyzing memory and CPU usage during training and then finding out the ec2 instance that is best suited to your needs. I wish they was a rule of thumb / turn-key solution for this but there are none to my knowledge