Small models can be trained on the developer machines, but to speed things up and to be able to run bigger nets we could use AWS GPU spot instances which cost 1$/hour. In my company with relatively small models we pay >1000$/month. We will probably reach that unless we are really successful.
Small models can be trained on the developer machines, but to speed things up and to be able to run bigger nets we could use AWS GPU spot instances which cost 1$/hour. In my company with relatively small models we pay >1000$/month. We will probably reach that unless we are really successful.