As the model is scaled up, compute requirements may grow beyond the "conventional". We should ensure that our early design decisions to do impede this, if possible: Liaise with HPC / GPU experts and obtain advice.
I have some experience with spark and azure batch jobs. Do we know what kind of options are open to us? Do we have an AWS account or something like that?
As the model is scaled up, compute requirements may grow beyond the "conventional". We should ensure that our early design decisions to do impede this, if possible: Liaise with HPC / GPU experts and obtain advice.