2 ms·
If you are training models that are intended to be used in production at scale then training is dirt cheap compared to inference. There is a reason why Google f
by Jensson 4y ago
If you are training models that are intended to be used in production at scale then training is dirt cheap compared to inference. There is a reason why Google focused on inference first with their TPU's even though Google does a lot of ML training.
- dllthomas 4y agoI think another part of the question is whether you're scaling on your own hardware or the customers' hardware.