Once trained and fine tuned, can the model be uploaded or transferred to a lower spec server somewhere (eg a cheap cloud server) to use as an on-demand service, or does inference carry the same fairly beefy hardware requirements?