Let's say once the model is trained. Is it way less expensive for smaller companies to just download it and run it on their own servers?
Yes, GPUs are crazy expensive, and will probably continue to be for this decade.
You do have to invest to retrain so it does not drift much.
It required $4.6 million in hardware costs to train. Many companies could afford that, as long as they have the labor expertise to execute also.
Are they using their own HPC cluster? How much electricity did they use? Some of these national lab supercomputers use MWs of energy
Probably this we’re running on cloud computing service, that’s how you can get a specific number.
I think serving such models is expensive too. Also all the $$ they have to pay the researchers. Compared to that, model training cost is nothing.
Tech Industry
Yesterday
2882
Tech companies to avoid as a white guy?
Tech Industry
Yesterday
2047
Lack of diversity in engineering division at X
Tech Industry
Yesterday
2492
1 vs 5 Million - no lifestyle change
Tech Industry
Yesterday
315
Bitcoin is the only possible future
Ask Blinders
Yesterday
477
Why is our country owned by Israel? I don't want my tax dollars fund genocide. How can we stop this nonsense?
It’s dealing with questions from Blind.