3 ms·
Inference costs can't be faked though, since the model can be run locally by anyone with capable hardware. Even if the whole story about the training cost was
by diego_sandoval 2y ago
Inference costs can't be faked though, since the model can be run locally by anyone with capable hardware.
Even if the whole story about the training cost was fake, R1 and the distilled models are still very efficient at inference.
- nextworddev 2y agounless OAI has all the optimization tricks already, which they probably do
- croes 2y agoThe shock for the industry was the claimed training costs and used hardware.
- creato 2y agoIs the model architecture actually that different from anything else? Or are you just saying that you can get away with smaller models now?