4 ms·
Not if the current architecture plateaus around the level of O1/3.5/R1. Then any further training is useless and inference is relatively cheap.
by csomar 2y ago
Not if the current architecture plateaus around the level of O1/3.5/R1. Then any further training is useless and inference is relatively cheap.