3 ms·
From their numbers 3 hours with 32 A100 80GBs. From lambda cloud: 3 hours * 4 ~22$/hr for 8x A100 ~= $265 So yeah not too expensive even for a native fine tu
by kirill5pol 3y ago
From their numbers 3 hours with 32 A100 80GBs.
From lambda cloud:
3 hours * 4 ~22$/hr for 8x A100 ~= $265
So yeah not too expensive even for a native fine tune (obviously this ignores all other costs other than the GPUs)