Comment by Ianjit

Comment by Ianjit 3 hours ago

0 replies

The incremental compute costs will scale with the incremental data added, therefore training costs will grow at a much slower rate compared to when training was GPU limited.