Closed DrewGalbraith closed 3 months ago
The reason the current situation is not ideal is depicted in the attached screenshot.
Though the higher-loss ckpts were generated first, they appear below. Additionally, if validation starts climbing again, it could even make a ckpt from several epochs later appear as .v18 or something following .v17 from much earlier in the training sequence, forcing us to rely on
ls -l
and like commands to divine creation order. This is way too much work.
CKPT files are being names like this rn:
Instead of vn, we want o=to order them in order of their appearance. So the implemented ouput will be: