Training Log output

[rohanchn]

Has the training log output that looked like the snippet below been discontinued in 4.3.9?

stage 6/∞ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 23621/23621 0:00:00 0:13:10 val_accuracy: 0.93735 early_stopping: 0/10 0.93735 stage 7/∞ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 23621/23621 0:00:00 0:13:11 val_accuracy: 0.93994 early_stopping: 0/10 0.93994 stage 8/∞ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 23621/23621 0:00:00 0:13:14 val_accuracy: 0.94252 early_stopping: 0/10 0.94252

I'm training a recognition model and each new epoch is overwriting the output of the previous epoch. It works perfectly well in -v mode.

Like this: Total params: 4.0 M Total estimated model params size (MB): 16 stage 1/∞ ━╺━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 712/25483 0:00:20 • 0:11:35 35.65it/s v_num: 0 val_accuracy: 0.74 early_stopping: 0/10 0.74048

[mittagessen]

I had to rewrite the progress bar code completely because the lightning internals of it changed considerably and it was fairly deep in there. I'll see if I can find an easy way to keep it from being overwritten.

[rohanchn]

It's particularly useful to quickly compare models in an ongoing training session, but not something absolutely critical. Plus, I am not sure whether everyone finds it useful, so you may want to let it be if it's too much work.

Just want to thank you for your amazing work!