Feature request: Add total steps & estimated/remaining duration to train and evaluate

[HarikalarKutusu]

Is your feature request related to a problem? Please describe. When running train and evaluate we don't know how long it will take and where we are wrt total. Knowing them would help a lot to answer such questions:

• Should I run the process overnight?
• Will it fit into the limits of my Google Colab subscription?
• How long more do I need to wait?

AFAIK, as the set train/test size and batch sizes are known one can calculate toral steps. Also from the executed steps and elapsed time the expected total duration can be estimated.

Describe the solution you'd like Instead of this:

Test epoch | Steps: 104 | Elapsed Time: 0:31:46

Such an output will be very helpful:

Test epoch | Steps: 104/128 | Elapsed Time: 0:31:10 (estimated total 0:45:20, remaning 0:14:10)

Describe alternatives you've considered Currently, I need to calculate them manually from time to time to decide if I need to go to sleep or wait a bit more :)

[reuben]

We can't always know the dataset size, for example with the recently introduced WebDataset input. I guess we could save the total after the first epoch and then use it for subsequent epochs. PRs welcome!

[HarikalarKutusu]

I was afraid of that :(