You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The logs of my jobs are missing some lines (I print one line every batch, approx 6 out of 7 are missing!) and sometimes the lines are scrambled (the outputs of different epochs are interleaved). Note that these are two separate issues, i.e., I checked that the missing lines are not just somewhere else in the log, they are actually missing.
It might have to do with progressbar2, but I have no evidence of this. The only reason for this claim is that there might be an incompatibility with progressbar rewriting multiple times the same line.
The text was updated successfully, but these errors were encountered:
I've never heard of this one. Might be related to progressbar2 which I've never tried with smartdispatch. I don't know if @MarcCote knows more about this issue. For now, can you tell us on which cluster you've seen this issue and do you have code to reproduce?
I was using Helios. You can almost certaintly reproduce it with this code (I'm using the same main loop, which is where the progressbar is called): https://github.com/fvisin/reseg
The logs of my jobs are missing some lines (I print one line every batch, approx 6 out of 7 are missing!) and sometimes the lines are scrambled (the outputs of different epochs are interleaved). Note that these are two separate issues, i.e., I checked that the missing lines are not just somewhere else in the log, they are actually missing.
It might have to do with progressbar2, but I have no evidence of this. The only reason for this claim is that there might be an incompatibility with progressbar rewriting multiple times the same line.
The text was updated successfully, but these errors were encountered: