Pytorch Lightning keeps track of all its outputs through one epoch. That is why, we observed the OOM error during the predict loop. If we disable the return_predictions flag, OOM errors don't occur anymore. For a detailed discussion on this, refer the comments on #19.
Pytorch Lightning keeps track of all its outputs through one epoch. That is why, we observed the OOM error during the predict loop. If we disable the
return_predictions
flag, OOM errors don't occur anymore. For a detailed discussion on this, refer the comments on #19.