I was wondering if I’m missing here something or not.
I can see the keep_checkpoint_max: n
variable in train parameters group but I was wondering if that can used under eval group of parameters? I’m evaluating every steps: 5000
and exporting a model with export_on_best: bleu
. As my training continues, now I have like 10-15 of about ~300MB size model folders from exports. So, I’m wondering if there’s any way perhaps I can limit this same as what we have with train keep_checkpoint_max
!
Is there anything I’m missing here? I think it would be convenient to only keep last n copies of model exports from disk space point of view. Thoughts?