Hi,
couldn’t find any description for the Train Output report.
What does total_flos mean?
Does it show the total amount of floating operations (mult, add etc) that took place during the training process? And is it modelled or based on real logging/counting?
If it’s the same as FLOPs (multiple ops, not seconds) why not call it that? Just curious.
Really need to know to estimate needed resources for an ongoing project.
Can someone explain to me this parameter, please.