Hi! Thank u for this wonderful work!
In evaluate.py-->compute_losses(self, dataloader), it seems that only one batch of data is used for evaluation.
But when it comes to loss computation,
losses = {k: sum(v) / len(dataloader) for k, v in storer.items()}
it uses len(dataloader) to average the loss. Should that be the length of element v?
I wonder if I misunderstand the above computation.
Any help will be appreciated!
Hi! Thank u for this wonderful work! In evaluate.py-->compute_losses(self, dataloader), it seems that only one batch of data is used for evaluation. But when it comes to loss computation,
it uses len(dataloader) to average the loss. Should that be the length of element v? I wonder if I misunderstand the above computation. Any help will be appreciated!