Calculating training loss - sum of avg batch losses or avg loss across all samples

[YASJAY (Srinath)]

Hi,

how is loss calculated during training,

is it batchwise average or overall data average?