video tba
article tba
previous (cascade training study)
we continue the training study by using the same dataset:
https://github.com/MushroomFleet/assassinKahb-1024
we will use PRODIGY :
V1 - used only class tokens for every image "assassinkahb style"
V2 - used all the caption pairs found in the dataset
as explained we are comparing selected early epochs in the training process for testing, these are marked with "-early".