Dynamic Data Reinsertion in Dataloader

Hi,

Hi there ! I am training a model requiring recursive predictions. After each step, some data is fully processed, while the rest needs to be reinserted into the dataloader for further iterations.

How can I dynamically reinsert partially predicted data into the dataloader during training ? Are there recommended hooks or approaches for this? Should I work on dataset / dataloader / batch sampler level ?

Thanks in advance!