Hi
I think for this it could be useful to implement a collate_fn function where you can define the concatenation of your data into a batch:
def collate_fn(samples):
# samples is the list of samples returned from your
# dataset, to be assembled into a batch
# [[{a1},{a2}...{an}],[{b1},{b2}...{bn}]]
return samples
dataloader = DataLoader(..., collate_fn=collate_fn)
Here are the PyTorch docs for this.
Hope this helps