I wish to use multiple GPUs for data parallel training. But unlike the general situation, I hope that each GPU can receive sequential data instead of randomly sampled data. That is, the dataset can provide batch data to devices according to their id.
How can I do this using lightning?
Thanks.