Is there any way of combining an IterDataPipe
dataset with DDP?
I have tried to convert my IterDataPipe to Map-style dataset with to_map_datapipe()
but without success.
Also, I am not sure if there are better ways to do this.
Thanks
Is there any way of combining an IterDataPipe
dataset with DDP?
I have tried to convert my IterDataPipe to Map-style dataset with to_map_datapipe()
but without success.
Also, I am not sure if there are better ways to do this.
Thanks
Could you please elaborate on it?
In the documentation it seems that using torch.utils.data.distributed.DistributedSampler
is not compatible with an Iter-style dataset. I was just wondering if there is any way of making it compatible.
Some people have suggested some solutions when using multiple workers in IterDataPipes
.