When is it safe to return gpu/cuda tensors. e.g. may I return it in my collate function after I collate everything into 1 tensor?
In that case do we still need to do pin_memory=True
?
When is it safe to return gpu/cuda tensors. e.g. may I return it in my collate function after I collate everything into 1 tensor?
In that case do we still need to do pin_memory=True
?