Hi,
What could be possible reasons for encountering a BUS error in DDP? My guess would be an OOM error but as far as I can see RAM should be sufficient for the task. Could the shuffling of the data play a role when creating a torch dataloader? This is my implementation:
data = DataLoader(
dataset,
batch_size=batch_size,
pin_memory=True,
shuffle=False,
num_workers=0,
sampler=DistributedSampler(dataset)
)