Pytorch DDP join without backward

Recently I try to handle uneven inputs with pytorch DDP.

The join() of DDP works well for training loop, but does not for evaluation loop.

The below code sniffet is from pytorch docs, but does not work although I only removed loss.backward()

How can I handle uneven input in evaluation loop?

import os
import torch
import torch.distributed as dist
import torch.multiprocessing as mp
from torch.distributed.algorithms.join import Join
from torch.nn.parallel import DistributedDataParallel as DDP

BACKEND = "nccl"
WORLD_SIZE = 2
NUM_INPUTS = 5

def worker(rank):
    os.environ['MASTER_ADDR'] = 'localhost'
    os.environ['MASTER_PORT'] = '29500'
    dist.init_process_group(BACKEND, rank=rank, world_size=WORLD_SIZE)

    model = DDP(torch.nn.Linear(1, 1).to(rank), device_ids=[rank])
    # Rank 1 gets one more input than rank 0
    inputs = [torch.tensor([1]).float() for _ in range(NUM_INPUTS + rank)]

    num_inputs = 0
    with Join([model]):
        for input in inputs:
            num_inputs += 1
            loss = model(input).sum()
            # loss.backward() (original example)
    print(f"Rank {rank} has exhausted all {num_inputs} of its inputs!")

def main():
    mp.spawn(worker, nprocs=WORLD_SIZE, join=True)

if __name__ == "__main__":
    main()