def get_engine(self):
return DistributedDataParallelEngine(port=DDP_ADDRESS +
random.randint(1, 100))
def get_engine(self):
return DistributedDataParallelEngine(
port=DDP_ADDRESS + random.randint(1, 100),
process_group_kwargs={"backend": "nccl"})
def test_ddp_load_on_stage_start():
with TemporaryDirectory() as logdir:
runner = CustomRunner(logdir, DistributedDataParallelEngine())
runner.run()