def test_train(ray_start_4_cpus):
def simple_fn(worker):
local_rank = _train()
return local_rank
setting = RayExecutor.create_settings(timeout_s=30)
hjob = RayExecutor(
setting, num_workers=3, use_gpu=torch.cuda.is_available())
hjob.start()
result = hjob.execute(simple_fn)
assert set(result) == {0, 1, 2}
result = ray.get(hjob.run_remote(simple_fn, args=[None]))
assert set(result) == {0, 1, 2}
hjob.shutdown()
def test_ray_exec_remote_func(ray_start_4_cpus):
def simple_fn(num_epochs):
import horovod.torch as hvd
hvd.init()
return hvd.rank() * num_epochs
setting = RayExecutor.create_settings(timeout_s=30)
hjob = RayExecutor(
setting, num_hosts=1, num_slots=4, use_gpu=torch.cuda.is_available())
hjob.start()
object_refs = hjob.run_remote(simple_fn, args=[0])
result = ray.get(object_refs)
assert len(set(result)) == 1
hjob.shutdown()