def run_mnist_2gpu(self, target_file_name, eager_mode=True):
if not fluid.core.is_compiled_with_cuda(
) or fluid.core.get_cuda_device_count() == 0:
return
selected_gpus = get_gpus('0,1')
cluster = None
pod = None
cluster, pod = get_cluster_from_args(selected_gpus)
procs = start_local_trainers(
cluster,
pod,
eager_mode=eager_mode,
training_script=target_file_name,
training_script_args=[])
while True:
alive = watch_local_trainers(procs, cluster.trainers_endpoints())
if not alive:
print("Local procs complete, POD info:{}".format(pod))
break
time.sleep(3)
def run_mnist_2cpu(self, target_file_name):
cluster, pod = get_cluster_from_args(
[0, 1]) #tmp use. for getting trainer_nranks()
procs = start_local_trainers_cpu(cluster.trainers_endpoints(),
training_script=target_file_name,
training_script_args=[])
while True:
alive = watch_local_trainers(procs, cluster.trainers_nranks())
if not alive:
print("Local procs complete, POD info:{}".format(pod))
break
time.sleep(3)