def run_launcher():
import ncluster
if args.aws:
ncluster.set_backend('aws')
job = ncluster.make_job(args.name, num_tasks=2, image_name=args.image)
job.upload(__file__)
job.upload('util.py')
sender, receiver = job.tasks
# kill python just for when tmux session reuse is on
if not ncluster.running_locally():
sender._run_raw('killall python', ignore_errors=True)
receiver._run_raw('killall python', ignore_errors=True)
if ncluster.get_backend() == 'aws':
# on AWS probably running in conda DLAMI, switch into TF-enabled env
job.run('source activate tensorflow_p36')
ip_config = f'--sender-ip={sender.ip} --receiver-ip={receiver.ip}'
receiver.run(f'python {__file__} --role=receiver {ip_config}',
non_blocking=True)
sender.run(
f'python {__file__} --role=sender {ip_config} --iters={args.iters} --size-mb={args.size_mb} --shards={args.shards}')
print(sender.read('out'))
def run_launcher():
import ncluster
job = ncluster.make_job('tf_adder', num_tasks=2)
job.upload(__file__)
sender, receiver = job.tasks
if ncluster.get_backend() == 'aws':
# on AWS probably are running in DLAMI, switch into TF-enabled env
job.run('source activate tensorflow_p36')
ip_config = f'--sender-ip={sender.ip} --receiver-ip={receiver.ip}'
job.tasks[1].run(f'python tf_adder.py --role=receiver {ip_config}', async=True)
job.tasks[0].run(f'python tf_adder.py --role=sender {ip_config}')
def run_launcher():
import ncluster
job = ncluster.make_job('tf_adder_tb', num_tasks=2)
job.upload(__file__)
this_file = os.path.basename(__file__)
sender, receiver = job.tasks
if ncluster.get_backend() == 'aws':
# on AWS probably are running in DLAMI, switch into TF-enabled env
job.run('source activate tensorflow_p36')
ip_config = f'--sender-ip={sender.ip} --receiver-ip={receiver.ip}'
job.tasks[1].run(f'python {this_file} --role=receiver {ip_config}', non_blocking=True)
job.tasks[0].run(f'python {this_file} --role=sender --logdir={job.logdir} {ip_config}')
job.tasks[0].run(f'tensorboard --logdir={job.logdir}/..', non_blocking=True)
print(f"Benchmark done, tensorboard at http://{job.tasks[0].public_ip}:6006")
def run_launcher():
import ncluster
if args.aws:
ncluster.set_backend('aws')
job = ncluster.make_job('tf_adder', num_tasks=2, image_name=args.image)
job.upload(__file__)
sender, receiver = job.tasks
if ncluster.get_backend() == 'aws':
# on AWS probably running in conda DLAMI, switch into TF-enabled env
job.run('source activate tensorflow_p36')
ip_config = f'--sender-ip={sender.ip} --receiver-ip={receiver.ip}'
receiver.run(f'python tf_adder.py --role=receiver {ip_config}',
non_blocking=True)
sender.run(f'python tf_adder.py --role=sender {ip_config} --iters={args.iters}')
def run_launcher():
import ncluster
if args.aws:
ncluster.set_backend('aws')
job = ncluster.make_job(args.name, num_tasks=2, image_name=args.image)
job.upload(__file__)
job.upload('util.py')
# kill python just for when tmux session reuse is on
if not ncluster.running_locally():
job._run_raw('killall python', ignore_errors=True)
if ncluster.get_backend() == 'aws':
# on AWS probably running in conda DLAMI, switch into TF-enabled env
job.run('source activate tensorflow_p36')
hosts = [task.public_ip for task in job.tasks]
host_str = ','.join(hosts)
os.system(
f'mpirun -np 2 --host {host_str} python {__file__} --role=worker')
print(job.tasks[0].read('/tmp/out'))
def run_launcher():
import ncluster
job = ncluster.make_job(args.name, num_tasks=2, image_name=args.image,
instance_type=args.instance_type)
job.upload(__file__)
job.upload('util.py')
# kill python just for when tmux session reuse is on
if not ncluster.running_locally():
job._run_raw('killall python', ignore_errors=True)
if ncluster.get_backend() == 'aws':
# on AWS probably running in conda DLAMI, switch into TF-enabled env
# TODO(y) switch to PyTorch enabled
job.run('source activate tensorflow_p36')
# TODO(y): this should be private ip
hosts = [task.ip for task in job.tasks]
host_str = ','.join(hosts)
os.system(f'/usr/local/mpi/bin/mpirun -np 2 --host {host_str} python {__file__} --role=worker')
print(job.tasks[0].read('/tmp/out'))