def create_experiment(args):
# to make it clear what are inside args
config_file = Path(args.config)
port = args.port
debug = args.debug
url_prefix = args.url_prefix
foreground = args.foreground
# it should finally be done in nnictl main function
# but for now don't break routines without logging support
init_logger_for_command_line()
logging.getLogger('nni').setLevel(logging.INFO)
if not config_file.is_file():
_logger.error(f'"{config_file}" is not a valid file.')
exit(1)
with config_file.open() as config:
config_content = yaml.safe_load(config)
v1_platform = config_content.get('trainingServicePlatform')
if v1_platform:
can_convert = True
if v1_platform == 'adl':
can_convert = False
if v1_platform in ['kubeflow', 'frameworkcontroller']:
reuse = config_content.get(v1_platform + 'Config', {}).get('reuse')
can_convert = (
reuse != False
) # if user does not explicitly specify it, convert to reuse mode
if not can_convert:
legacy_launcher.create_experiment(args)
exit()
try:
v2_config = convert.to_v2(config_content)
except Exception:
_logger.error(
'You are using legacy config format with incorrect fields or values, '
'to get more accurate error message please update it to the new format.'
)
_logger.error(
'Reference: https://nni.readthedocs.io/en/stable/reference/experiment_config.html'
)
exit(1)
_logger.warning(
f'You are using legacy config file, please update it to latest format:'
)
# use `print` here because logging will add timestamp and make it hard to copy paste
print(Fore.YELLOW + '=' * 80 + Fore.RESET)
print(yaml.dump(v2_config, sort_keys=False).strip())
print(Fore.YELLOW + '=' * 80 + Fore.RESET)
print(
Fore.YELLOW +
'Reference: https://nni.readthedocs.io/en/stable/reference/experiment_config.html'
+ Fore.RESET)
utils.set_base_path(config_file.parent)
config = ExperimentConfig(**v2_config)
utils.unset_base_path()
else:
config = ExperimentConfig.load(config_file)
if config.use_annotation:
path = Path(tempfile.gettempdir(), getuser(), 'nni', 'annotation')
path.mkdir(parents=True, exist_ok=True)
path = tempfile.mkdtemp(dir=path)
code_dir = expand_annotations(config.trial_code_directory, path)
config.trial_code_directory = code_dir
config.search_space = generate_search_space(code_dir)
assert config.search_space, 'ERROR: Generated search space is empty'
config.use_annotation = False
exp = Experiment(config)
exp.url_prefix = url_prefix
run_mode = RunMode.Foreground if foreground else RunMode.Detach
exp.start(port, debug, run_mode)
_logger.info(
f'To stop experiment run "nnictl stop {exp.id}" or "nnictl stop --all"'
)
_logger.info(
'Reference: https://nni.readthedocs.io/en/stable/Tutorial/Nnictl.html')
experiment.config.max_trial_number = 100
experiment.config.max_experiment_duration = '60d'
experiment.config.nni_manager_ip = '10.221.90.21'
experiment.config.search_space = search_space
experiment.config.trial_prepare_command = 'source /home/igor.quintanilha/miniconda3/bin/activate dsc'
experiment.config.trial_command = 'python main.py --gpus 1 data/brtd --vocab data/brtd/b3922f0904f4f1b7b258a9488132f2e6480cf936493be53f74fd7aaa07e14781.8f9337.vocab --batch-size 64 --max_epochs 10 --terminate_on_nan --num-embedding 400 --num-layers 3 --num-hidden 1150 --model awd --bptt 20 --max_steps 150000 --val_check_interval .25'
experiment.config.trial_code_directory = Path(__file__).parent.parent
experiment.config.trial_concurrency = 2
experiment.config.trial_gpu_number = 1
experiment.config.training_service[0].use_active_gpu = True
experiment.config.training_service[0].max_trial_number_per_gpu = True
experiment.config.training_service[1].reuse_mode = True
remote_confs = []
for ip in ['10.221.70.3', '10.221.70.15', '10.221.90.20']:
rm_conf = RemoteMachineConfig()
rm_conf.host = ip
rm_conf.user = '******'
rm_conf.ssh_key_file = '/home/igor.quintanilha/.ssh/id_rsa'
rm_conf.use_active_gpu = True
rm_conf.max_trial_number_per_gpu = 1
remote_confs.append(rm_conf)
experiment.config.training_service[1].machine_list = remote_confs
experiment.start(26780, debug=False)