def _get_weight_path(path):
env = os.environ
if 'PADDLE_TRAINERS_NUM' in env and 'PADDLE_TRAINER_ID' in env:
trainer_id = int(env['PADDLE_TRAINER_ID'])
num_trainers = int(env['PADDLE_TRAINERS_NUM'])
if num_trainers <= 1:
path = get_weights_path(path)
else:
from ppdet.utils.download import map_path, WEIGHTS_HOME
weight_path = map_path(path, WEIGHTS_HOME)
lock_path = weight_path + '.lock'
if not os.path.exists(weight_path):
try:
os.makedirs(os.path.dirname(weight_path))
except OSError as e:
if e.errno != errno.EEXIST:
raise
with open(lock_path, 'w'): # touch
os.utime(lock_path, None)
if trainer_id == 0:
get_weights_path(path)
os.remove(lock_path)
else:
while os.path.exists(lock_path):
time.sleep(1)
path = weight_path
else:
path = get_weights_path(path)
return path
def get_weights_path_dist(path):
env = os.environ
if 'PADDLE_TRAINERS_NUM' in env and 'PADDLE_TRAINER_ID' in env:
trainer_id = int(env['PADDLE_TRAINER_ID'])
num_trainers = int(env['PADDLE_TRAINERS_NUM'])
if num_trainers <= 1:
path = get_weights_path(path)
else:
from ppdet.utils.download import map_path, WEIGHTS_HOME
weight_path = map_path(path, WEIGHTS_HOME)
lock_path = weight_path + '.lock'
if not os.path.exists(weight_path):
from paddle.distributed import ParallelEnv
unique_endpoints = _get_unique_endpoints(
ParallelEnv().trainer_endpoints[:])
try:
os.makedirs(os.path.dirname(weight_path))
except OSError as e:
if e.errno != errno.EEXIST:
raise
with open(lock_path, 'w'): # touch
os.utime(lock_path, None)
if ParallelEnv().current_endpoint in unique_endpoints:
get_weights_path(path)
os.remove(lock_path)
else:
while os.path.exists(lock_path):
time.sleep(1)
path = weight_path
else:
path = get_weights_path(path)
return path