def _init_runtime(self):
""" Call _init_runtime under different CUDA_VISIBLE_DEVICES, you'll
have workers that run on multiGPUs
"""
if self.idx != 0:
from tensorpack.models.common import disable_layer_logging
disable_layer_logging()
self.predictor = OfflinePredictor(self.config)
if self.idx == 0:
with self.predictor.graph.as_default():
describe_model()
def _init_runtime(self):
""" Call _init_runtime under different CUDA_VISIBLE_DEVICES, you'll
have workers that run on multiGPUs
"""
if self.idx != 0:
from tensorpack.models.common import disable_layer_logging
disable_layer_logging()
self.predictor = OfflinePredictor(self.config)
if self.idx == 0:
with self.predictor.graph.as_default():
describe_model()
def _prepare(self):
disable_layer_logging()
self.predictor = OfflinePredictor(self.pred_config)
with self.predictor.graph.as_default():
vars_to_update = self._params_to_update()
self.sess_updater = SessionUpdate(
self.predictor.session, vars_to_update)
# TODO setup callback for explore?
self.predictor.graph.finalize()
self.weight_lock = threading.Lock()
# start a thread to wait for notification
def func():
self.condvar.acquire()
while True:
self.condvar.wait()
self._trigger_evt()
self.evt_th = threading.Thread(target=func)
self.evt_th.daemon = True
self.evt_th.start()
def _prepare(self):
disable_layer_logging()
self.predictor = OfflinePredictor(self.pred_config)
with self.predictor.graph.as_default():
vars_to_update = self._params_to_update()
self.sess_updater = SessionUpdate(
self.predictor.session, vars_to_update)
# TODO setup callback for explore?
self.predictor.graph.finalize()
self.weight_lock = threading.Lock()
# start a thread to wait for notification
def func():
self.condvar.acquire()
while True:
self.condvar.wait()
self._trigger_evt()
self.evt_th = threading.Thread(target=func)
self.evt_th.daemon = True
self.evt_th.start()