def __init__(self, monitor_interval=1):
self._num_workers = current_cluster_size()
self._step = counter()
self._interval = monitor_interval
self._summed_variance = None
self._variances = None
def __init__(self, device_batch_size, monitor_interval=1):
self._num_workers = current_cluster_size()
self._step = counter()
self._interval = monitor_interval
self._device_batch_size = tf.cast(device_batch_size, dtype=tf.float32)
self._global_batch_size = self._device_batch_size * self._num_workers
def test_counter():
from kungfu.tensorflow.ops import counter
c = counter()
with tf.Session() as sess:
for i in range(10):
n = sess.run(c)
assert (n == i)
def test_counter():
c = counter()
with tf.Session() as sess:
for i in range(10):
n = sess.run(c)
print(n)
if n != i:
raise RuntimeError('counter failed!')
def build_ops():
init_step = int(_get_init_step())
print('init_step is %d' % (init_step))
step = counter(init_step)
schedule = step_based_schedule(config, step)
ckpt_tensor = tf.as_string(step + 1)
resize_op = resize_cluster(ckpt_tensor, schedule)
return init_step, resize_op
def __init__(self,
device_batch_size,
monitor_interval=1,
alpha=0.9,
verbose=False):
self._num_workers = tf.cast(cluster_size(), tf.float32)
self._alpha = alpha
self._step = counter()
self._verbose = verbose
self._interval = monitor_interval
self._device_batch_size = tf.cast(device_batch_size, dtype=tf.float32)
self._global_batch_size = self._device_batch_size * self._num_workers
def __init__(self, fuse_requests, fused_model_name=None):
self._step = counter()
self._fuse_requests = fuse_requests
self._fused_model_name = fused_model_name
def _build_resize_op(self, config, init_step):
step = counter(init_step)
new_size = step_based_schedule(config, step)
ckpt_tensor = tf.as_string(step + 1)
resize_op = resize_cluster(ckpt_tensor, new_size)
return resize_op
def test_counter():
c = counter()
with tf.Session() as sess:
for i in range(10):
n = sess.run(c)
assert (n == i)
def test_counter_init():
c = counter(init=1)
with tf.Session() as sess:
for i in range(10):
n = sess.run(c)
assert (n == i + 1)
