def main():
args = parse_arguments()
use_ideep = False
if is_ideep_available():
if args.ideep:
use_ideep = True
else:
print('>> you can use ideep to accelerate inference speed')
print('>> with optional argument --ideep')
predict(args.arch, args.trained, use_ideep)
def __init__(self):
self.chainer_version = chainer.__version__
self.numpy_version = numpy.__version__
self.platform_version = platform.platform()
if cuda.available:
self.cuda_info = cuda.cupyx.get_runtime_info()
else:
self.cuda_info = None
if intel64.is_ideep_available():
self.ideep_version = intel64.ideep.__version__
else:
self.ideep_version = None
def accuracy(self, backend_config):
model = self.model
optimizer = self.optimizer
optimizer.setup(model)
if backend_config.use_ideep == 'always':
if not intel64.is_ideep_available():
# TODO(niboshi): This is temporary workaround.
# See the comment on Skipped.
raise Skipped('ideep is required to run this test.')
model.to_device(backend_config.device)
with chainer.using_device(backend_config.device):
return self._train_linear_classifier(
model, optimizer, backend_config)
def accuracy(self, backend_config, gpu_device=None):
model = self.model
optimizer = self.optimizer
optimizer.setup(model)
if backend_config.use_cuda:
model.to_gpu(device=gpu_device)
elif backend_config.use_ideep == 'always':
if not intel64.is_ideep_available():
# TODO(niboshi): This is temporary workaround.
# See the comment on Skipped.
raise Skipped('ideep is required to run this test.')
model.to_intel64()
with backend_config:
return self._train_linear_classifier(
model, optimizer, backend_config.use_cuda)
def accuracy(self, backend_config, loss_scaling=False):
model = self.model
optimizer = self.optimizer
optimizer.setup(model)
_optimizer_loss_scaling(optimizer, loss_scaling)
if backend_config.use_ideep == 'always':
if not intel64.is_ideep_available():
# TODO(niboshi): This is temporary workaround.
# See the comment on Skipped.
raise Skipped('ideep is required to run this test.')
model.to_device(backend_config.device)
with chainer.using_device(backend_config.device):
return self._train_linear_classifier(
model, optimizer, backend_config)
def accuracy(self, backend_config):
# TODO(niboshi): Support it
if backend_config.use_chainerx and self.dtype == numpy.float16:
raise unittest.SkipTest('ChainerX does not support float16')
model = self.model
optimizer = self.optimizer
optimizer.setup(model)
if backend_config.use_ideep == 'always':
if not intel64.is_ideep_available():
# TODO(niboshi): This is temporary workaround.
# See the comment on Skipped.
raise Skipped('ideep is required to run this test.')
model.to_device(backend_config.device)
with chainer.using_device(backend_config.device):
return self._train_linear_classifier(model, optimizer,
backend_config)
def accuracy(self, backend_config):
# TODO(niboshi): Support it
if backend_config.use_chainerx and self.dtype == numpy.float16:
raise unittest.SkipTest('ChainerX does not support float16')
model = self.model
optimizer = self.optimizer
optimizer.setup(model)
if backend_config.use_ideep == 'always':
if not intel64.is_ideep_available():
# TODO(niboshi): This is temporary workaround.
# See the comment on Skipped.
raise Skipped('ideep is required to run this test.')
model.to_device(backend_config.device)
with chainer.using_device(backend_config.device):
return self._train_linear_classifier(
model, optimizer, backend_config)
def main():
enable_ideep = is_ideep_available()
model = VGG()
mode = "never"
if enable_ideep:
model.to_intel64()
mode = "always"
nb_itr = 20
timings = []
for i in tqdm.tqdm(range(nb_itr)):
data = np.random.randn(1, 3, 224, 224).astype(np.float32)
start_time = time.time()
with chainer.using_config('train', False):
with chainer.using_config('enable_backprop', False):
with chainer.using_config('use_ideep', mode):
ret = F.softmax(model(chainer.Variable(data)))
print(ret.data.ravel()[0])
timings.append(time.time() - start_time)
print(
'%10s : %f (sd %f)' %
('chainer-vgg-16', np.array(timings).mean(), np.array(timings).std()))
x = self.conv_ds_4(x)
x = self.conv_ds_5(x)
x = self.conv_ds_6(x)
x = self.conv_ds_7(x)
x = self.conv_ds_8(x)
x = self.conv_ds_9(x)
x = self.conv_ds_10(x)
x = self.conv_ds_11(x)
x = self.conv_ds_12(x)
x = self.conv_ds_13(x)
x = self.conv_ds_14(x)
x = F.average_pooling_2d(x, 7, stride=1)
return F.softmax(x)
enable_ideep = is_ideep_available()
model = MobileNet()
mode = "never"
if enable_ideep:
model.to_intel64()
mode = "always"
nb_itr = 20
timings = []
for i in tqdm.tqdm(range(nb_itr)):
data = np.random.randn(1, 3, 224, 224).astype(np.float32)
start_time = time.time()
with chainer.using_config('train', False):
with chainer.using_config('enable_backprop', False):
with chainer.using_config('use_ideep', mode):
