def run_test(output_name, output_nodes, recreate=True):
batch_size = 1
source_shapes = {
ph.name: [
int(d.value) if d.value is not None else batch_size
for d in ph.shape.dims
]
for ph in get_placeholders()
}
pb_path = os.path.join('out', 'pb', output_name)
network_name = output_name.rstrip('/').replace('/', '_')
with tf.Session() as sess:
init = tf.global_variables_initializer()
sess.run(init)
save_protobuf(pb_path, output_nodes, sess, recreate)
sess.close()
for prefer_nhwc in [True, False]:
test_activations(filename=os.path.join(pb_path, 'test.pb'),
source_shapes=source_shapes,
feed_dict={
utils.anystr_to_str(k): np.random.random(v)
for k, v in six.iteritems(source_shapes)
},
prefer_nhwc=prefer_nhwc,
network_name=network_name,
delete_after_each=False,
export_io_only=True)
return nnef.parse_file(
os.path.join('out', network_name + '_nhwc', 'nnef',
network_name + '_nnef', 'graph.nnef'))
def main(args):
if args.seed is not None:
np.random.seed(args.seed)
distributions = {
'scalar': uniform(0.0, 1.0),
'integer': integers(0, 100),
'logical': bernoulli(0.5),
}
try:
random = eval(args.random)
if isinstance(random, dict):
distributions.update(
{key: _ensure_lambda(value)
for key, value in random.items()})
else:
random = _ensure_lambda(random)
if args.random.startswith('integers'):
distributions['integer'] = random
elif args.random.startswith('bernoulli'):
distributions['logical'] = random
else:
distributions['scalar'] = random
except Exception as e:
print("Could not evaluate distribution: " + str(e), file=sys.stderr)
return -1
graph = nnef.parse_file(os.path.join(args.model, 'graph.nnef'))
for op in graph.operations:
if args.weights and op.name == 'variable':
label = op.attribs['label']
shape = op.attribs['shape']
data = distributions[op.dtype](shape)
filename = os.path.join(args.model, label + '.dat')
os.makedirs(os.path.split(filename)[0], exist_ok=True)
with open(filename, 'wb') as file:
nnef.write_tensor(file, data)
if args.verbose:
print("Generated weight '{}'".format(filename))
if args.inputs and op.name == 'external':
name = op.outputs['output']
shape = op.attribs['shape']
data = distributions[op.dtype](shape)
filename = os.path.join(args.model, args.inputs, name + '.dat')
os.makedirs(os.path.split(filename)[0], exist_ok=True)
with open(filename, 'wb') as file:
nnef.write_tensor(file, data)
if args.verbose:
print("Generated input '{}'".format(filename))
return 0
def run(self):
# Changing dir to NNEF's graph parent folder.
network_dir, model_filename = os.path.split(self.input_model)
if network_dir != '':
prevdir = os.getcwd()
os.chdir(network_dir)
try:
attr, ops = nnef.parse_file(model_filename)
except Exception as ex:
print('WARNING: converter ignoring exception:', ex)
nnef._register_layer_ops()
try:
attr, ops = nnef.parse_file(model_filename)
except Exception as ex:
raise Exception('failed to open %s: %s' %
(self.input_model, ex))
self.graph_name = attr['graph'].name
for operand in ops:
if hasattr(self, "import_" + operand[0][0]):
func = getattr(self, "import_" + operand[0][0])
returned_node = func(operand)
self.node_pool[returned_node.name] = returned_node
else:
self.import_UNKNOWN(operand)
input_nodes = self.get_input_nodes()
output_nodes = self.get_output_nodes()
graph = NNEFGraph(os.path.basename(self.input_model).split('.')[0],
input_nodes,
output_nodes,
node_pool=self.node_pool)
if prevdir is not None:
os.chdir(prevdir)
return graph
def __init__(self,
path,
device=None,
decomposed=None,
custom_operators=None):
self._nnef_graph = nnef.parse_file(os.path.join(path, 'graph.nnef'),
lowered=decomposed)
self._init_input_shapes(self._nnef_graph)
self._nnef_module = NNEFModule(path=path,
custom_operators=custom_operators,
decomposed=decomposed)
if device is None:
device = 'cuda' if torch.cuda.is_available() else 'cpu'
self._nnef_module.to(device)
self._nnef_module.eval()
self._device = device
def main(args):
if args.seed is not None:
np.random.seed(args.seed)
try:
distribution = eval(args.random)
if not _is_lambda(distribution):
distribution = distribution()
except Exception as e:
print("Could not evaluate distribution: " + str(e), file=sys.stderr)
return -1
graph = nnef.parse_file(os.path.join(args.model, 'graph.nnef'))
for op in graph.operations:
if args.weights and op.name == 'variable':
label = op.attribs['label']
shape = op.attribs['shape']
data = distribution(shape).astype(_nnef_dtype_to_numpy[op.dtype])
filename = os.path.join(args.model, label + '.dat')
os.makedirs(os.path.split(filename)[0], exist_ok=True)
with open(filename, 'wb') as file:
nnef.write_tensor(file, data)
if args.verbose:
print("Generated weight '{}'".format(filename))
if args.inputs and op.name == 'external':
name = op.outputs['output']
shape = op.attribs['shape']
data = distribution(shape).astype(_nnef_dtype_to_numpy[op.dtype])
filename = os.path.join(args.model, args.inputs, name + '.dat')
os.makedirs(os.path.split(filename)[0], exist_ok=True)
with open(filename, 'wb') as file:
nnef.write_tensor(file, data)
if args.verbose:
print("Generated input '{}'".format(filename))
return 0
def parse_file(file_name):
with default_config:
return nnef.parse_file(file_name, atomics=AtomicOperations)
def test_vggnet(self):
nnef.parse_file("../examples/vgg.txt")
def test_resnet(self):
nnef.parse_file("../examples/resnet.txt")
def test_googlenet(self):
nnef.parse_file("../examples/googlenet.txt")
def test_alexnet(self):
nnef.parse_file("../examples/alexnet.txt")