def run(input_path: str,
dest_dir_path: str,
project_name: str,
activate_hard_quantization: bool,
threshold_skipping: bool = False,
num_pe: int = 16,
use_tvm: bool = False,
use_onnx: bool = False,
debug: bool = False,
cache_dma: bool = False):
output_dlk_test_dir = path.join(dest_dir_path, f'{project_name}.test')
optimized_pb_path = path.join(dest_dir_path, f'{project_name}')
optimized_pb_path += '.onnx' if use_onnx else '.pb'
output_project_path = path.join(dest_dir_path, f'{project_name}.prj')
config = Config(num_pe=num_pe,
activate_hard_quantization=activate_hard_quantization,
threshold_skipping=threshold_skipping,
tvm_path=(path.join(ROOT_DIR, 'tvm') if use_tvm else None),
test_dir=output_dlk_test_dir,
optimized_pb_path=optimized_pb_path,
output_pj_path=output_project_path,
debug=debug,
cache_dma=cache_dma
)
dest_dir_path = path.abspath(dest_dir_path)
utils.make_dirs(dest_dir_path)
click.echo('import pb file')
if use_onnx:
try:
__import__('onnx')
except ImportError:
raise ImportError('ONNX is required but not installed.')
from frontend.base import BaseIO
from frontend.onnx import OnnxIO
io: BaseIO = OnnxIO()
else:
io = TensorFlowIO()
model: Model = io.read(input_path)
click.echo('optimize graph step: start')
optimize_graph_step(model, config)
click.echo('optimize graph step: done!')
click.echo('generate code step: start')
generate_code_step(model, config)
click.echo(f'generate code step: done!')
def test_tf_import(self) -> None:
"""Test code for importing Tensorflow file with TensorflowIO."""
tf_path = path.join('examples', 'classification', 'lmnet_v1_toy_graph',
'minimal_graph_with_shape.pb')
tf_io = TensorFlowIO()
graph: Graph = tf_io.read(tf_path)
outputs = graph.get_outputs()
self.assertEqual(len(outputs), 1)
self.assertEqual(outputs[0].shape, [1, 10])
print("TF file import test passed!")
def test_import_classification(self) -> None:
"""Test code for importing Tensorflow file with TensorflowIO."""
tf_path = path.join('examples', 'classification',
'lmnet_quantize_cifar10_stride_2.20180523.3x3',
'minimal_graph_with_shape.pb')
tf_io = TensorFlowIO()
graph: Graph = tf_io.read(tf_path)
outputs = graph.get_outputs()
self.assertEqual(len(outputs), 1)
self.assertEqual(outputs[0].shape, [1, 10])
print("TF file import test passed for classification!")
def test_import_group_convolution_classification(self) -> None:
"""Test code for importing Tensorflow file with TensorflowIO."""
tf_path = path.join('examples', 'classification',
'lmnet_v1_group_conv',
'minimal_graph_with_shape.pb')
tf_io = TensorFlowIO()
model = tf_io.read(tf_path)
graph: Graph = model.graph
outputs = graph.get_outputs()
self.assertEqual(len(outputs), 1)
self.assertEqual(outputs[0].shape, [1, 10])
print("TF file import test passed for group convolution!")
def test_tf_export2(self) -> None:
"""Test code for exporting Tensorflow file with TensorflowIO #2."""
tmpdir = 'tmp'
tf_path = path.join(tmpdir, 'test2.pb')
if not path.exists(tmpdir):
makedirs(tmpdir)
elif not path.isdir(tmpdir):
raise ValueError('tmp directory is not a directory.')
model: Model = make_model()
tf_io = TensorFlowIO()
tf_io.write(model, tf_path)
new_model = tf_io.read(tf_path)
self.assertEqual(model.graph, new_model.graph)
print("TF file export test #2 passed!")
def test_import_object_detection_lack_shape_info(self) -> None:
"""Test code for importing tf pb file of object detection
(lack of shape info for some operator) with TensorflowIO.
"""
tf_path = path.join(
'examples', 'object_detection',
'fyolo_quantize_17_v14_some_1x1_wide_pact_add_conv',
'minimal_graph_with_shape.pb')
tf_io = TensorFlowIO()
graph: Graph = tf_io.read(tf_path)
outputs = graph.get_outputs()
self.assertEqual(len(outputs), 1)
self.assertEqual(outputs[0].shape, [1, 10, 10, 125])
print("TF file import test passed for object detection!")
def run(input_path: str,
dest_dir_path: str,
project_name: str,
activate_hard_quantization: bool,
threshold_skipping: bool = False,
num_pe: int = 16,
debug: bool = False,
cache_dma: bool = False):
output_dlk_test_dir = path.join(dest_dir_path, f'{project_name}.test')
optimized_pb_path = path.join(dest_dir_path, f'{project_name}')
optimized_pb_path += '.pb'
output_project_path = path.join(dest_dir_path, f'{project_name}.prj')
config = Config(num_pe=num_pe,
activate_hard_quantization=activate_hard_quantization,
threshold_skipping=threshold_skipping,
test_dir=output_dlk_test_dir,
optimized_pb_path=optimized_pb_path,
output_pj_path=output_project_path,
debug=debug,
cache_dma=cache_dma)
dest_dir_path = path.abspath(dest_dir_path)
utils.make_dirs(dest_dir_path)
click.echo('import pb file')
io = TensorFlowIO()
model: Model = io.read(input_path)
click.echo('optimize graph step: start')
optimize_graph_step(model, config)
click.echo('optimize graph step: done!')
click.echo('generate code step: start')
generate_code_step(model, config)
click.echo(f'generate code step: done!')