def _test_relu_graph(self, X, batch_size, trt_max_batch_size):
node_def = make_node("Relu", ["X"], ["Y"])
Y_c2 = c2.run_node(node_def, {"X": X})
graph_def = make_graph(
[node_def],
name="test",
inputs=[
make_tensor_value_info("X", onnx.TensorProto.FLOAT,
[batch_size, 1, 3, 2])
],
outputs=[
make_tensor_value_info("Y", onnx.TensorProto.FLOAT,
[batch_size, 1, 3, 2])
])
model_def = make_model(graph_def, producer_name='relu-test')
op_outputs = [x.name for x in model_def.graph.output]
op = convert_onnx_model_to_trt_op(model_def,
max_batch_size=trt_max_batch_size)
device_option = core.DeviceOption(caffe2_pb2.CUDA, 0)
op.device_option.CopyFrom(device_option)
Y_trt = None
ws = Workspace()
with core.DeviceScope(device_option):
ws.FeedBlob("X", X)
ws.RunOperatorsOnce([op])
output_values = [ws.FetchBlob(name) for name in op_outputs]
Y_trt = namedtupledict('Outputs', op_outputs)(*output_values)
np.testing.assert_almost_equal(Y_c2, Y_trt)
def _test_onnx_importer(self,
model_name,
data_input_index,
opset_version=onnx.defs.onnx_opset_version()):
model_dir = _download_onnx_model(model_name, opset_version)
model_def = onnx.load(os.path.join(model_dir, 'model.onnx'))
input_blob_dims = [
int(x.dim_value) for x in
model_def.graph.input[data_input_index].type.tensor_type.shape.dim
]
op_inputs = [x.name for x in model_def.graph.input]
op_outputs = [x.name for x in model_def.graph.output]
print("{}".format(op_inputs))
data = np.random.randn(*input_blob_dims).astype(np.float32)
Y_c2 = c2.run_model(model_def, {op_inputs[data_input_index]: data})
op = convert_onnx_model_to_trt_op(model_def, verbosity=3)
device_option = core.DeviceOption(caffe2_pb2.CUDA, 0)
op.device_option.CopyFrom(device_option)
Y_trt = None
ws = Workspace()
with core.DeviceScope(device_option):
ws.FeedBlob(op_inputs[data_input_index], data)
if opset_version >= 5:
# Some newer models from ONNX Zoo come with pre-set "data_0" input
ws.FeedBlob("data_0", data)
ws.RunOperatorsOnce([op])
output_values = [ws.FetchBlob(name) for name in op_outputs]
Y_trt = namedtupledict('Outputs', op_outputs)(*output_values)
np.testing.assert_allclose(Y_c2, Y_trt, rtol=1e-3)
def test_resnet50(self):
input_blob_dims = (1, 3, 224, 224)
model_dir = _download_onnx_model('resnet50')
model_def = onnx.load(os.path.join(model_dir, 'model.onnx'))
op_inputs = [x.name for x in model_def.graph.input]
op_outputs = [x.name for x in model_def.graph.output]
n, c, h, w = input_blob_dims
data = np.random.randn(n, c, h, w).astype(np.float32)
Y_c2 = c2.run_model(model_def, {op_inputs[0]: data})
op = convert_onnx_model_to_trt_op(model_def)
device_option = core.DeviceOption(caffe2_pb2.CUDA, 0)
op.device_option.CopyFrom(device_option)
Y_trt = None
ws = Workspace()
with core.DeviceScope(device_option):
ws.FeedBlob(op_inputs[0], data)
ws.RunOperatorsOnce([op])
output_values = [ws.FetchBlob(name) for name in op_outputs]
Y_trt = namedtupledict('Outputs', op_outputs)(*output_values)
np.testing.assert_allclose(Y_c2, Y_trt, rtol=1e-3)
def _test_onnx_importer(self, model_name, data_input_index = 0):
model_dir = _download_onnx_model(model_name)
model_def = onnx.load(os.path.join(model_dir, 'model.onnx'))
input_blob_dims = [int(x.dim_value) for x in model_def.graph.input[data_input_index].type.tensor_type.shape.dim]
op_inputs = [x.name for x in model_def.graph.input]
op_outputs = [x.name for x in model_def.graph.output]
print("{}".format(op_inputs))
data = np.random.randn(*input_blob_dims).astype(np.float32)
Y_c2 = c2.run_model(model_def, {op_inputs[data_input_index]: data})
op = convert_onnx_model_to_trt_op(model_def, verbosity=3)
device_option = core.DeviceOption(caffe2_pb2.CUDA, 0)
op.device_option.CopyFrom(device_option)
Y_trt = None
ws = Workspace()
with core.DeviceScope(device_option):
ws.FeedBlob(op_inputs[data_input_index], data)
ws.RunOperatorsOnce([op])
output_values = [ws.FetchBlob(name) for name in op_outputs]
Y_trt = namedtupledict('Outputs', op_outputs)(*output_values)
np.testing.assert_allclose(Y_c2, Y_trt, rtol=1e-3)
def _test_relu_graph(self, X, batch_size, trt_max_batch_size):
node_def = make_node("Relu", ["X"], ["Y"])
Y_c2 = c2.run_node(node_def, {"X": X})
graph_def = make_graph(
[node_def],
name="test",
inputs=[make_tensor_value_info("X", onnx.TensorProto.FLOAT, [batch_size, 1, 3, 2])],
outputs=[make_tensor_value_info("Y", onnx.TensorProto.FLOAT, [batch_size, 1, 3, 2])])
model_def = make_model(graph_def, producer_name='relu-test')
op_outputs = [x.name for x in model_def.graph.output]
op = convert_onnx_model_to_trt_op(model_def, max_batch_size=trt_max_batch_size)
device_option = core.DeviceOption(caffe2_pb2.CUDA, 0)
op.device_option.CopyFrom(device_option)
Y_trt = None
ws = Workspace()
with core.DeviceScope(device_option):
ws.FeedBlob("X", X)
ws.RunOperatorsOnce([op])
output_values = [ws.FetchBlob(name) for name in op_outputs]
Y_trt = namedtupledict('Outputs', op_outputs)(*output_values)
np.testing.assert_almost_equal(Y_c2, Y_trt)
