def test_nn_functional_interpolate_nearest_date_format():
"""
api: paddle.nn.functional.interpolate
op version: 11
"""
op = Net(size=[4, 12], data_format='NHWC')
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_functional_interpolate', [11])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [2, 2, 2, 10]).astype('float32')))
obj.run()
def test_nn_functional_interpolate_bicubic_align_corners():
"""
api: paddle.nn.functional.interpolate
op version: 11
"""
op = Net(mode='bicubic', scale_factor=1.5, align_corners=True)
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_functional_interpolate', [11])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [1, 2, 2, 5]).astype('float32')))
obj.run()
def test_AdaptiveAvgPool1D_base():
"""
api: paddle.nn.AdaptiveAvgPool1D
op version: 9, 10, 11, 12
"""
op = Net()
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_AdaptiveAvgPool1D', [9, 10, 11, 12])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 10]).astype('float32')))
obj.run()
def test_LayerNorm_11():
"""
api: paddle.nn.LayerNorm
op version: 11
"""
op = Net()
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_LayerNorm', [11])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 10, 10]).astype('float32')))
obj.run()
def test_MaxPool2D_base_Padding_0():
"""
api: paddle.MaxPool2D
op version: 9, 10, 11, 12
"""
op = Net(kernel_size=5, padding=[[0, 0], [0, 0], [1, 2], [3, 4]])
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_MaxPool2D', [9, 10, 11, 12])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 10, 10]).astype('float32')))
obj.run()
def test_transpose_10():
"""
api: paddle.transpose
op version: 10
"""
op = Net()
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'transpose', [10])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [1, 3, 10]).astype('float32')))
obj.run()
def test_Pad1D_replicate():
"""
api: paddle.nn.Pad1D
op version: 12
"""
op = Net(mode='replicate')
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_Pad1D', [12])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 10]).astype('float32')))
obj.run()
def test_Conv2D_padding_2_9():
"""
api: paddle.nn.Conv2D
op version: 9
"""
op = Net(padding=[[0, 0], [0, 0], [1, 2], [2, 3]])
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'Conv2D', [9])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 10, 10]).astype('float32')))
obj.run()
def test_InstanceNorm_base():
"""
api: paddle.InstanceNorm
op version: 9, 10, 11, 12
"""
op = Net()
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_InstanceNorm', [9, 10, 11, 12])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [2, 2, 2, 2, 3]).astype('float32')))
obj.run()
def test_avg_pool2d_12():
"""
api: paddle.nn.functional.avg_pool2d
op version: 12
"""
op = Net()
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'avg_pool2d', [12])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 10, 10]).astype('float32')))
obj.run()
def test_nn_functional_selu_scale():
"""
api: paddle.nn.functional.selu
op version: 12
"""
op = Net(scale=2)
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn.functional.selu', [12])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 3, 3]).astype('float32')))
obj.run()
def test_squeeze_9_None():
"""
api: paddle.squeeze
op version: 12
"""
op = Net(axis=None)
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'squeeze', [9, 10, 11, 12, 13])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [1, 3, 10, 1]).astype('float32')))
obj.run()
def test_Conv3DTranspose_padding_1_9_10_11_12():
"""
api: paddle.Conv3DTranspose
op version: 9,10,11,12
"""
op = Net(padding=[1, 2, 3, 4, 5, 6])
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_Conv3DTranspose', [9, 10, 11, 12])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 10, 10, 10]).astype('float32')))
obj.run()
def test_Conv3DTranspose_output_padding_2_9_10_11_12():
"""
api: paddle.Conv3DTranspose
op version: 9,10,11,12
"""
op = Net(in_channels=16, out_channels=16, stride=3, output_padding=2)
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_Conv3DTranspose', [9, 10, 11, 12])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 16, 10, 10, 10]).astype('float32')))
obj.run()
def test_Hardsigmoid_10():
"""
api: paddle.nn.Hardsigmoid
op version: 10
"""
op = Net()
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_Hardsigmoid', [10])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 10, 10]).astype('float32')))
obj.run()
def test_GroupNorm_12():
"""
api: paddle.nn.GroupNorm
op version: 12
"""
op = Net()
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_GroupNorm', [12])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [5, 10, 8, 8]).astype('float32')))
obj.run()
def test_BatchNorm2D_9():
"""
api: paddle.nn.BatchNorm2D
op version: 9
"""
op = Net()
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_BatchNorm2D', [9])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 10, 10]).astype('float32')))
obj.run()
def test_Conv2D_dilation_2_9():
"""
api: paddle.nn.Conv2D
op version: 9
"""
op = Net(in_channels=16, out_channels=16, dilation=3)
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'Conv2D', [9])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 16, 10, 10]).astype('float32')))
obj.run()
def test_Conv1DTranspose_12_Padding_tuple1():
"""
api: paddle.nn.Conv1DTranspose
op version: 12
"""
op = Net(padding=(1, 2))
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_Conv1DTranspose', [9, 10, 11, 12, 13])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 10]).astype('float32')))
obj.run()
def test_initializer_Uniform_base():
"""
api: paddle.initializer.Uniform
op version: 9
"""
op = Net()
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_initializer_Uniform', [9, 10, 11, 12])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 2]).astype('float32')))
obj.run()
def test_Conv3D_10():
"""
api: paddle.Conv3D
op version: 10
"""
op = Net()
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'Conv2D_Dropout', [10])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 5, 10, 10]).astype('float32')))
obj.run()
def test_square_9():
"""
api: paddle.square
op version: 9
"""
op = Net()
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'square', [9])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 3, 3]).astype('float32')))
obj.run()
def test_argmin_axis_keepdim():
"""
api: paddle.argmin
op version: 12
"""
op = Net(axis=1, keepdim=True)
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'argmin', [12])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 3, 10]).astype('float32')))
obj.run()
def test_neg_12():
"""
api: paddle.neg
op version: 12
"""
op = Net()
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'neg', [12])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 3, 3]).astype('float32')))
obj.run()
def test_nn_functional_interpolate_nearest_scale_factor_tuple():
"""
api: paddle.nn.functional.interpolate
op version: 11
"""
op = Net(scale_factor=(1, 2))
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_functional_interpolate', [11])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [2, 3, 6, 10]).astype('float32')))
obj.run()
def test_argsort_descending_1():
"""
api: paddle.argsort
op version: 12
"""
op = Net(descending=True)
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'argsort', [7])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 3, 10]).astype('float32')))
obj.run()
def test_Conv2DTranspose_9_output_padding_3():
"""
api: paddle.Conv2DTranspose
op version: 9
"""
op = Net(output_padding=1,
stride=[3, 2],
padding=[[0, 0], [0, 0], [1, 2], [2, 3]])
op.eval()
# net, name, ver_list, delta=1e-6, rtol=1e-5
obj = APIOnnx(op, 'nn_Conv2DTranspose', [9])
obj.set_input_data(
"input_data",
paddle.to_tensor(
randtool("float", -1, 1, [3, 1, 10, 10]).astype('float32')))
obj.run()
def generator_data():
input_data = randtool("int", 0, 10, input2_shape)
return input_data
def run_test(self, configs):
config, models = configs
logging.info("Run configs: {}".format(config))
assert "op_names" in config.keys(
), "config must include op_names in dict keys"
assert "test_data_shapes" in config.keys(
), "config must include test_data_shapes in dict keys"
assert "test_data_types" in config.keys(
), "config must include test_data_types in dict keys"
assert "opset_version" in config.keys(
), "config must include opset_version in dict keys"
assert "input_spec_shape" in config.keys(
), "config must include input_spec_shape in dict keys"
op_names = config["op_names"]
test_data_shapes = config["test_data_shapes"]
test_data_types = config["test_data_types"]
opset_version = config["opset_version"]
input_specs = config["input_spec_shape"]
use_gpu = True
if "use_gpu" in config.keys():
use_gpu = config["use_gpu"]
self.num_ran_models += 1
if not isinstance(models, (tuple, list)):
models = [models]
if not isinstance(op_names, (tuple, list)):
op_names = [op_names]
if not isinstance(opset_version[0], (tuple, list)):
opset_version = [opset_version]
if len(opset_version) == 1 and len(models) != len(opset_version):
opset_version = opset_version * len(models)
assert len(models) == len(
op_names), "Length of models should be equal to length of op_names"
input_type_list = None
if len(test_data_types) > 1:
input_type_list = list(product(*test_data_types))
elif len(test_data_types) == 1:
if isinstance(test_data_types[0], str):
input_type_list = [test_data_types[0]]
else:
input_type_list = test_data_types
elif len(test_data_types) == 0:
input_type_list = [["float32"] * len(test_data_shapes)]
delta = 1e-5
rtol = 1e-5
if "delta" in config.keys():
delta = config["delta"]
if "rtol" in config.keys():
rtol = config["rtol"]
for i, model in enumerate(models):
model.eval()
obj = APIOnnx(model, op_names[i], opset_version[i], op_names[i],
input_specs, delta, rtol, use_gpu)
for input_type in input_type_list:
input_tensors = list()
for j, shape in enumerate(test_data_shapes):
# Determine whether it is a user-defined data generation function
if isfunction(shape):
data = shape()
data = data.astype(input_type[j])
input_tensors.append(paddle.to_tensor(data))
continue
if input_type[j].count('int') > 0:
input_tensors.append(
paddle.to_tensor(
randtool("int", -20, 20,
shape).astype(input_type[j])))
elif input_type[j].count('bool') > 0:
input_tensors.append(
paddle.to_tensor(
randtool("bool", -2, 2,
shape).astype(input_type[j])))
else:
input_tensors.append(
paddle.to_tensor(
randtool("float", -2, 2,
shape).astype(input_type[j])))
obj.set_input_data("input_data", tuple(input_tensors))
logging.info("Now Run >>> dtype: {}, op_name: {}".format(
input_type, op_names[i]))
obj.run()
if len(input_type_list) == 0:
obj.run()
logging.info("Run Successfully!")
def generator3_data():
input_data3 = randtool("int", 1, 10, [input_shape[2]])
return input_data3
