forked from mindspore-Ecosystem/mindspore
187 lines
8.3 KiB
Python
187 lines
8.3 KiB
Python
# Copyright 2022 Huawei Technologies Co., Ltd
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
# ============================================================================
|
|
|
|
import numpy as np
|
|
import pytest
|
|
|
|
import mindspore.context as context
|
|
import mindspore.nn as nn
|
|
from mindspore import Tensor
|
|
from mindspore.ops import composite as C
|
|
from mindspore.ops import operations as P
|
|
from mindspore.ops import functional as F
|
|
|
|
|
|
class MishNet(nn.Cell):
|
|
def __init__(self):
|
|
super(MishNet, self).__init__()
|
|
self.mish = P.Mish()
|
|
|
|
def construct(self, x):
|
|
return self.mish(x)
|
|
|
|
|
|
class MishGradNet(nn.Cell):
|
|
def __init__(self, network):
|
|
super(MishGradNet, self).__init__()
|
|
self.grad = C.GradOperation(get_all=True, sens_param=True)
|
|
self.network = network
|
|
|
|
def construct(self, x, dy):
|
|
gout = self.grad(self.network)(x, dy)
|
|
return gout
|
|
|
|
|
|
class MishGradDynamicShapeNet(nn.Cell):
|
|
def __init__(self, network):
|
|
super(MishGradDynamicShapeNet, self).__init__()
|
|
self.unique = P.Unique()
|
|
self.reshape = P.Reshape()
|
|
self.grad = C.GradOperation(get_all=True, sens_param=True)
|
|
self.network = network
|
|
|
|
def construct(self, x, dy):
|
|
x_unique, _ = self.unique(x)
|
|
x_unique = self.reshape(x_unique, (3, 3))
|
|
dy_unique, _ = self.unique(dy)
|
|
dy_unique = self.reshape(dy_unique, (3, 3))
|
|
return self.grad(self.network)(x_unique, dy_unique)
|
|
|
|
|
|
@pytest.mark.level0
|
|
@pytest.mark.platform_x86_cpu
|
|
@pytest.mark.env_onecard
|
|
@pytest.mark.parametrize('mode', [context.GRAPH_MODE, context.PYNATIVE_MODE])
|
|
@pytest.mark.parametrize('dtype, tol', [(np.float16, 1e-3), (np.float32, 1e-4)])
|
|
def test_mish_grad(mode, dtype, tol):
|
|
"""
|
|
Feature: ALL To ALL
|
|
Description: test cases for MishGrad
|
|
Expectation: the result match to the expect
|
|
"""
|
|
context.set_context(mode=context.GRAPH_MODE, device_target="CPU")
|
|
x = Tensor(np.array([[[[1.7641, 0.4002, 0.9787],
|
|
[2.2409, 1.8676, -0.9773]],
|
|
[[0.9501, -0.1514, -0.1032],
|
|
[0.4106, 0.1440, 1.4543]]],
|
|
[[[0.7610, 0.1217, 0.4439],
|
|
[0.3337, 1.4941, -0.2052]],
|
|
[[0.3131, -0.8541, -2.5530],
|
|
[0.6536, 0.8644, -0.7422]]]]).astype(dtype))
|
|
dy = Tensor(np.array([[[[2.2698, -1.4544, 0.0458],
|
|
[-0.1872, 1.5328, 1.4694]],
|
|
[[0.1549, 0.3782, -0.8878],
|
|
[-1.9808, -0.3479, 0.1563]]],
|
|
[[[1.2303, 1.2024, -0.3873],
|
|
[-0.3023, -1.0486, -1.4200]],
|
|
[[-1.7063, 1.9508, -0.5097],
|
|
[-0.4381, -1.2528, 0.7775]]]]).astype(dtype))
|
|
expect = np.array([[[[2.4551, -1.2174, 0.0478],
|
|
[-0.1975, 1.6503, 0.0989]],
|
|
[[0.1610, 0.1901, -0.4737],
|
|
[-1.6688, -0.2403, 0.1702]]],
|
|
[[[1.2171, 0.8138, -0.3328],
|
|
[-0.2423, -1.1413, -0.6649]],
|
|
[[-1.3482, 0.2244, 0.0553],
|
|
[-0.4169, -1.2767, 0.1278]]]]).astype(dtype)
|
|
net = MishNet()
|
|
grad = MishGradNet(net)
|
|
output = grad(x, dy)
|
|
assert np.allclose(output[0].asnumpy(), expect, atol=tol, rtol=tol, equal_nan=True)
|
|
|
|
|
|
@pytest.mark.level0
|
|
@pytest.mark.platform_x86_gpu_training
|
|
@pytest.mark.env_onecard
|
|
@pytest.mark.parametrize('mode', [context.GRAPH_MODE, context.PYNATIVE_MODE])
|
|
def test_mish_grad_vmap(mode):
|
|
"""
|
|
Feature: test mish_grad vmap feature.
|
|
Description: test mish_grad vmap feature.
|
|
Expectation: Success.
|
|
"""
|
|
context.set_context(mode=mode, device_target="CPU")
|
|
x = Tensor(np.array([[[[1.7641, 0.4002, 0.9787],
|
|
[2.2409, 1.8676, -0.9773]],
|
|
[[0.9501, -0.1514, -0.1032],
|
|
[0.4106, 0.1440, 1.4543]]],
|
|
[[[0.7610, 0.1217, 0.4439],
|
|
[0.3337, 1.4941, -0.2052]],
|
|
[[0.3131, -0.8541, -2.5530],
|
|
[0.6536, 0.8644, -0.7422]]]]).astype(np.float32))
|
|
dout = Tensor(np.array([[[[2.2698, -1.4544, 0.0458],
|
|
[-0.1872, 1.5328, 1.4694]],
|
|
[[0.1549, 0.3782, -0.8878],
|
|
[-1.9808, -0.3479, 0.1563]]],
|
|
[[[1.2303, 1.2024, -0.3873],
|
|
[-0.3023, -1.0486, -1.4200]],
|
|
[[-1.7063, 1.9508, -0.5097],
|
|
[-0.4381, -1.2528, 0.7775]]]]).astype(np.float32))
|
|
# Case 1
|
|
output = F.vmap(MishGradNet(MishNet()), (0, 0), 0)(x, dout)
|
|
expect_output = np.array([[[[2.4551497, -1.2175097, 0.0478603],
|
|
[-0.1975334, 1.6502883, 0.09884691]],
|
|
[[0.16096735, 0.19009684, -0.47376704],
|
|
[-1.6688112, -0.24026634, 0.17010784]]],
|
|
[[[1.2171272, 0.81384104, -0.33282074],
|
|
[-0.24231759, -1.1413976, -0.6648671]],
|
|
[[-1.3482722, 0.22441024, 0.05531986],
|
|
[-0.41696107, -1.2767013, 0.1277946]]]]).astype(np.float32)
|
|
assert np.allclose(output[0].asnumpy(), expect_output, atol=1e-4, rtol=1e-4, equal_nan=True)
|
|
|
|
# # Case 2
|
|
output = F.vmap(MishGradNet(MishNet()), (0, 1), 0)(x, dout)
|
|
expect_output = np.array([[[[2.4551497, -1.2175097, 0.0478603],
|
|
[-0.1975334, 1.6502883, 0.09884691]],
|
|
[[1.2784901, 0.6043692, -0.20667942],
|
|
[-0.2546858, -0.724183, -1.5454454]]],
|
|
[[[0.15324152, 0.2559836, -0.7629183],
|
|
[-1.5877694, -0.378688, 0.0731822]],
|
|
[[-1.3482722, 0.22441024, 0.05531986],
|
|
[-0.41696107, -1.2767013, 0.1277946]]]]).astype(np.float32)
|
|
assert np.allclose(output[0].asnumpy(), expect_output, atol=1e-4, rtol=1e-4, equal_nan=True)
|
|
|
|
# # Case 3
|
|
output = F.vmap(MishGradNet(MishNet()), (0, 0), 1)(x, dout)
|
|
expect_output = np.array([[[[2.4551497, -1.2175097, 0.0478603],
|
|
[-0.1975334, 1.6502883, 0.09884691]],
|
|
[[1.2171272, 0.81384104, -0.33282074],
|
|
[-0.24231759, -1.1413976, -0.6648671]]],
|
|
[[[0.16096735, 0.19009684, -0.47376704],
|
|
[-1.6688112, -0.24026634, 0.17010784]],
|
|
[[-1.3482722, 0.22441024, 0.05531986],
|
|
[-0.41696107, -1.2767013, 0.1277946]]]]).astype(np.float32)
|
|
assert np.allclose(output[0].asnumpy(), expect_output, atol=1e-4, rtol=1e-4, equal_nan=True)
|
|
|
|
|
|
@pytest.mark.level0
|
|
@pytest.mark.platform_x86_cpu
|
|
@pytest.mark.env_onecard
|
|
@pytest.mark.parametrize('mode', [context.GRAPH_MODE, context.PYNATIVE_MODE])
|
|
def test_mish_grad_dynamic_shape(mode):
|
|
"""
|
|
Feature: test mish_grad dynamic_shape feature.
|
|
Description: test mish_grad dynamic_shape feature.
|
|
Expectation: Success.
|
|
"""
|
|
context.set_context(mode=mode, device_target="CPU")
|
|
x = Tensor(np.array([8., -3., 0., 0., 10., 1., 21., -3., 11., 4., -2., 10., 8.]).astype(np.float32))
|
|
dout = Tensor(np.array([18., -1.3, 0., 0., 12., 1., 2.1, -1.3, 11., 4.2, -2., 12., 18.]).astype(np.float32))
|
|
output = MishGradDynamicShapeNet(MishNet())(x, dout)
|
|
expect_output = np.array([[18.000065, 0.12141099, 0.],
|
|
[12., 1.0490363, 2.1],
|
|
[11., 4.218617, 0.2167282]], dtype=np.float32)
|
|
assert np.allclose(output[0].asnumpy(), expect_output, atol=1e-4, rtol=1e-4, equal_nan=True)
|