169 lines
5.6 KiB
Python
169 lines
5.6 KiB
Python
# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# 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.
|
|
|
|
# [AUTO-GENERATED] Unit test for paddle.nn.clip
|
|
# 自动生成的单测,覆盖 paddle.nn.clip 模块中未覆盖的代码
|
|
|
|
"""
|
|
测试模块:paddle.nn.clip (ClipGradByValue, ClipGradByNorm, ClipGradByGlobalNorm, clip_by_norm)
|
|
Test Module: paddle.nn.clip
|
|
|
|
本测试覆盖以下功能:
|
|
This test covers the following functions:
|
|
1. clip_by_norm - 按范数裁剪 / Clip by norm in dynamic mode
|
|
2. ClipGradByValue - 按值裁剪梯度 / Clip gradient by value
|
|
3. ClipGradByNorm - 按范数裁剪梯度 / Clip gradient by norm
|
|
4. ClipGradByGlobalNorm - 按全局范数裁剪梯度 / Clip gradient by global norm
|
|
|
|
覆盖的未覆盖行:88-110 (clip_by_norm static), 143-151 (merge_selected_rows)
|
|
"""
|
|
|
|
import unittest
|
|
|
|
import numpy as np
|
|
|
|
import paddle
|
|
|
|
|
|
class TestClipByNorm(unittest.TestCase):
|
|
"""测试clip_by_norm功能
|
|
Test clip_by_norm function"""
|
|
|
|
def setUp(self):
|
|
paddle.disable_static()
|
|
|
|
def test_clip_by_norm_no_clip(self):
|
|
"""范数小于max_norm时不裁剪 / No clipping when norm < max_norm"""
|
|
x = paddle.to_tensor([[0.1, 0.2], [0.3, 0.4]], dtype='float32')
|
|
out = paddle.nn.clip.clip_by_norm(x, max_norm=10.0)
|
|
np.testing.assert_allclose(out.numpy(), x.numpy(), rtol=1e-5)
|
|
|
|
def test_clip_by_norm_with_clip(self):
|
|
"""范数大于max_norm时裁剪 / Clipping when norm > max_norm"""
|
|
x = paddle.to_tensor([[3.0, 4.0]], dtype='float32')
|
|
out = paddle.nn.clip.clip_by_norm(x, max_norm=1.0)
|
|
norm = np.linalg.norm(out.numpy())
|
|
self.assertAlmostEqual(norm, 1.0, places=5)
|
|
|
|
def test_clip_by_norm_zero(self):
|
|
"""零tensor裁剪 / Clip zero tensor"""
|
|
x = paddle.zeros([3, 3], dtype='float32')
|
|
out = paddle.nn.clip.clip_by_norm(x, max_norm=1.0)
|
|
np.testing.assert_allclose(out.numpy(), np.zeros([3, 3]), atol=1e-7)
|
|
|
|
|
|
class TestClipGradByValue(unittest.TestCase):
|
|
"""测试ClipGradByValue梯度裁剪
|
|
Test ClipGradByValue gradient clipping"""
|
|
|
|
def setUp(self):
|
|
paddle.disable_static()
|
|
|
|
def test_clip_grad_by_value_basic(self):
|
|
"""基本按值裁剪 / Basic clip by value"""
|
|
clip = paddle.nn.ClipGradByValue(min=-0.5, max=0.5)
|
|
linear = paddle.nn.Linear(3, 3)
|
|
x = paddle.randn([2, 3])
|
|
y = linear(x)
|
|
loss = y.sum()
|
|
loss.backward()
|
|
|
|
params_grads = []
|
|
for p in linear.parameters():
|
|
if p.grad is not None:
|
|
params_grads.append((p, p.grad))
|
|
|
|
clipped = clip(params_grads)
|
|
for p, g in clipped:
|
|
self.assertTrue(g.numpy().max() <= 0.5 + 1e-6)
|
|
self.assertTrue(g.numpy().min() >= -0.5 - 1e-6)
|
|
|
|
def test_clip_grad_by_value_symmetric(self):
|
|
"""对称裁剪 / Symmetric clipping with only max"""
|
|
clip = paddle.nn.ClipGradByValue(max=1.0)
|
|
self.assertIsNotNone(clip)
|
|
|
|
|
|
class TestClipGradByNorm(unittest.TestCase):
|
|
"""测试ClipGradByNorm梯度裁剪
|
|
Test ClipGradByNorm gradient clipping"""
|
|
|
|
def setUp(self):
|
|
paddle.disable_static()
|
|
|
|
def test_clip_grad_by_norm_basic(self):
|
|
"""基本按范数裁剪 / Basic clip by norm"""
|
|
clip = paddle.nn.ClipGradByNorm(clip_norm=1.0)
|
|
linear = paddle.nn.Linear(3, 3)
|
|
x = paddle.randn([2, 3])
|
|
y = linear(x)
|
|
loss = y.sum() * 100
|
|
loss.backward()
|
|
|
|
params_grads = []
|
|
for p in linear.parameters():
|
|
if p.grad is not None:
|
|
params_grads.append((p, p.grad))
|
|
|
|
clipped = clip(params_grads)
|
|
for p, g in clipped:
|
|
norm = float(paddle.linalg.norm(g).numpy())
|
|
self.assertLessEqual(norm, 1.0 + 1e-5)
|
|
|
|
|
|
class TestClipGradByGlobalNorm(unittest.TestCase):
|
|
"""测试ClipGradByGlobalNorm全局范数裁剪
|
|
Test ClipGradByGlobalNorm global norm clipping"""
|
|
|
|
def setUp(self):
|
|
paddle.disable_static()
|
|
|
|
def test_clip_grad_by_global_norm_basic(self):
|
|
"""基本全局范数裁剪 / Basic global norm clipping"""
|
|
clip = paddle.nn.ClipGradByGlobalNorm(clip_norm=1.0)
|
|
linear = paddle.nn.Linear(10, 10)
|
|
x = paddle.randn([2, 10])
|
|
y = linear(x)
|
|
loss = y.sum() * 100
|
|
loss.backward()
|
|
|
|
params_grads = []
|
|
for p in linear.parameters():
|
|
if p.grad is not None:
|
|
params_grads.append((p, p.grad))
|
|
|
|
clipped = clip(params_grads)
|
|
self.assertIsNotNone(clipped)
|
|
|
|
def test_clip_grad_by_global_norm_no_clip(self):
|
|
"""全局范数未超限不裁剪 / No clipping when global norm is within limit"""
|
|
clip = paddle.nn.ClipGradByGlobalNorm(clip_norm=1000.0)
|
|
linear = paddle.nn.Linear(3, 3)
|
|
x = paddle.randn([2, 3])
|
|
y = linear(x)
|
|
loss = y.sum()
|
|
loss.backward()
|
|
|
|
params_grads = []
|
|
for p in linear.parameters():
|
|
if p.grad is not None:
|
|
params_grads.append((p, p.grad.clone()))
|
|
|
|
clipped = clip(params_grads)
|
|
self.assertIsNotNone(clipped)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|