forked from mindspore-Ecosystem/mindspore
231 lines
9.5 KiB
Python
231 lines
9.5 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.common import dtype as mstype
|
|
from mindspore.ops import operations as P
|
|
from mindspore.common.parameter import Parameter
|
|
from mindspore.nn import layer
|
|
from mindspore.ops import functional as F
|
|
from mindspore.ops import composite as C
|
|
from mindspore.ops.operations import _inner_ops as inner
|
|
|
|
num_one = Tensor(np.ones([1]), mstype.float32)
|
|
data_type = np.float32
|
|
param_val = np.array([[1, 1, 2], [3, 4, 5], [5, 6, 7]]).astype(data_type)
|
|
m_val = np.array([[1, 1, 2], [3, 4, 5], [5, 6, 7]]).astype(data_type)
|
|
v_val = np.array([[1, 1, 2], [3, 4, 5], [5, 6, 7]]).astype(data_type)
|
|
grad_val = np.array([[1, 1, 2], [3, 4, 5], [5, 6, 7]]).astype(data_type)
|
|
|
|
beta1_val = Tensor(np.array([0.9]).astype(data_type))
|
|
beta2_val = Tensor(np.array([0.999]).astype(data_type))
|
|
eps_val = Tensor(np.array([1e-6]).astype(data_type))
|
|
global_step_val = Tensor(np.array([2]).astype(np.int32))
|
|
lr_val = Tensor(np.array([0.01]).astype(data_type))
|
|
weight_decay_val = Tensor(np.array([0.0]).astype(data_type))
|
|
|
|
|
|
class LambGPUOrigin(nn.Cell):
|
|
def __init__(self, param, m, v, gradient):
|
|
super().__init__()
|
|
self.param = Parameter(param, name="param")
|
|
self.m = Parameter(m, name="m")
|
|
self.v = Parameter(v, name="v")
|
|
self.gradient = Parameter(gradient, name="grad")
|
|
self.op_mul = P.Mul()
|
|
self.op_sqrt = P.Sqrt()
|
|
self.op_rsqrt = P.Rsqrt()
|
|
self.op_square = P.Square()
|
|
self.op_cast = P.Cast()
|
|
self.op_reshape = P.Reshape()
|
|
self.op_shape = P.Shape()
|
|
self.op_pow = P.Pow()
|
|
self.op_norm = layer.Norm()
|
|
self.op_select = P.Select()
|
|
self.op_greater = P.Greater()
|
|
self.op_fill = P.Fill()
|
|
self.op_dtype = P.DType()
|
|
|
|
def construct(self, beta1, beta2, eps, global_step, lr, weight_decay, decay_flag):
|
|
param_fp32 = self.op_cast(self.param, mstype.float32)
|
|
m_fp32 = self.op_cast(self.m, mstype.float32)
|
|
v_fp32 = self.op_cast(self.v, mstype.float32)
|
|
gradient_fp32 = self.op_cast(self.gradient, mstype.float32)
|
|
|
|
next_m = self.op_mul(beta1, m_fp32) + self.op_mul(self.op_cast(num_one, mstype.float32) - beta1, gradient_fp32)
|
|
|
|
next_v = self.op_mul(beta2, v_fp32) + self.op_mul(self.op_cast(num_one, mstype.float32) - beta2,
|
|
self.op_square(gradient_fp32))
|
|
|
|
next_mm = next_m / (self.op_cast(num_one, mstype.float32) -
|
|
self.op_pow(beta1, self.op_cast(global_step, mstype.float32)))
|
|
next_vv = next_v / (self.op_cast(num_one, mstype.float32) -
|
|
self.op_pow(beta2, self.op_cast(global_step, mstype.float32)))
|
|
w_norm = self.op_norm(param_fp32)
|
|
g_norm = self.op_norm(gradient_fp32)
|
|
|
|
g_norm_hat = self.op_norm(self.op_mul(next_mm, self.op_rsqrt(next_vv + eps)) + weight_decay * param_fp32)
|
|
zeros = F.zeros_like(w_norm)
|
|
ones = self.op_fill(self.op_dtype(w_norm), self.op_shape(w_norm), 1.0)
|
|
trust_ratio = self.op_select(
|
|
self.op_greater(w_norm, zeros),
|
|
self.op_select(self.op_greater(g_norm, zeros), w_norm / g_norm_hat, ones),
|
|
ones)
|
|
tens = self.op_fill(self.op_dtype(trust_ratio), self.op_shape(trust_ratio), 10.0)
|
|
trust_ratio = C.clip_by_value(trust_ratio, zeros, tens)
|
|
update = next_mm / (self.op_sqrt(next_vv) + eps)
|
|
|
|
if decay_flag:
|
|
update = update + self.op_mul(weight_decay, param_fp32)
|
|
|
|
update_with_lr = self.op_mul(self.op_mul(trust_ratio, lr), update)
|
|
|
|
next_param = param_fp32 - self.op_reshape(update_with_lr, self.op_shape(param_fp32))
|
|
|
|
next_param = F.depend(next_param, F.assign(self.param, self.op_cast(next_param, F.dtype(self.param))))
|
|
next_param = F.depend(next_param, F.assign(self.m, self.op_cast(next_m, F.dtype(self.m))))
|
|
next_param = F.depend(next_param, F.assign(self.v, self.op_cast(next_v, F.dtype(self.v))))
|
|
|
|
return self.op_cast(next_param, F.dtype(self.param))
|
|
|
|
|
|
class LambAscendOrigin(nn.Cell):
|
|
def __init__(self, param, m, v, gradient):
|
|
super().__init__()
|
|
self.param = Parameter(param, name="param")
|
|
self.m = Parameter(m, name="m")
|
|
self.v = Parameter(v, name="v")
|
|
self.gradient = Parameter(gradient, name="grad")
|
|
self.op_lamb_apply_optimizer_assign = P.LambApplyOptimizerAssign()
|
|
self.op_lamb_apply_weight_assign = P.LambApplyWeightAssign()
|
|
self.op_cast = P.Cast()
|
|
self.op_norm = layer.Norm()
|
|
|
|
def construct(self, beta1, beta2, eps, global_step, lr, weight_decay, decay_flag):
|
|
param_fp32 = self.op_cast(self.param, mstype.float32)
|
|
gradient_fp32 = self.op_cast(self.gradient, mstype.float32)
|
|
new_global_step = self.op_cast(global_step, mstype.float32)
|
|
weight_decay_flag = self.op_cast(decay_flag, mstype.float32)
|
|
|
|
update, _, _ = self.op_lamb_apply_optimizer_assign(gradient_fp32, self.v, self.m, param_fp32,
|
|
beta1, 1.0 - beta1, beta2, 1.0 - beta2, eps,
|
|
new_global_step, weight_decay_flag, weight_decay)
|
|
w_norm = self.op_norm(param_fp32)
|
|
g_norm = self.op_norm(update)
|
|
|
|
update = F.depend(update, self.op_lamb_apply_weight_assign(w_norm, g_norm, lr, update, self.param))
|
|
return self.param
|
|
|
|
|
|
class MyLamb(nn.Cell):
|
|
def __init__(self, param, m, v, gradient):
|
|
super().__init__()
|
|
self.param = Parameter(param, name="param")
|
|
self.m = Parameter(m, name="m")
|
|
self.v = Parameter(v, name="v")
|
|
self.gradient = Parameter(gradient, name="grad")
|
|
self.lamb = inner.Lamb()
|
|
|
|
def construct(self, beta1, beta2, eps, global_step, lr, weight_decay):
|
|
return self.lamb(self.param, self.m, self.v, lr, beta1, beta2, eps, weight_decay, global_step, self.gradient)
|
|
|
|
|
|
def test_gpu_net():
|
|
"""
|
|
Feature: gpu testcase for Lamb
|
|
Description: fixed input when using gpu
|
|
Expectation: get the same result when use new lamb kernel and old kernel
|
|
"""
|
|
my_lamb = MyLamb(param_val, m_val, v_val, grad_val)
|
|
my_lamb(beta1_val, beta2_val, eps_val, global_step_val, lr_val, weight_decay_val)
|
|
|
|
lamb_gpu_origin = LambGPUOrigin(param_val, m_val, v_val, grad_val)
|
|
lamb_gpu_origin(beta1_val, beta2_val, eps_val, global_step_val, lr_val, weight_decay_val, True)
|
|
|
|
assert np.allclose(my_lamb.param.asnumpy(), lamb_gpu_origin.param.asnumpy())
|
|
|
|
|
|
def test_ascend_net():
|
|
"""
|
|
Feature: ascend testcase for Lamb
|
|
Description: fixed input when using ascend
|
|
Expectation: get the same result when use new lamb kernel and old kernel
|
|
"""
|
|
my_lamb = MyLamb(param_val, m_val, v_val, grad_val)
|
|
my_lamb(beta1_val, beta2_val, eps_val, global_step_val, lr_val, weight_decay_val)
|
|
|
|
lamb_ascend_origin = LambAscendOrigin(param_val, m_val, v_val, grad_val)
|
|
lamb_ascend_origin(beta1_val, beta2_val, eps_val, global_step_val, lr_val, weight_decay_val, True)
|
|
|
|
assert np.allclose(my_lamb.param.asnumpy(), lamb_ascend_origin.param.asnumpy())
|
|
|
|
|
|
@pytest.mark.level0
|
|
@pytest.mark.platform_x86_gpu_training
|
|
@pytest.mark.env_onecard
|
|
def test_gpu_graph_net():
|
|
"""
|
|
Feature: graph kernel testcase for Lamb
|
|
Description: fixed input when using ascend in graph mode
|
|
Expectation: get the same result when use new lamb kernel and old kernel
|
|
"""
|
|
context.set_context(mode=context.GRAPH_MODE, device_target="GPU")
|
|
test_gpu_net()
|
|
|
|
|
|
@pytest.mark.level0
|
|
@pytest.mark.platform_x86_gpu_training
|
|
@pytest.mark.env_onecard
|
|
def test_gpu_pynative_net():
|
|
"""
|
|
Feature: pynative kernel testcase for Lamb
|
|
Description: fixed input when using ascend in pynative mode
|
|
Expectation: get the same result when use new lamb kernel and old kernel
|
|
"""
|
|
context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU")
|
|
test_gpu_net()
|
|
|
|
|
|
@pytest.mark.level0
|
|
@pytest.mark.platform_arm_ascend_training
|
|
@pytest.mark.platform_x86_ascend_training
|
|
@pytest.mark.env_onecard
|
|
def test_ascend_graph_net():
|
|
"""
|
|
Feature: graph kernel testcase for Lamb
|
|
Description: fixed input when using ascend in graph mode
|
|
Expectation: get the same result when use new lamb kernel and old kernel
|
|
"""
|
|
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend")
|
|
test_ascend_net()
|
|
|
|
|
|
@pytest.mark.level0
|
|
@pytest.mark.platform_arm_ascend_training
|
|
@pytest.mark.platform_x86_ascend_training
|
|
@pytest.mark.env_onecard
|
|
def test_ascend_pynative_net():
|
|
"""
|
|
Feature: pynative kernel testcase for Lamb
|
|
Description: fixed input when using ascend in pynative mode
|
|
Expectation: get the same result when use new lamb kernel and old kernel
|
|
"""
|
|
context.set_context(mode=context.PYNATIVE_MODE, device_target="Ascend")
|
|
test_ascend_net()
|