[Ref_Impl] Add GroupNormalization-12 reference implementation (#18343)
* Add GroupNorm reference implementation * Clean up * Handle GroupNorm in ov namespace only * Simplify the flow * Add various scale and bias tests * Fix batches case * Add more float tests * Set output shape * Set output shape * Fix missing includes * Move to ov namespace --------- Co-authored-by: Michal Lukaszewski <michal.lukaszewski@intel.com>
This commit is contained in:
Tomasz JankowskiandMichal Lukaszewski authored and 
GitHub committed
1 parent
0cd4ea29ce
commit
e81bcacf33
7 files changed
+378
-2
No files matched your search
@@ -0,0 +1,58 @@
|
||||
// Copyright (C) 2018-2023 Intel Corporation
|
||||
// SPDX-License-Identifier: Apache-2.0
|
||||
//
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <cmath>
|
||||
#include <numeric>
|
||||
|
||||
#include "openvino/core/shape.hpp"
|
||||
|
||||
using namespace std;
|
||||
|
||||
namespace ov {
|
||||
namespace reference {
|
||||
|
||||
template <typename T>
|
||||
void group_normalization(const T* const data,
|
||||
const T* const scale,
|
||||
const T* const bias,
|
||||
T* const out,
|
||||
const Shape& data_shape,
|
||||
const size_t num_groups,
|
||||
const double epsilon) {
|
||||
const auto num_batches = data_shape[0];
|
||||
const auto num_channels = data_shape[1];
|
||||
const auto num_channels_in_group = num_channels / num_groups;
|
||||
const auto data_size = shape_size(data_shape);
|
||||
const auto batch_size = data_size / num_batches;
|
||||
const auto channel_size = batch_size / num_channels;
|
||||
const auto group_size = num_channels_in_group * channel_size;
|
||||
const auto eps = static_cast<T>(epsilon);
|
||||
|
||||
for (size_t n = 0; n < num_batches; ++n) {
|
||||
for (size_t g = 0; g < num_groups; ++g) {
|
||||
const auto group_begin = data + n * batch_size + g * group_size;
|
||||
const auto group_end = group_begin + group_size;
|
||||
const auto mean = accumulate(group_begin, group_end, static_cast<T>(0)) / group_size;
|
||||
const auto variance = accumulate(group_begin,
|
||||
group_end,
|
||||
static_cast<T>(0),
|
||||
[mean](const T acc, const T d) {
|
||||
return acc + pow(d - mean, 2);
|
||||
}) /
|
||||
group_size;
|
||||
const auto standard_deviation = sqrt(variance + eps);
|
||||
|
||||
for (size_t s = 0; s < num_channels_in_group; ++s) {
|
||||
const auto c = g * num_channels_in_group + s;
|
||||
const auto channel_offset = n * batch_size + c * channel_size;
|
||||
for (size_t i = channel_offset; i < channel_offset + channel_size; ++i)
|
||||
out[i] = ((data[i] - mean) / standard_deviation) * scale[c] + bias[c];
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
} // namespace reference
|
||||
} // namespace ov
|
||||
@@ -38,6 +38,6 @@ ov::Coordinate& ov::Coordinate::operator=(const Coordinate& v) {
|
||||
}
|
||||
|
||||
ov::Coordinate& ov::Coordinate::operator=(Coordinate&& v) noexcept {
|
||||
static_cast<std::vector<size_t>*>(this)->operator=(v);
|
||||
static_cast<std::vector<size_t>*>(this)->operator=(std::move(v));
|
||||
return *this;
|
||||
}
|
||||
@@ -5,6 +5,7 @@
|
||||
#pragma once
|
||||
#include "ngraph/ops.hpp"
|
||||
#include "openvino/core/except.hpp"
|
||||
#include "openvino/op/ops.hpp"
|
||||
|
||||
template <ngraph::element::Type_t ET>
|
||||
bool evaluate(std::shared_ptr<ngraph::Node> op,
|
||||
|
||||
@@ -0,0 +1,43 @@
|
||||
// Copyright (C) 2018-2023 Intel Corporation
|
||||
// SPDX-License-Identifier: Apache-2.0
|
||||
//
|
||||
|
||||
#include "ngraph/runtime/reference/group_normalization.hpp"
|
||||
|
||||
#include "evaluate_node.hpp"
|
||||
#include "openvino/op/group_normalization.hpp"
|
||||
|
||||
using namespace ov;
|
||||
|
||||
template <element::Type_t DATA_ET>
|
||||
bool evaluate(const std::shared_ptr<ov::op::v12::GroupNormalization>& node,
|
||||
const ov::HostTensorVector& outputs,
|
||||
const ov::HostTensorVector& inputs) {
|
||||
outputs[0]->set_shape(inputs[0]->get_shape());
|
||||
ov::reference::group_normalization(inputs[0]->get_data_ptr<DATA_ET>(),
|
||||
inputs[1]->get_data_ptr<DATA_ET>(),
|
||||
inputs[2]->get_data_ptr<DATA_ET>(),
|
||||
outputs[0]->get_data_ptr<DATA_ET>(),
|
||||
inputs[0]->get_shape(),
|
||||
static_cast<size_t>(node->get_num_groups()),
|
||||
node->get_epsilon());
|
||||
return true;
|
||||
}
|
||||
|
||||
template <>
|
||||
bool evaluate_node<op::v12::GroupNormalization>(std::shared_ptr<ov::Node> node,
|
||||
const ov::HostTensorVector& outputs,
|
||||
const ov::HostTensorVector& inputs) {
|
||||
switch (node->get_input_element_type(0)) {
|
||||
case element::Type_t::bf16:
|
||||
return evaluate<element::Type_t::bf16>(as_type_ptr<op::v12::GroupNormalization>(node), outputs, inputs);
|
||||
case element::Type_t::f16:
|
||||
return evaluate<element::Type_t::f16>(as_type_ptr<op::v12::GroupNormalization>(node), outputs, inputs);
|
||||
case element::Type_t::f64:
|
||||
return evaluate<element::Type_t::f64>(as_type_ptr<op::v12::GroupNormalization>(node), outputs, inputs);
|
||||
case element::Type_t::f32:
|
||||
return evaluate<element::Type_t::f32>(as_type_ptr<op::v12::GroupNormalization>(node), outputs, inputs);
|
||||
default:
|
||||
OPENVINO_THROW("Unhandled data type ", node->get_element_type().get_type_name(), "in evaluate_node()");
|
||||
}
|
||||
}
|
||||
@@ -445,10 +445,14 @@ extern template bool evaluate_node<ngraph::op::v10::Unique>(std::shared_ptr<ngra
|
||||
const ngraph::HostTensorVector& outputs,
|
||||
const ngraph::HostTensorVector& inputs);
|
||||
|
||||
extern template bool evaluate_node<ov::op::v12::GroupNormalization>(std::shared_ptr<ov::Node> node,
|
||||
const ov::HostTensorVector& outputs,
|
||||
const ov::HostTensorVector& inputs);
|
||||
|
||||
extern template bool evaluate_node<ov::op::internal::AUGRUCell>(std::shared_ptr<ngraph::Node> node,
|
||||
const ngraph::HostTensorVector& outputs,
|
||||
const ngraph::HostTensorVector& inputs);
|
||||
|
||||
extern template bool evaluate_node<ov::op::internal::AUGRUSequence>(std::shared_ptr<ngraph::Node> node,
|
||||
const ngraph::HostTensorVector& outputs,
|
||||
const ngraph::HostTensorVector& inputs);
|
||||
const ngraph::HostTensorVector& inputs);
|
||||
@@ -148,5 +148,7 @@ _OPENVINO_OP_REG(Unique, op::v10)
|
||||
|
||||
_OPENVINO_OP_REG(Interpolate, op::v11)
|
||||
|
||||
_OPENVINO_OP_REG(GroupNormalization, ov::op::v12)
|
||||
|
||||
_OPENVINO_OP_REG(AUGRUCell, ov::op::internal)
|
||||
_OPENVINO_OP_REG(AUGRUSequence, ov::op::internal)
|
||||
@@ -0,0 +1,268 @@
|
||||
// Copyright (C) 2022 Intel Corporation
|
||||
// SPDX-License-Identifier: Apache-2.0
|
||||
//
|
||||
|
||||
#include "openvino/op/group_normalization.hpp"
|
||||
|
||||
#include "base_reference_test.hpp"
|
||||
#include "gtest/gtest.h"
|
||||
#include "openvino/op/group_conv.hpp"
|
||||
|
||||
using namespace std;
|
||||
using namespace ov;
|
||||
using namespace reference_tests;
|
||||
|
||||
namespace {
|
||||
struct GroupNormalizationParams {
|
||||
GroupNormalizationParams(const reference_tests::Tensor& data,
|
||||
const reference_tests::Tensor& scale,
|
||||
const reference_tests::Tensor& bias,
|
||||
const reference_tests::Tensor& expected,
|
||||
int64_t num,
|
||||
double eps,
|
||||
string name)
|
||||
: data_tensor{data},
|
||||
scale_tensor{scale},
|
||||
bias_tensor{bias},
|
||||
expected_tensor{expected},
|
||||
num_groups{num},
|
||||
epsilon{eps},
|
||||
test_case_name{move(name)} {}
|
||||
|
||||
reference_tests::Tensor data_tensor;
|
||||
reference_tests::Tensor scale_tensor;
|
||||
reference_tests::Tensor bias_tensor;
|
||||
reference_tests::Tensor expected_tensor;
|
||||
int64_t num_groups;
|
||||
double epsilon;
|
||||
string test_case_name;
|
||||
};
|
||||
|
||||
class ReferenceGroupNormalization : public testing::TestWithParam<GroupNormalizationParams>,
|
||||
public CommonReferenceTest {
|
||||
public:
|
||||
void SetUp() override {
|
||||
const auto& params = GetParam();
|
||||
function = CreateFunction(params);
|
||||
inputData = {params.data_tensor.data, params.scale_tensor.data, params.bias_tensor.data};
|
||||
refOutData = {params.expected_tensor.data};
|
||||
}
|
||||
|
||||
static string getTestCaseName(const testing::TestParamInfo<GroupNormalizationParams>& obj) {
|
||||
std::ostringstream name;
|
||||
name << obj.param.test_case_name;
|
||||
name << "_";
|
||||
name << obj.param.data_tensor.type;
|
||||
return name.str();
|
||||
}
|
||||
|
||||
private:
|
||||
static shared_ptr<Model> CreateFunction(const GroupNormalizationParams& params) {
|
||||
const auto in_data = make_shared<op::v0::Parameter>(params.data_tensor.type, params.data_tensor.shape);
|
||||
const auto in_scale = make_shared<op::v0::Parameter>(params.scale_tensor.type, params.scale_tensor.shape);
|
||||
const auto in_bias = make_shared<op::v0::Parameter>(params.bias_tensor.type, params.bias_tensor.shape);
|
||||
const auto group_norm =
|
||||
make_shared<op::v12::GroupNormalization>(in_data, in_scale, in_bias, params.num_groups, params.epsilon);
|
||||
return make_shared<Model>(NodeVector{group_norm}, ParameterVector{in_data, in_scale, in_bias});
|
||||
}
|
||||
};
|
||||
|
||||
vector<GroupNormalizationParams> generateBasicParams() {
|
||||
constexpr auto et = element::f32;
|
||||
using vt = typename element_type_traits<et>::value_type;
|
||||
|
||||
const Shape data_shape{1, 4, 2, 2};
|
||||
reference_tests::Tensor scale{{4}, et, vector<vt>{1, 1, 1, 1}};
|
||||
reference_tests::Tensor bias{{4}, et, vector<vt>{0, 0, 0, 0}};
|
||||
|
||||
// clang-format off
|
||||
reference_tests::Tensor data{data_shape, et,
|
||||
vector<vt>{0.001, 0.002, 0.003, 0.004,
|
||||
1.01, 1.02, 1.03, 1.04,
|
||||
2.1, 2.2, 2.3, 2.4,
|
||||
11, 12, 13, 14}};
|
||||
reference_tests::Tensor output_4_groups{data_shape, et,
|
||||
vector<vt>{-0.4472, -0.1491, 0.1491, 0.4472,
|
||||
-1.2910, -0.4303, 0.4303, 1.2910,
|
||||
-1.3411, -0.4470, 0.4470, 1.3411,
|
||||
-1.3416, -0.4472, 0.4472, 1.3416}};
|
||||
reference_tests::Tensor output_2_groups{data_shape, et,
|
||||
vector<vt>{-1.0028, -1.0008, -0.9989, -0.9969,
|
||||
0.9705, 0.9901, 1.0096, 1.0292,
|
||||
-1.0171, -0.9978, -0.9786, -0.9593,
|
||||
0.6990, 0.8918, 1.0846, 1.2774}};
|
||||
reference_tests::Tensor output_1_group{data_shape, et,
|
||||
vector<vt>{-0.7832, -0.7830, -0.7828, -0.7826,
|
||||
-0.5828, -0.5808, -0.5789, -0.5769,
|
||||
-0.3663, -0.3465, -0.3266, -0.3067,
|
||||
1.4014, 1.6000, 1.7986, 1.9973}};
|
||||
// clang-format on
|
||||
|
||||
vector<GroupNormalizationParams> params;
|
||||
params.emplace_back(data, scale, bias, output_4_groups, 4, 1e-5, "basic_4groups");
|
||||
params.emplace_back(data, scale, bias, output_2_groups, 2, 1e-5, "basic_2groups");
|
||||
params.emplace_back(data, scale, bias, output_1_group, 1, 1e-5, "basic_1group");
|
||||
return params;
|
||||
}
|
||||
|
||||
template <element::Type_t et>
|
||||
vector<GroupNormalizationParams> generateVariousScaleBiasParams() {
|
||||
using vt = typename element_type_traits<et>::value_type;
|
||||
|
||||
const Shape data_shape{1, 2, 2, 2};
|
||||
reference_tests::Tensor scale{{2}, et, vector<vt>{0.7, 1.2}};
|
||||
reference_tests::Tensor bias{{2}, et, vector<vt>{0.2, -0.3}};
|
||||
|
||||
reference_tests::Tensor data{data_shape, et, vector<vt>{0.001, 0.002, 0.003, 0.004, 2.1, 2.2, 2.3, 2.4}};
|
||||
reference_tests::Tensor output_2_groups{
|
||||
data_shape,
|
||||
et,
|
||||
vector<vt>{-0.5937, -0.0646, 0.4646, 0.9937, -1.9099, -0.8366, 0.2366, 1.3099}};
|
||||
reference_tests::Tensor output_1_group{
|
||||
data_shape,
|
||||
et,
|
||||
vector<vt>{-0.4992, -0.4986, -0.4980, -0.4973, 0.7373, 0.8438, 0.9503, 1.0568}};
|
||||
|
||||
vector<GroupNormalizationParams> params;
|
||||
params.emplace_back(data, scale, bias, output_2_groups, 2, 5e-7, "scale_bias_2groups");
|
||||
params.emplace_back(data, scale, bias, output_1_group, 1, 5e-7, "scale_bias_1group");
|
||||
return params;
|
||||
}
|
||||
|
||||
vector<GroupNormalizationParams> generateComplexParams() {
|
||||
constexpr auto et = element::f32;
|
||||
using vt = typename element_type_traits<et>::value_type;
|
||||
|
||||
vector<GroupNormalizationParams> params;
|
||||
{
|
||||
const Shape data_shape{2, 2, 1, 2};
|
||||
reference_tests::Tensor scale{{2}, et, vector<vt>{2, 3}};
|
||||
reference_tests::Tensor bias{{2}, et, vector<vt>{-1, 10}};
|
||||
|
||||
reference_tests::Tensor data{data_shape, et, vector<vt>{1, 11, 2, 22, 33, 3, 44, 4}};
|
||||
|
||||
reference_tests::Tensor output_2_groups{data_shape, et, vector<vt>{-3, 1, 7, 13, 1, -3, 13, 7}};
|
||||
reference_tests::Tensor output_1_group{
|
||||
data_shape,
|
||||
et,
|
||||
vector<vt>{-2.8922, -0.5270, 7.5165, 14.6122, 0.3385, -3.0078, 13.8482, 7.1557}};
|
||||
params.emplace_back(data, scale, bias, output_2_groups, 2, 1e-5, "2batches_2groups");
|
||||
params.emplace_back(data, scale, bias, output_1_group, 1, 1e-5, "2batches_1group");
|
||||
}
|
||||
|
||||
{
|
||||
const Shape data_shape{3, 3, 7, 5};
|
||||
reference_tests::Tensor scale{{3}, et, vector<vt>{0.77, 0.99, 1.11}};
|
||||
reference_tests::Tensor bias{{3}, et, vector<vt>{0.21, 0.07, -0.37}};
|
||||
reference_tests::Tensor data{
|
||||
data_shape,
|
||||
et,
|
||||
vector<vt>{0.1648, 0.2497, 0.1195, 0.1584, 0.2611, 0.3091, 0.1737, 0.0421, 0.9458, 0.3296, 0.8453, 0.1896,
|
||||
0.1267, 0.2665, 0.2749, 0.2594, 0.4134, 0.7526, 0.6499, 0.7584, 0.8176, 0.9045, 0.8179, 0.9137,
|
||||
0.5692, 0.6913, 0.1530, 0.8242, 0.1903, 0.2448, 0.9984, 0.9623, 0.5121, 0.2807, 0.5177,
|
||||
|
||||
0.3564, 0.6398, 0.5228, 0.3958, 0.1639, 0.5372, 0.5363, 0.5431, 0.5481, 0.7458, 0.5475, 0.6672,
|
||||
0.8525, 0.7553, 0.1790, 0.5535, 0.4606, 0.9114, 0.9227, 0.1119, 0.5071, 0.2426, 0.2356, 0.7997,
|
||||
0.8595, 0.5270, 0.4007, 0.7677, 0.1941, 0.1273, 0.8571, 0.5119, 0.8057, 0.8024, 0.2653,
|
||||
|
||||
0.7026, 0.0503, 0.4928, 0.1142, 0.2521, 0.5274, 0.9813, 0.1118, 0.6919, 0.9727, 0.4152, 0.7918,
|
||||
0.2413, 0.3088, 0.3133, 0.0472, 0.1392, 0.1585, 0.7208, 0.3988, 0.9662, 0.9470, 0.4756, 0.7606,
|
||||
0.9366, 0.8194, 0.6172, 0.8799, 0.9407, 0.4477, 0.4529, 0.8176, 0.1426, 0.4921, 0.1570,
|
||||
|
||||
0.9935, 0.0889, 0.3966, 0.9366, 0.0620, 0.2797, 0.6732, 0.0448, 0.4008, 0.5622, 0.2622, 0.1329,
|
||||
0.9235, 0.3959, 0.2991, 0.7200, 0.8702, 0.9584, 0.4113, 0.5059, 0.4774, 0.6776, 0.6542, 0.1194,
|
||||
0.1748, 0.6760, 0.0938, 0.7508, 0.6926, 0.6281, 0.8262, 0.2689, 0.0053, 0.3767, 0.9926,
|
||||
|
||||
0.0568, 0.2568, 0.8414, 0.0434, 0.5292, 0.0985, 0.2029, 0.5948, 0.3747, 0.5926, 0.2130, 0.0519,
|
||||
0.9413, 0.2119, 0.2858, 0.2752, 0.9895, 0.1642, 0.9411, 0.8528, 0.5139, 0.3473, 0.8534, 0.3050,
|
||||
0.1338, 0.7669, 0.2127, 0.7881, 0.5265, 0.3529, 0.5342, 0.5154, 0.4277, 0.1556, 0.2562,
|
||||
|
||||
0.9280, 0.6243, 0.4608, 0.3487, 0.8103, 0.8956, 0.8416, 0.3014, 0.5215, 0.9945, 0.3381, 0.7508,
|
||||
0.6548, 0.8600, 0.5781, 0.7989, 0.0082, 0.9144, 0.4307, 0.1504, 0.3410, 0.7372, 0.6724, 0.4242,
|
||||
0.0205, 0.9671, 0.5956, 0.1512, 0.5713, 0.6840, 0.2899, 0.3118, 0.6995, 0.8726, 0.0138,
|
||||
|
||||
0.2557, 0.5081, 0.6399, 0.0994, 0.8553, 0.8695, 0.0975, 0.1074, 0.2329, 0.4659, 0.4810, 0.3403,
|
||||
0.6434, 0.1760, 0.0869, 0.7490, 0.4383, 0.2731, 0.3274, 0.0870, 0.2499, 0.6138, 0.2206, 0.6845,
|
||||
0.6387, 0.7274, 0.5181, 0.8421, 0.4487, 0.2840, 0.9888, 0.9848, 0.4048, 0.1550, 0.5963,
|
||||
|
||||
0.5346, 0.8775, 0.6872, 0.6386, 0.4324, 0.7045, 0.0228, 0.7691, 0.8856, 0.7495, 0.6414, 0.6308,
|
||||
0.2641, 0.5891, 0.1852, 0.0247, 0.0910, 0.8219, 0.0344, 0.6146, 0.1786, 0.1071, 0.8895, 0.4652,
|
||||
0.3224, 0.2132, 0.0159, 0.6958, 0.2965, 0.6573, 0.6049, 0.4310, 0.6706, 0.6910, 0.8486,
|
||||
|
||||
0.2045, 0.8496, 0.1481, 0.7015, 0.1985, 0.3038, 0.9332, 0.4544, 0.2214, 0.5375, 0.1587, 0.0834,
|
||||
0.8334, 0.1958, 0.2984, 0.8155, 0.6043, 0.8328, 0.5758, 0.2238, 0.9458, 0.9503, 0.1430, 0.5495,
|
||||
0.1429, 0.4320, 0.1902, 0.6482, 0.1235, 0.8223, 0.8314, 0.6919, 0.8652, 0.1149, 0.3165}};
|
||||
reference_tests::Tensor output{
|
||||
data_shape,
|
||||
et,
|
||||
vector<vt>{-0.7244, -0.4964, -0.8461, -0.7416, -0.4657, -0.3368, -0.7005, -1.0541, 1.3736,
|
||||
-0.2817, 1.1036, -0.6578, -0.8268, -0.4512, -0.4287, -0.4703, -0.0566, 0.8546,
|
||||
0.5787, 0.8702, 1.0292, 1.2627, 1.0300, 1.2874, 0.3619, 0.6899, -0.7561,
|
||||
1.0469, -0.6559, -0.5095, 1.5149, 1.4179, 0.2085, -0.4131, 0.2236,
|
||||
|
||||
-0.4697, 0.5092, 0.1051, -0.3336, -1.1345, 0.1548, 0.1517, 0.1752, 0.1925,
|
||||
0.8753, 0.1904, 0.6038, 1.2438, 0.9081, -1.0824, 0.2111, -0.1098, 1.4473,
|
||||
1.4863, -1.3141, 0.0508, -0.8627, -0.8869, 1.0615, 1.2680, 0.1196, -0.3166,
|
||||
0.9509, -1.0302, -1.2609, 1.2597, 0.0674, 1.0822, 1.0708, -0.7843,
|
||||
|
||||
0.3656, -2.1605, -0.4469, -1.9130, -1.3790, -0.3129, 1.4449, -1.9223, 0.3242,
|
||||
1.4116, -0.7474, 0.7110, -1.4208, -1.1594, -1.1420, -2.1725, -1.8162, -1.7414,
|
||||
0.4361, -0.8109, 1.3864, 1.3121, -0.5135, 0.5902, 1.2718, 0.8179, 0.0349,
|
||||
1.0522, 1.2877, -0.6215, -0.6014, 0.8109, -1.8030, -0.4496, -1.7473,
|
||||
|
||||
1.5069, -0.8523, -0.0498, 1.3585, -0.9224, -0.3547, 0.6716, -0.9673, -0.0388,
|
||||
0.3821, -0.4003, -0.7375, 1.3244, -0.0516, -0.3041, 0.7936, 1.1854, 1.4154,
|
||||
-0.0115, 0.2353, 0.1609, 0.6830, 0.6220, -0.7727, -0.6283, 0.6789, -0.8395,
|
||||
0.8740, 0.7222, 0.5540, 1.0706, -0.3828, -1.0703, -0.1017, 1.5046,
|
||||
|
||||
-1.4034, -0.7328, 1.2275, -1.4484, 0.1806, -1.2636, -0.9135, 0.4006, -0.3375,
|
||||
0.3932, -0.8797, -1.4199, 1.5624, -0.8834, -0.6356, -0.6711, 1.7241, -1.0433,
|
||||
1.5618, 1.2657, 0.1293, -0.4293, 1.2677, -0.5712, -1.1452, 0.9776, -0.8807,
|
||||
1.0487, 0.1715, -0.4106, 0.1974, 0.1343, -0.1597, -1.0721, -0.7348,
|
||||
|
||||
1.2533, 0.1115, -0.5032, -0.9246, 0.8108, 1.1315, 0.9285, -1.1024, -0.2749,
|
||||
1.5033, -0.9645, 0.5871, 0.2262, 0.9977, -0.0622, 0.7680, -2.2047, 1.2022,
|
||||
-0.6163, -1.6701, -0.9535, 0.5360, 0.2924, -0.6408, -2.1585, 1.4003, 0.0036,
|
||||
-1.6671, -0.0877, 0.3360, -1.1457, -1.0633, 0.3943, 1.0450, -2.1837,
|
||||
|
||||
-0.3997, 0.2887, 0.6482, -0.8260, 1.2357, 1.2744, -0.8311, -0.8041, -0.4619,
|
||||
0.1736, 0.2148, -0.1689, 0.6577, -0.6170, -0.8600, 0.9457, 0.0983, -0.3522,
|
||||
-0.2041, -0.8598, -0.4155, 0.5770, -0.4954, 0.7698, 0.6449, 0.8868, 0.3160,
|
||||
1.1997, 0.1267, -0.3225, 1.5998, 1.5888, 0.0070, -0.6743, 0.5293,
|
||||
|
||||
0.2641, 1.4665, 0.7992, 0.6288, -0.0942, 0.8599, -1.5305, 1.0864, 1.4949,
|
||||
1.0177, 0.6386, 0.6015, -0.6844, 0.4552, -0.9611, -1.5239, -1.2914, 1.2716,
|
||||
-1.4899, 0.5447, -0.9842, -1.2349, 1.5086, 0.0208, -0.4800, -0.8629, -1.5547,
|
||||
0.8294, -0.5708, 0.6944, 0.5106, -0.0992, 0.7410, 0.8126, 1.3652,
|
||||
|
||||
-1.4502, 1.0861, -1.6719, 0.5038, -1.4738, -1.0598, 1.4148, -0.4677, -1.3837,
|
||||
-0.1409, -1.6302, -1.9263, 1.0224, -1.4844, -1.0810, 0.9521, 0.1217, 1.0201,
|
||||
0.0096, -1.3743, 1.4643, 1.4820, -1.6920, -0.0938, -1.6924, -0.5557, -1.5064,
|
||||
0.2943, -1.7686, 0.9788, 1.0146, 0.4661, 1.1475, -1.8025, -1.0098}};
|
||||
|
||||
params.emplace_back(data, scale, bias, output, 1, 1e-5, "3batches_1group");
|
||||
}
|
||||
return params;
|
||||
}
|
||||
|
||||
vector<GroupNormalizationParams> generateGroupNormalizationParams() {
|
||||
vector<vector<GroupNormalizationParams>> combo_params{generateBasicParams(),
|
||||
generateVariousScaleBiasParams<element::f16>(),
|
||||
generateVariousScaleBiasParams<element::f32>(),
|
||||
generateVariousScaleBiasParams<element::f64>(),
|
||||
generateComplexParams()};
|
||||
vector<GroupNormalizationParams> test_params;
|
||||
for (auto& params : combo_params)
|
||||
move(params.begin(), params.end(), back_inserter(test_params));
|
||||
return test_params;
|
||||
}
|
||||
} // namespace
|
||||
|
||||
TEST_P(ReferenceGroupNormalization, LayerTest) {
|
||||
Exec();
|
||||
}
|
||||
|
||||
INSTANTIATE_TEST_SUITE_P(smoke,
|
||||
ReferenceGroupNormalization,
|
||||
::testing::ValuesIn(generateGroupNormalizationParams()),
|
||||
ReferenceGroupNormalization::getTestCaseName);
|
||||
Reference in new issue
Block a user