Skip to content

Add stable parameter to make CPU result correct #32217

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 6 commits into
base: master
Choose a base branch
from
+32 −22
Open

Add stable parameter to make CPU result correct #32217

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
8a9ccb6
Add stable parameter to make CPU result correct
sgbihu Sep 26, 2025
4662abe
Delete the unused Convert Operator
sgbihu Sep 28, 2025
f8454ac
Fix make TopK output i64
sgbihu Sep 29, 2025
be87552
Change the return type based on review inputs
sgbihu Sep 29, 2025
36db638
Modify test to test the updated code
sgbihu Sep 29, 2025
cf02b03
Update the test input
sgbihu Sep 29, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
36 changes: 23 additions & 13 deletions src/frontends/onnx/frontend/src/utils/arg_min_max_factory.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,15 +28,15 @@ ArgMinMaxFactory::ArgMinMaxFactory(const Node& node)
m_axis{node.get_attribute_value<std::int64_t>("axis", 0)},
m_select_last_index{node.get_attribute_value<std::int64_t>("select_last_index", 0)} {}

std::shared_ptr<ov::Node> ArgMinMaxFactory::make_arg_max() const {
ov::Output<ov::Node> ArgMinMaxFactory::make_arg_max() const {
return make_topk_subgraph(v11::TopK::Mode::MAX);
}

std::shared_ptr<ov::Node> ArgMinMaxFactory::make_arg_min() const {
ov::Output<ov::Node> ArgMinMaxFactory::make_arg_min() const {
return make_topk_subgraph(v11::TopK::Mode::MIN);
}

std::shared_ptr<ov::Node> ArgMinMaxFactory::make_topk_subgraph(v11::TopK::Mode mode) const {
ov::Output<ov::Node> ArgMinMaxFactory::make_topk_subgraph(v11::TopK::Mode mode) const {
const auto k_node = v0::Constant::create(ov::element::i64, ov::Shape{}, {1});

if (m_select_last_index == 1) {
Expand Down Expand Up @@ -70,40 +70,50 @@ std::shared_ptr<ov::Node> ArgMinMaxFactory::make_topk_subgraph(v11::TopK::Mode m
const auto axis_node = v0::Constant::create(ov::element::i64, ov::Shape{1}, {normalized_axis});
const auto reverse = std::make_shared<v1::Reverse>(m_input_node, axis_node, v1::Reverse::Mode::INDEX);

const auto topk = std::make_shared<v11::TopK>(reverse, k_node, normalized_axis, mode, v1::TopK::SortType::NONE);
const auto topk = std::make_shared<v11::TopK>(reverse,
k_node,
normalized_axis,
mode,
v1::TopK::SortType::SORT_VALUES,
element::i64,
true);

const auto data_shape = std::make_shared<v0::ShapeOf>(m_input_node);
const auto dims_on_axis =
std::make_shared<v1::Gather>(data_shape,
axis_node,
v0::Constant::create(ov::element::i64, ov::Shape{}, {0}));

const auto res_index =
std::make_shared<v1::Subtract>(dims_on_axis,
std::make_shared<v0::Convert>(topk->output(1), ov::element::i64));
const auto res_index = std::make_shared<v1::Subtract>(dims_on_axis, topk->output(1));
const auto result =
std::make_shared<v1::Subtract>(res_index, v0::Constant::create(ov::element::i64, ov::Shape{1}, {1}));

if (m_keep_dims == 0) {
const auto axis_to_remove = v0::Constant::create(ov::element::u64, ov::Shape{}, {topk->get_axis()});

return std::make_shared<v0::Squeeze>(result, axis_to_remove);
return {std::make_shared<v0::Squeeze>(result, axis_to_remove)};
}

return result;
return {result};
}

const auto topk = std::make_shared<v11::TopK>(m_input_node, k_node, m_axis, mode, v11::TopK::SortType::NONE);
const auto topk = std::make_shared<v11::TopK>(m_input_node,
k_node,
m_axis,
mode,
v11::TopK::SortType::SORT_VALUES,
element::i64,
true);

const auto result = std::make_shared<v0::Convert>(topk->output(1), ov::element::i64);
ov::Output<ov::Node> result = topk->output(1);

if (m_keep_dims == 0) {
const auto axis_to_remove = v0::Constant::create(ov::element::u64, ov::Shape{}, {topk->get_axis()});

return std::make_shared<v0::Squeeze>(result, axis_to_remove);
return {std::make_shared<v0::Squeeze>(result, axis_to_remove)};
}

return result;
return {result};
}
} // namespace utils
} // namespace onnx
Expand Down
6 changes: 3 additions & 3 deletions src/frontends/onnx/frontend/src/utils/arg_min_max_factory.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,14 +24,14 @@ class ArgMinMaxFactory {

/// \brief Creates ArgMax ONNX operation.
/// \return Sub-graph representing ArgMax op.
std::shared_ptr<ov::Node> make_arg_max() const;
ov::Output<ov::Node> make_arg_max() const;

/// \brief Creates ArgMin ONNX operation.
/// \return Sub-graph representing ArgMin op.
std::shared_ptr<ov::Node> make_arg_min() const;
ov::Output<ov::Node> make_arg_min() const;

private:
std::shared_ptr<ov::Node> make_topk_subgraph(ov::op::v11::TopK::Mode mode) const;
ov::Output<ov::Node> make_topk_subgraph(ov::op::v11::TopK::Mode mode) const;

const std::int64_t m_keep_dims;
ov::Output<ov::Node> m_input_node;
Expand Down
12 changes: 6 additions & 6 deletions src/frontends/onnx/tests/onnx_import.in.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2744,16 +2744,16 @@ OPENVINO_TEST(${BACKEND_NAME}, onnx_model_argmax_int32) {
auto model = convert_model("argmax_int32.onnx");

auto test_case = ov::test::TestCase(model, s_device);
test_case.add_input<std::int32_t>({1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12});
test_case.add_expected_output<std::int64_t>({1, 1, 1, 1, 1, 1});
test_case.add_input<std::int32_t>({3, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12});
test_case.add_expected_output<std::int64_t>({0, 1, 1, 1, 1, 1});
test_case.run();
}

OPENVINO_TEST(${BACKEND_NAME}, onnx_model_argmin_int32) {
auto model = convert_model("argmin_int32.onnx");

auto test_case = ov::test::TestCase(model, s_device);
test_case.add_input<std::int32_t>({1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12});
test_case.add_input<std::int32_t>({2, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12});
test_case.add_expected_output<std::int64_t>({0, 0, 0, 0});
test_case.run();
}
Expand All @@ -2762,7 +2762,7 @@ OPENVINO_TEST(${BACKEND_NAME}, onnx_model_argmax_float) {
auto model = convert_model("argmax_float.onnx");

auto test_case = ov::test::TestCase(model, s_device);
test_case.add_input<float>({4.f, 0.1f, 2.f, 3.f, -3.f, 1.f, -0.9f, 0.f, 1.f, 2.f, 3.f, 0.f});
test_case.add_input<float>({4.f, 0.1f, 2.f, 4.f, -3.f, 1.f, -0.9f, 0.f, 1.f, 2.f, 3.f, 0.f});
test_case.add_expected_output<std::int64_t>({0, 3, 0});
test_case.run();
}
Expand All @@ -2771,8 +2771,8 @@ OPENVINO_TEST(${BACKEND_NAME}, onnx_model_argmin_float) {
auto model = convert_model("argmin_float.onnx");

auto test_case = ov::test::TestCase(model, s_device);
test_case.add_input<float>({4.f, 0.1f, 2.f, 3.f, -3.f, 1.f, -0.9f, 0.f, 1.f, 2.f, 3.f, 0.f});
test_case.add_expected_output<std::int64_t>({1, 1, 0, 2});
test_case.add_input<float>({0.1f, 0.1f, 2.f, 3.f, -3.f, 1.f, -0.9f, 0.f, 1.f, 2.f, 3.f, 0.f});
test_case.add_expected_output<std::int64_t>({0, 1, 0, 2});
test_case.run();
}

Expand Down
Loading