Skip to content

PR#252 rebased #253

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

Merged
ravi9 merged 2 commits into from
Jul 14, 2026
+58 −45
Merged

PR#252 rebased #253

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4b30e3e
OpenVINO Backend: move parameter node creating from decoder into tran…
zhaixuejun1993 Jul 3, 2026
0733933
OpenVINO Backend: create extra input ov node move from decoder to tra…
zhaixuejun1993 Jul 3, 2026
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
31 changes: 5 additions & 26 deletions ggml/src/ggml-openvino/ggml-decoder.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,6 @@
#include <openvino/core/type/float16.hpp>
#include <openvino/op/constant.hpp>
#include <openvino/op/convert.hpp>
#include <openvino/op/parameter.hpp>
#include <openvino/runtime/tensor.hpp>
#include <ostream>
#include <set>
Expand Down Expand Up @@ -708,21 +707,7 @@ void GgmlOvDecoder::add_extra_inputs() {
// 2. `n_seq_active` and `seq_active_start`, used in FLASH_ATTN_EXT to indicate the active sequences in the batch

auto create_1d_input = [this](const std::string & name, int64_t value) {
if (m_is_static) {
auto constant =
std::make_shared<ov::op::v0::Constant>(ov::element::i64, ov::Shape{1}, std::vector<int64_t>{value});
constant->set_friendly_name(name);
m_model_extra_inputs[name] = constant;
} else {
auto param_node = std::make_shared<ov::op::v0::Parameter>(ov::element::i64, ov::Shape{1});
param_node->set_friendly_name(name);
param_node->output(0).get_tensor().set_names({name});
m_model_extra_inputs[name] = param_node;

auto tensor = std::make_shared<ov::Tensor>(ov::element::i64, ov::Shape{1});
*tensor->data<int64_t>() = value;
m_model_extra_input_values[name] = tensor;
}
m_model_extra_inputs[name] = {ov::element::i64, ov::Shape{1}, value, !m_is_static};
};

if (m_compute_params.attention_size != -1) {
Expand Down Expand Up @@ -773,11 +758,8 @@ void GgmlOvDecoder::compute_model_inputs() {
std::string node_name = get_tensor_ov_name(m_cgraph, node);
if (m_model_weights.find(node_name) == m_model_weights.end()) {
m_inputs[node_name] = node;
auto param_node = std::make_shared<ov::op::v0::Parameter>(
get_ov_type(node), get_graph_input_shape(node, nullptr, m_node_dynamic_dims[node]));
param_node->set_friendly_name(node_name);
param_node->output(0).get_tensor().set_names({node_name});
m_model_inputs[node_name] = param_node;
m_model_inputs[node_name] = {get_ov_type(node),
get_graph_input_shape(node, nullptr, m_node_dynamic_dims[node])};
}
continue;
}
Expand Down Expand Up @@ -824,11 +806,8 @@ void GgmlOvDecoder::compute_model_inputs() {
src_name = get_tensor_ov_name(m_cgraph, src);
}
m_inputs[src_name] = src;
ov::PartialShape param_shape = get_graph_input_shape(node, src, m_node_dynamic_dims[src]);
auto param_node = std::make_shared<ov::op::v0::Parameter>(get_ov_type(src), param_shape);
param_node->set_friendly_name(src_name);
param_node->output(0).get_tensor().set_names({src_name});
m_model_inputs[src_name] = param_node;
m_model_inputs[src_name] = {get_ov_type(src),
get_graph_input_shape(node, src, m_node_dynamic_dims[src])};
}
}
}
Expand Down
13 changes: 4 additions & 9 deletions ggml/src/ggml-openvino/ggml-decoder.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -207,18 +207,14 @@ class GgmlOvDecoder : public ov::frontend::ggml::GgmlDecoder {

virtual int get_op_case(int node_idx) const override { return m_node_info_list[node_idx].node_op_case; }

virtual const std::map<std::string, std::shared_ptr<ov::Node>> & get_model_inputs() const override {
virtual const std::map<std::string, ov::frontend::ggml::ModelInputInfo> & get_model_inputs() const override {
return m_model_inputs;
}

virtual const std::map<std::string, std::shared_ptr<ov::Node>> & get_model_extra_inputs() const override {
virtual const std::map<std::string, ov::frontend::ggml::ModelExtraInputInfo> & get_model_extra_inputs() const override {
return m_model_extra_inputs;
}

virtual const std::map<std::string, std::shared_ptr<ov::Tensor>> & get_model_extra_input_values() const {
return m_model_extra_input_values;
}

virtual const std::map<std::string, std::shared_ptr<ov::Node>> & get_model_weights() const override {
return m_model_weights;
}
Expand Down Expand Up @@ -379,9 +375,8 @@ class GgmlOvDecoder : public ov::frontend::ggml::GgmlDecoder {
ggml_cgraph * m_cgraph = nullptr;
std::map<std::string, ggml_tensor *> m_inputs;

std::map<std::string, std::shared_ptr<ov::Node>> m_model_inputs;
std::map<std::string, std::shared_ptr<ov::Node>> m_model_extra_inputs;
std::map<std::string, std::shared_ptr<ov::Tensor>> m_model_extra_input_values;
std::map<std::string, ov::frontend::ggml::ModelInputInfo> m_model_inputs;
std::map<std::string, ov::frontend::ggml::ModelExtraInputInfo> m_model_extra_inputs;
std::map<std::string, std::shared_ptr<ov::Node>> m_model_weights;
std::map<std::string, ggml_tensor *> m_model_outputs;
std::set<std::string> m_model_output_names;
Expand Down
19 changes: 16 additions & 3 deletions ggml/src/ggml-openvino/openvino/decoder.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,12 +6,25 @@
#include <openvino/core/partial_shape.hpp>
#include <openvino/core/shape.hpp>
#include <openvino/frontend/decoder.hpp>
#include <set>
#include <string>

namespace ov {
namespace frontend {
namespace ggml {

struct ModelInputInfo {
element::Type type;
PartialShape shape;
};

struct ModelExtraInputInfo {
element::Type type;
Shape shape;
int64_t value;
bool is_parameter;
};

class GgmlDecoder : public DecoderBase {
public:
virtual ov::Any get_attribute(const std::string & name) const = 0;
Expand Down Expand Up @@ -91,9 +104,9 @@ class GgmlDecoder : public DecoderBase {

virtual int get_op_case(int node_idx) const = 0;

virtual const std::map<std::string, std::shared_ptr<ov::Node>>& get_model_inputs() const = 0;
virtual const std::map<std::string, std::shared_ptr<ov::Node>>& get_model_extra_inputs() const = 0;
virtual const std::map<std::string, std::shared_ptr<ov::Node>>& get_model_weights() const = 0;
virtual const std::map<std::string, ModelInputInfo> & get_model_inputs() const = 0;
virtual const std::map<std::string, ModelExtraInputInfo> & get_model_extra_inputs() const = 0;
virtual const std::map<std::string, std::shared_ptr<ov::Node>> & get_model_weights() const = 0;
virtual std::set<std::string> get_model_output_names() const = 0;

virtual int32_t * get_rope_params() const = 0;
Expand Down
34 changes: 29 additions & 5 deletions ggml/src/ggml-openvino/openvino/translate_session.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,28 @@ using namespace ov::op;

namespace {

std::shared_ptr<ov::op::v0::Parameter> create_parameter(const std::string & name,
const ModelInputInfo & input_info) {
auto param_node = std::make_shared<ov::op::v0::Parameter>(input_info.type, input_info.shape);
param_node->set_friendly_name(name);
param_node->output(0).get_tensor().set_names({name});
return param_node;
}

std::shared_ptr<ov::Node> create_extra_input(const std::string & name, const ModelExtraInputInfo & input_info) {
if (input_info.is_parameter) {
auto param_node = std::make_shared<ov::op::v0::Parameter>(input_info.type, input_info.shape);
param_node->set_friendly_name(name);
param_node->output(0).get_tensor().set_names({name});
return param_node;
}

auto constant = std::make_shared<ov::op::v0::Constant>(input_info.type, input_info.shape,
std::vector<int64_t>{input_info.value});
constant->set_friendly_name(name);
return constant;
}

ov::pass::MakeStateful::ParamResPairs get_kv_param_res_pairs(
const std::shared_ptr<ov::Model> & model,
const std::map<std::string, std::string> & kv_param_res_names) {
Expand Down Expand Up @@ -177,15 +199,17 @@ std::shared_ptr<Model> TranslateSession::translate_graph(const frontend::InputMo
std::shared_ptr<GgmlDecoder> ggml_model_decoder = ggml_model->get_model_decoder();

for (const auto & it : ggml_model_decoder->get_model_inputs()) {
params.push_back(std::dynamic_pointer_cast<ov::op::v0::Parameter>(it.second));
(*tensor_map)[it.first] = it.second;
auto param_node = create_parameter(it.first, it.second);
params.push_back(param_node);
(*tensor_map)[it.first] = param_node;
}

for (const auto & it : ggml_model_decoder->get_model_extra_inputs()) {
if (std::dynamic_pointer_cast<ov::op::v0::Parameter>(it.second)) {
params.push_back(std::dynamic_pointer_cast<ov::op::v0::Parameter>(it.second));
auto input_node = create_extra_input(it.first, it.second);
if (it.second.is_parameter) {
params.push_back(std::dynamic_pointer_cast<ov::op::v0::Parameter>(input_node));
}
(*tensor_map)[it.first] = it.second;
(*tensor_map)[it.first] = input_node;
}

for (const auto & it : ggml_model_decoder->get_model_weights()) {
Expand Down
6 changes: 4 additions & 2 deletions ggml/src/ggml-openvino/utils.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -847,8 +847,10 @@ ov::Tensor convert_ggml_input_to_ov(std::shared_ptr<GgmlOvDecoder> ggml_decoder,

ov::Tensor get_ov_input_tensor(std::shared_ptr<GgmlOvDecoder> ggml_decoder, const std::string & param_name) {
ov::Tensor input_tensor;
if (ggml_decoder->get_model_extra_inputs().find(param_name) != ggml_decoder->get_model_extra_inputs().end()) {
input_tensor = *ggml_decoder->get_model_extra_input_values().at(param_name);
auto extra_input = ggml_decoder->get_model_extra_inputs().find(param_name);
if (extra_input != ggml_decoder->get_model_extra_inputs().end()) {
input_tensor = ov::Tensor(extra_input->second.type, extra_input->second.shape);
*input_tensor.data<int64_t>() = extra_input->second.value;
} else {
input_tensor = convert_ggml_input_to_ov(ggml_decoder, param_name);
}
Expand Down
Loading