Skip to content
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

[CPU]Check runtime_options from IR model #27765

Merged
merged 7 commits into from
Nov 30, 2024
Merged

[CPU]Check runtime_options from IR model #27765

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
247441f
[CPU]Check runtime_options from IR model
zhangYiIntel Nov 27, 2024
c6eeb4d
fix code style
zhangYiIntel Nov 27, 2024
0d95480
Revert "[CPU]Check runtime_options from IR model"
zhangYiIntel Nov 28, 2024
b64d31b
[CPU]move runtim_option check to cpu plugin
zhangYiIntel Nov 28, 2024
88fec83
apply review comments
zhangYiIntel Nov 28, 2024
5450057
[CPU]make Config check runtime_options
zhangYiIntel Nov 29, 2024
3949fc8
Merge branch 'master' into yi3/check_u8_rtinfo
zhangYiIntel Nov 30, 2024
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
21 changes: 19 additions & 2 deletions src/inference/src/dev/core_impl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -736,7 +736,7 @@ ov::SoPtr<ov::ICompiledModel> ov::CoreImpl::compile_model(const std::shared_ptr<
ov::AnyMap config_with_batch = config;
// if auto-batching is applicable, the below function will patch the device name and config accordingly:
auto model = apply_auto_batching(model_, deviceName, config_with_batch);

apply_rt_info(model_, config_with_batch);
auto parsed = parseDeviceNameIntoConfig(deviceName, config_with_batch, is_proxy_device(device_name));
auto plugin = get_plugin(parsed._deviceName);
ov::SoPtr<ov::ICompiledModel> res;
Expand Down Expand Up @@ -769,7 +769,7 @@ ov::SoPtr<ov::ICompiledModel> ov::CoreImpl::compile_model(const std::shared_ptr<
ov::AnyMap config_with_batch = config;
// if auto-batching is applicable, the below function will patch the device name and config accordingly:
auto model = apply_auto_batching(model_, deviceName, config_with_batch);

apply_rt_info(model_, config_with_batch);
auto parsed = parseDeviceNameIntoConfig(deviceName, config_with_batch, is_proxy_device(deviceName));
auto plugin = get_plugin(parsed._deviceName);
ov::SoPtr<ov::ICompiledModel> res;
Expand Down Expand Up @@ -1098,6 +1098,23 @@ std::shared_ptr<const ov::Model> ov::CoreImpl::apply_auto_batching(const std::sh
return ov::details::apply_batch_affinity(model, deviceNameWithoutBatch);
}

void ov::CoreImpl::apply_rt_info(const std::shared_ptr<const ov::Model>& model, ov::AnyMap& config) const {
if (model->has_rt_info({"runtime_options", "KV_CACHE_PRECISION"})) {
ilya-lavrenov marked this conversation as resolved.
Show resolved Hide resolved
if (config.find("KV_CACHE_PRECISION") == config.end()) {
const auto kv_cache_precision =
model->get_rt_info<ov::element::Type>({"runtime_options", "KV_CACHE_PRECISION"});
config.insert(ov::hint::kv_cache_precision(kv_cache_precision));
}
}
if (model->has_rt_info({"runtime_options", "DYNAMIC_QUANTIZATION_GROUP_SIZE"})) {
if (config.find("DYNAMIC_QUANTIZATION_GROUP_SIZE") == config.end()) {
const auto dyn_quant_group_size =
model->get_rt_info<uint64_t>({"runtime_options", "DYNAMIC_QUANTIZATION_GROUP_SIZE"});
config.insert(ov::hint::dynamic_quantization_group_size(dyn_quant_group_size));
}
}
}

void ov::CoreImpl::set_property(const std::string& device_name, const AnyMap& properties) {
OPENVINO_ASSERT(device_name.find("HETERO:") != 0,
"set_property is supported only for HETERO itself (without devices). "
Expand Down
6 changes: 6 additions & 0 deletions src/inference/src/dev/core_impl.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -200,6 +200,12 @@ class CoreImpl : public ov::ICore, public std::enable_shared_from_this<ov::ICore
std::string& deviceName,
ov::AnyMap& config) const;

/*
* @brief Apply rt_info from IR model
*/
void apply_rt_info(const std::shared_ptr<const ov::Model>& model,
ov::AnyMap& config) const;

/*
* @brief Register plugins according to the build configuration
*/
Expand Down
14 changes: 14 additions & 0 deletions src/plugins/intel_cpu/tests/functional/custom/behavior/ov_executable_network/properties.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -327,4 +327,18 @@ TEST_F(OVClassConfigTestCPU, smoke_CpuExecNetworkCheckCPUExecutionDevice) {
ASSERT_EQ(value.as<std::string>(), "CPU");
}

TEST_F(OVClassConfigTestCPU, smoke_CpuExecNetworkCheckCPURuntimOptions) {
zhangYiIntel marked this conversation as resolved.
Show resolved Hide resolved
ov::Core ie;
ov::Any type;
ov::Any size;
ov::CompiledModel compiledModel;
model->set_rt_info("f16", "runtime_options", "KV_CACHE_PRECISION");
model->set_rt_info("0", "runtime_options", "DYNAMIC_QUANTIZATION_GROUP_SIZE");
OV_ASSERT_NO_THROW(compiledModel = ie.compile_model(model, deviceName));
OV_ASSERT_NO_THROW(type = compiledModel.get_property(ov::hint::kv_cache_precision));
OV_ASSERT_NO_THROW(size = compiledModel.get_property(ov::hint::dynamic_quantization_group_size));
ASSERT_EQ(type.as<ov::element::Type>(), ov::element::f16);
ASSERT_EQ(size.as<uint64_t>(), 0);
}

} // namespace
Loading