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 all 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
10 changes: 10 additions & 0 deletions src/plugins/intel_cpu/src/config.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -460,5 +460,15 @@ void Config::updateProperties() {
_config.insert({ov::hint::num_requests.name(), std::to_string(hintNumRequests)});
}

void Config::applyRtInfo(const std::shared_ptr<const ov::Model>& model) {
if (model->has_rt_info({"runtime_options", ov::hint::kv_cache_precision.name()})) {
this->kvCachePrecision = model->get_rt_info<ov::element::Type>({"runtime_options", ov::hint::kv_cache_precision.name()});
}
if (model->has_rt_info({"runtime_options", ov::hint::dynamic_quantization_group_size.name()})) {
this->fcDynamicQuantizationGroupSize =
model->get_rt_info<uint64_t>({"runtime_options", ov::hint::dynamic_quantization_group_size.name()});
}
}

} // namespace intel_cpu
} // namespace ov
2 changes: 2 additions & 0 deletions src/plugins/intel_cpu/src/config.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,6 +106,8 @@ struct Config {

void updateProperties();

void applyRtInfo(const std::shared_ptr<const ov::Model>& model);

std::map<std::string, std::string> _config;

int modelPreferThreads = -1;
Expand Down
4 changes: 3 additions & 1 deletion src/plugins/intel_cpu/src/plugin.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -247,6 +247,7 @@ std::shared_ptr<ov::ICompiledModel> Plugin::compile_model(const std::shared_ptr<
// update the props after the perf mode translated to configs
// TODO: Clarify the behavior of SetConfig method. Skip eng_config or not?
Config conf = engConfig;
conf.applyRtInfo(cloned_model);
conf.readProperties(config, modelType);

Transformations transformations(cloned_model, conf);
Expand Down Expand Up @@ -520,6 +521,7 @@ ov::SupportedOpsMap Plugin::query_model(const std::shared_ptr<const ov::Model>&

Config conf = engConfig;
Config::ModelType modelType = getModelType(model);
conf.applyRtInfo(model);
conf.readProperties(config, modelType);

auto context = std::make_shared<GraphContext>(conf, fake_w_cache, false);
Expand Down Expand Up @@ -575,7 +577,7 @@ std::shared_ptr<ov::ICompiledModel> Plugin::import_model(std::istream& model_str

Config conf = engConfig;
Config::ModelType modelType = getModelType(model);

conf.applyRtInfo(model);
// check ov::loaded_from_cache property and erase it to avoid exception in readProperties.
auto _config = config;
const auto& it = _config.find(ov::loaded_from_cache.name());
Expand Down
1 change: 0 additions & 1 deletion src/plugins/intel_cpu/src/plugin.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,7 +50,6 @@ class Plugin : public ov::IPlugin {

void get_performance_streams(Config& config, const std::shared_ptr<ov::Model>& model) const;
void calculate_streams(Config& conf, const std::shared_ptr<ov::Model>& model, bool imported = false) const;

Config engConfig;
/* Explicily configured streams have higher priority than performance hints.
So track if streams is set explicitly (not auto-configured) */
Expand Down
31 changes: 31 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 +327,35 @@ 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", ov::hint::kv_cache_precision.name());
model->set_rt_info("0", "runtime_options", ov::hint::dynamic_quantization_group_size.name());
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);
}

TEST_F(OVClassConfigTestCPU, smoke_CpuExecNetworkCheckCPURuntimOptionsWithCompileConfig) {
ov::Core ie;
ov::Any type;
ov::Any size;
ov::CompiledModel compiledModel;
model->set_rt_info("f16", "runtime_options", ov::hint::kv_cache_precision.name());
model->set_rt_info("0", "runtime_options", ov::hint::dynamic_quantization_group_size.name());
ov::AnyMap config;
config[ov::hint::kv_cache_precision.name()] = "u8";
config[ov::hint::dynamic_quantization_group_size.name()] = "16";
OV_ASSERT_NO_THROW(compiledModel = ie.compile_model(model, deviceName, config));
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::u8);
ASSERT_EQ(size.as<uint64_t>(), 16);
}

} // namespace
Loading