Skip to content

[NPU] Fix: Caching Properties in offline compilation #33909

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
sbutnari wants to merge 5 commits into
base: master
Choose a base branch
from
+21 −12
Open

[NPU] Fix: Caching Properties in offline compilation #33909

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
13eb9ca
Remove property as it could be read in offline compilation environments
sbutnari Jan 30, 2026
f964b78
Capture properties throw and return default value
sbutnari Jan 30, 2026
5dec351
Capture throw from deviceName and return default value, moved from me…
sbutnari Feb 2, 2026
156d013
Apply suggestions from code review
sbutnari Feb 2, 2026
b83ae26
Add warning messages
sbutnari Feb 2, 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
3 changes: 2 additions & 1 deletion src/plugins/intel_npu/src/plugin/include/properties.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,7 +67,7 @@ class Properties final {
std::map<std::string, std::tuple<bool, ov::PropertyMutability, std::function<ov::Any(const Config&)>>> _properties;
std::vector<ov::PropertyName> _supportedProperties;

// internal registration functions basd on client object
// internal registration functions based on client object
void registerPluginProperties();
void registerCompiledModelProperties();

Expand All @@ -89,6 +89,7 @@ class Properties final {
ov::intel_npu::dynamic_shape_to_static.name(),
ov::intel_npu::enable_strides_for.name(),
ov::intel_npu::max_tiles.name(),
ov::intel_npu::stepping.name(),
ov::intel_npu::tiles.name(),
ov::intel_npu::turbo.name(),
ov::intel_npu::qdq_optimization.name(),
Expand Down
24 changes: 16 additions & 8 deletions src/plugins/intel_npu/src/plugin/src/properties.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -393,19 +393,27 @@ void Properties::registerPluginProperties() {

TRY_REGISTER_CUSTOMFUNC_PROPERTY(ov::intel_npu::stepping, STEPPING, [&](const Config& config) {
if (!config.has<STEPPING>()) {
const auto specifiedDeviceName = get_specified_device_name(config);
return static_cast<int64_t>(_metrics->GetSteppingNumber(specifiedDeviceName));
} else {
return config.get<STEPPING>();
try {
const auto specifiedDeviceName = get_specified_device_name(config);
return static_cast<int64_t>(_metrics->GetSteppingNumber(specifiedDeviceName));
} catch (...) {
Logger("Properties", ov::log::Level::WARNING)
.warning("Metrics GetSteppingNumber failed to get value from device.");
}
}
return config.get<STEPPING>();
});
TRY_REGISTER_CUSTOMFUNC_PROPERTY(ov::intel_npu::max_tiles, MAX_TILES, [&](const Config& config) {
if (!config.has<MAX_TILES>()) {
const auto specifiedDeviceName = get_specified_device_name(config);
return static_cast<int64_t>(_metrics->GetMaxTiles(specifiedDeviceName));
} else {
return config.get<MAX_TILES>();
try {
const auto specifiedDeviceName = get_specified_device_name(config);
return static_cast<int64_t>(_metrics->GetMaxTiles(specifiedDeviceName));
} catch (...) {
Logger("Properties", ov::log::Level::WARNING)
.warning("Metrics GetMaxTiles failed to get value from device.");
}
}
return config.get<MAX_TILES>();
});

TRY_REGISTER_VARPUB_PROPERTY(ov::intel_npu::run_inferences_sequentially, RUN_INFERENCES_SEQUENTIALLY, [&] {
Expand Down
6 changes: 3 additions & 3 deletions ...ns/intel_npu/tests/functional/shared_tests_instances/behavior/ov_plugin/caching_tests.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -125,7 +125,7 @@ const std::vector<ov::AnyMap> cachingProperties = {
{ov::hint::execution_mode(ov::hint::ExecutionMode::PERFORMANCE)},
{ov::hint::inference_precision(ov::element::i8)},
{ov::hint::performance_mode(ov::hint::PerformanceMode::THROUGHPUT)},
{ov::intel_npu::batch_compiler_mode_settings("NPU_BATCH_SIZE=4")},
{ov::intel_npu::batch_compiler_mode_settings("batch-compile-method=unroll")},
{ov::intel_npu::batch_mode(ov::intel_npu::BatchMode::COMPILER)},
{ov::intel_npu::compilation_mode("ReferenceSW")},
{ov::intel_npu::compilation_mode_params("dummy-op-replacement=true")},
Expand All @@ -134,11 +134,11 @@ const std::vector<ov::AnyMap> cachingProperties = {
{ov::intel_npu::dma_engines(1)},
{ov::intel_npu::dynamic_shape_to_static("true")},
{ov::intel_npu::max_tiles(64)},
{ov::intel_npu::stepping(1)},
{ov::intel_npu::tiles(2)},
{ov::intel_npu::turbo(true)},
{ov::intel_npu::qdq_optimization(true)},
{ov::intel_npu::qdq_optimization_aggressive(true)}
};
{ov::intel_npu::qdq_optimization_aggressive(true)}};

INSTANTIATE_TEST_SUITE_P(smoke_BehaviorTests_CachingSupportCase_NPU_Check_Config,
CompileModelLoadFromCacheTest,
Expand Down
Loading