Skip to content

[NPU][WS] Use the "ONE_SHOT" weights separation version as default when the compiler-in-plugin is used #32089

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
moslex merged 1 commit into from
Sep 17, 2025
Merged

[NPU][WS] Use the "ONE_SHOT" weights separation version as default when the compiler-in-plugin is used #32089

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
20 changes: 13 additions & 7 deletions src/plugins/intel_npu/src/compiler_adapter/src/plugin_compiler_adapter.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -143,16 +143,22 @@ std::shared_ptr<IGraph> PluginCompilerAdapter::compileWS(const std::shared_ptr<o
return starts_with(name, "main");
};

_logger.info("SEPARATE_WEIGHTS_VERSION: ", config.get<SEPARATE_WEIGHTS_VERSION>());
Config localConfig = config;
if (!localConfig.has<SEPARATE_WEIGHTS_VERSION>()) {
localConfig.update({{ov::intel_npu::separate_weights_version.name(), "ONE_SHOT"}});
}

_logger.info("SEPARATE_WEIGHTS_VERSION: %s",
SEPARATE_WEIGHTS_VERSION::toString(localConfig.get<SEPARATE_WEIGHTS_VERSION>()).c_str());

int64_t compile_model_mem_start = 0;
if (_logger.level() >= ov::log::Level::INFO) {
compile_model_mem_start = get_peak_memory_usage();
}
switch (config.get<SEPARATE_WEIGHTS_VERSION>()) {
switch (localConfig.get<SEPARATE_WEIGHTS_VERSION>()) {
case ov::intel_npu::WSVersion::ONE_SHOT: {
std::vector<std::shared_ptr<NetworkDescription>> initMainNetworkDescriptions =
_compiler->compileWsOneShot(model, config);
_compiler->compileWsOneShot(model, localConfig);

#if 0 // TODO: it is not clear whether we should change the name
OPENVINO_ASSERT(isMain(initMainNetworkDescriptions.back()->metadata.name),
Expand All @@ -170,7 +176,7 @@ std::shared_ptr<IGraph> PluginCompilerAdapter::compileWS(const std::shared_ptr<o
size_t i = 0;

while (auto networkDescription =
std::make_shared<NetworkDescription>(_compiler->compileWsIterative(targetModel, config, i++))) {
std::make_shared<NetworkDescription>(_compiler->compileWsIterative(targetModel, localConfig, i++))) {
if (isInit(networkDescription->metadata.name)) {
initNetworkDescriptions.push_back(networkDescription);
targetModel = originalModel->clone();
Expand All @@ -185,8 +191,8 @@ std::shared_ptr<IGraph> PluginCompilerAdapter::compileWS(const std::shared_ptr<o
}
} break;
default:
OPENVINO_THROW("Invalid \"SEPARATE_WEIGHTS_VERSION\" value found within the \"compileWS\" call:",
config.get<SEPARATE_WEIGHTS_VERSION>());
OPENVINO_THROW("Invalid \"SEPARATE_WEIGHTS_VERSION\" value found within the \"compileWS\" call: ",
localConfig.get<SEPARATE_WEIGHTS_VERSION>());
break;
}

Expand Down Expand Up @@ -244,7 +250,7 @@ std::shared_ptr<IGraph> PluginCompilerAdapter::compileWS(const std::shared_ptr<o
std::move(initNetworkMetadata),
tensorsInits,
model,
config,
localConfig,
/* persistentBlob = */ true, // exporting the blob shall be available in such a scenario
_compiler);
}
Expand Down
Loading