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

[LLM] [NPU] StaticLLMPipeline: support weightless caching #1635

Open
wants to merge 4 commits into
base: master
Choose a base branch
from
+24 −3
Open

[LLM] [NPU] StaticLLMPipeline: support weightless caching #1635

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
d67bad2
Depending on the cache mode call different compile_model overload
smirnov-alexey Jan 27, 2025
2575a64
Merge branch 'master' of https://github.com/openvinotoolkit/openvino.…
smirnov-alexey Jan 27, 2025
a62b527
Address review comment
smirnov-alexey Jan 31, 2025
aa613b3
Add config to import
smirnov-alexey Jan 31, 2025
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
22 changes: 19 additions & 3 deletions src/cpp/src/llm_pipeline_static.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -712,16 +712,23 @@ StatefulLLMPipeline::StatefulLLMPipeline(
if (!fin.is_open()) {
OPENVINO_THROW("Blob file can't be opened: " + blob_path);
}
auto compiled = genai::utils::singleton_core().import_model(fin, device, {});
auto compiled = genai::utils::singleton_core().import_model(fin, device, config);
m_max_prompt_len = compiled.get_property("NPUW_LLM_MAX_PROMPT_LEN").as<uint32_t>();
auto min_resp_len = compiled.get_property("NPUW_LLM_MIN_RESPONSE_LEN").as<uint32_t>();
m_kvcache_total = m_max_prompt_len + min_resp_len;
m_request = compiled.create_infer_request();
} else {
auto model = genai::utils::singleton_core().read_model(models_path / "openvino_model.xml", {}, config);
ModelConfigDesc model_desc = get_modeldesc_from_json(models_path / "config.json");
ov::AnyMap properties = config;
auto compiled = setupAndCompileModel(model, model_desc, properties);
std::shared_ptr<ov::CompiledModel> compiled;
// CACHE_DIR + weightless flow support
auto cache_mode = get_option<CacheMode>(config, "CACHE_MODE");
if (cache_mode.has_value() && *cache_mode == CacheMode::OPTIMIZE_SIZE) {
compiled = setupAndCompileModel(models_path / "openvino_model.xml", model_desc, properties);
} else {
auto model = genai::utils::singleton_core().read_model(models_path / "openvino_model.xml", {}, config);
compiled = setupAndCompileModel(model, model_desc, properties);
}
// Also export compiled model if required
if (export_blob) {
if (blob_path.empty()) {
Expand Down Expand Up @@ -802,6 +809,15 @@ std::shared_ptr<ov::CompiledModel> StatefulLLMPipeline::setupAndCompileModel(
return std::make_shared<ov::CompiledModel>(genai::utils::singleton_core().compile_model(model, "NPU", pipeline_config));
}

std::shared_ptr<ov::CompiledModel> StatefulLLMPipeline::setupAndCompileModel(
const std::filesystem::path& model_path,
const ModelConfigDesc& model_desc,
ov::AnyMap& pipeline_config) {
updateStatefulConfig(model_desc, pipeline_config);

return std::make_shared<ov::CompiledModel>(genai::utils::singleton_core().compile_model(model_path, "NPU", pipeline_config));
}

DecodedResults StatefulLLMPipeline::generate(
StringInputs inputs,
OptionalGenerationConfig generation_config,
Expand Down
5 changes: 5 additions & 0 deletions src/cpp/src/llm_pipeline_static.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,6 +59,11 @@ class StatefulLLMPipeline : public LLMPipelineImplBase {
const ModelConfigDesc& model_desc,
ov::AnyMap& pipeline_config);

std::shared_ptr<ov::CompiledModel> setupAndCompileModel(
const std::filesystem::path& model_path,
const ModelConfigDesc& model_desc,
ov::AnyMap& pipeline_config);

void updateStatefulConfig(
const ModelConfigDesc& model_desc,
ov::AnyMap& pipeline_config);
Expand Down
Loading