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.

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: Compiler DQ update #1515

Merged
merged 8 commits into from
Jan 14, 2025
Merged
Changes from all commits
Commits
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
47 changes: 27 additions & 20 deletions src/cpp/src/llm_pipeline_static.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -475,17 +475,16 @@ std::optional<NPUDesc> extract_npu_descriptor(ov::Core& core) {
}
const auto arch = core.get_property("NPU", ov::device::architecture);
const auto max_tiles = core.get_property("NPU", ov::intel_npu::max_tiles);

bool compiler_dq = false;
const auto device_caps = core.get_property("NPU", ov::device::capabilities);
if (std::find(device_caps.begin(), device_caps.end(),
"COMPILER_DYNAMIC_QUANTIZATION") != device_caps.end()) {
const auto supported_properties = core.get_property("NPU", ov::supported_properties);
if (std::find(supported_properties.begin(), supported_properties.end(),

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This looks like a sub-string search. Perhaps there is some OV utility to tokenize the list first?

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Seems already simple enough

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How it is a sub-string search, looks more like a container. At least std::find works on it (this auto sometimes makes things harder to understand)

"NPU_COMPILER_DYNAMIC_QUANTIZATION") != supported_properties.end()) {
compiler_dq = true;
}
return std::make_optional(NPUDesc{arch, max_tiles, compiler_dq});
}

ov::AnyMap get_baseline_common_config() {
ov::AnyMap get_baseline_common_config(const std::optional<NPUDesc>& npudesc) {
ov::AnyMap config = {
{ "NPU_COMPILATION_MODE_PARAMS", "compute-layers-with-higher-precision=Sqrt,Power,ReduceMean,Add_RMSNorm" },
{ "NPUW_DEVICES", "NPU" },
Expand All @@ -497,11 +496,20 @@ ov::AnyMap get_baseline_common_config() {
{ "NPUW_SLICE_OUT", "YES" },
{ "NPUW_FUNCALL_ASYNC", "YES" }
};
// FIXME: this config logic is getting more and more complex
if (npudesc.has_value() && npudesc->compiler_dq) {
config.emplace("NPUW_DQ", "YES");
config.emplace("NPUW_DQ_FULL", "NO");
config.emplace("NPU_COMPILER_DYNAMIC_QUANTIZATION", "YES");
config.erase("NPUW_DCOFF_TYPE");
config.erase("NPUW_DCOFF_SCALE");
}
return config;
}

ov::AnyMap get_default_common_config(const std::shared_ptr<ov::Model>& model) {
auto config = get_baseline_common_config();
ov::AnyMap get_default_common_config(const std::shared_ptr<ov::Model>& model,
const std::optional<NPUDesc>& npudesc) {
auto config = get_baseline_common_config(npudesc);
const char* npu_l0 = std::getenv("DISABLE_OPENVINO_GENAI_NPU_L0");
if (npu_l0 && std::atoi(npu_l0) == 1) {
config.emplace("NPUW_WEIGHTS_BANK_ALLOC", "CPU");
Expand All @@ -513,40 +521,39 @@ ov::AnyMap get_default_common_config(const std::shared_ptr<ov::Model>& model) {

ov::AnyMap get_default_prefill_config(const std::shared_ptr<ov::Model>& model,
const std::optional<NPUDesc>& npudesc) {
auto config = get_default_common_config(model);
if (is_cw_compressed(model)) {
config.emplace("NPUW_DQ", "YES");
} else {
config.emplace("NPUW_PMM", "NO");
}
auto config = get_default_common_config(model, npudesc);
if (npudesc.has_value() &&
npudesc->arch == "4000" &&
npudesc->max_tiles != -1) {
config.emplace("NPU_DPU_GROUPS", npudesc->max_tiles);
}
if (npudesc.has_value() && npudesc->compiler_dq) {
config.emplace("NPUW_DQ_FULL", "NO");
// Specify NPUW DQ if Compiler DQ is not enabled
if (!npudesc.has_value() || !npudesc->compiler_dq) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not gonna lie @TolyaTalamanov, it was much better WITH that tiny one-liner than without that.

if (is_cw_compressed(model)) {
config.emplace("NPUW_DQ", "YES");
} else {
config.emplace("NPUW_PMM", "NO");
}
}
return config;
}

ov::AnyMap get_default_generate_config(const std::shared_ptr<ov::Model>& model,
const std::optional<NPUDesc>& npudesc,
const GenerateHint hint) {
auto config = get_default_common_config(model);
auto config = get_default_common_config(model, npudesc);
if (hint == GenerateHint::BEST_PERF) {
config.emplace("NPUW_ONLINE_PIPELINE", "NONE");
}
// NB: Unconditionally set for generation model
config.emplace("NPUW_DQ", "YES");
if (npudesc.has_value() && npudesc->arch == "4000") {
config.emplace("NPU_DPU_GROUPS", 4);
}
if (hint == GenerateHint::FAST_COMPILE) {
config.emplace("NPUW_UNFOLD_IREQS", "YES");
}
if (npudesc.has_value() && npudesc->compiler_dq) {
config.emplace("NPUW_DQ_FULL", "NO");
// Specify NPUW DQ if Compiler DQ is not enabled
if (!npudesc.has_value() || !npudesc->compiler_dq) {
config.emplace("NPUW_DQ", "YES");
}
return config;
}
Expand Down
Loading