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

[GPU] Fix weightless caching with low precision models #28641

Open
wants to merge 9 commits into
base: master
Choose a base branch
from
+233 −173
Open

[GPU] Fix weightless caching with low precision models #28641

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
5700e0e
[GPU] Fix weightless caching with int4 models
tkrupa-intel Jan 21, 2025
4ee6437
Bring back std::dynamic_pointer_cast for ov::Node
tkrupa-intel Jan 23, 2025
81aca3e
clang-format
tkrupa-intel Jan 23, 2025
1a185cb
Add missing nullptr check
tkrupa-intel Jan 23, 2025
f741706
Remove unnecessary structures and add precision conversion WA
tkrupa-intel Jan 28, 2025
38aceaf
Remove ConvertPrecision WA and use Convert op + ConstantFolding for b…
tkrupa-intel Jan 28, 2025
4bbd8f5
Remove missed field
tkrupa-intel Jan 28, 2025
29b0bd0
Don't force reorder impl
tkrupa-intel Jan 29, 2025
57361a7
Support all possible reorders
tkrupa-intel 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
8 changes: 1 addition & 7 deletions src/common/transformations/src/transformations/convert_precision.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1417,13 +1417,7 @@ bool fuse_type_to_constant(const std::shared_ptr<ov::Node>& node,
new_const->validate_and_infer_types();
new_const->set_friendly_name(constant->get_friendly_name());
ov::copy_runtime_info(constant, new_const);

const auto& rt_info = node->get_rt_info();
auto weightless_caching_attr = rt_info.find(ov::WeightlessCacheAttribute::get_type_info_static());
if (weightless_caching_attr != rt_info.end()) {
new_const->get_rt_info()[ov::WeightlessCacheAttribute::get_type_info_static()] =
weightless_caching_attr->second;
}
ov::copy_weightless_cache_attr(constant, new_const);
return true;
}
return false;
Expand Down
2 changes: 2 additions & 0 deletions src/core/dev_api/openvino/core/rt_info/weightless_caching_attributes.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,8 @@

namespace ov {

OPENVINO_API void copy_weightless_cache_attr(const std::shared_ptr<Node>& from, const std::shared_ptr<Node>& to);

/**
* @brief Holds weightless caching attributes of a single constant.
*
Expand Down
10 changes: 10 additions & 0 deletions src/core/src/op/util/weightless_caching_attributes.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,3 +7,13 @@
bool ov::WeightlessCacheAttribute::is_copyable() const {
return false;
}

OPENVINO_API void ov::copy_weightless_cache_attr(const std::shared_ptr<ov::Node>& from,
const std::shared_ptr<ov::Node>& to) {
const auto& rt_info = from->get_rt_info();
auto weightless_caching_attr = rt_info.find(ov::WeightlessCacheAttribute::get_type_info_static());

if (weightless_caching_attr != rt_info.end()) {
to->get_rt_info()[ov::WeightlessCacheAttribute::get_type_info_static()] = weightless_caching_attr->second;
}
}
2 changes: 2 additions & 0 deletions src/core/src/pass/constant_folding.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@
#include "openvino/cc/pass/itt.hpp"
#include "openvino/core/constant_fold_utils.hpp"
#include "openvino/core/rt_info.hpp"
#include "openvino/core/rt_info/weightless_caching_attributes.hpp"
#include "openvino/op/constant.hpp"
#include "openvino/op/convert.hpp"
#include "openvino/op/util/op_types.hpp"
Expand Down Expand Up @@ -153,6 +154,7 @@ bool ov::pass::ConstantFolding::run_on_model(const std::shared_ptr<ov::Model>& m
copy_runtime_info_from_input_values(original_node);
// Propagate runtime info attributes to replacement
copy_runtime_info(original_node, replacement_ptr);
ov::copy_weightless_cache_attr(original_node, replacement_ptr);

rewritten = true;
}
Expand Down
Loading
Loading