[GPU] Apply #27407 and #27553 PRs to 2024.5 release (#27665)

### Details:
- PR includes improvement and fix to weightless cache feature added in
2024.5.
 - Changes combine #27407 and #27553 code.

### Tickets:
 - https://jira.devtools.intel.com/browse/CVS-157364

src/inference/src/dev/core_impl.cpp

@@ -1447,6 +1447,18 @@ ov::SoPtr<ov::ICompiledModel> ov::CoreImpl::load_model_from_cache(
 
                 ov::AnyMap update_config = config;
                 update_config[ov::loaded_from_cache.name()] = true;
+
+                if (util::contains(plugin.get_property(ov::supported_properties), ov::weights_path)) {
+                    std::string weights_path = cacheContent.modelPath;
+                    auto pos = weights_path.rfind('.');
+                    if (pos != weights_path.npos && weights_path.substr(pos) == ".xml") {
+                        weights_path = weights_path.substr(0, pos);
+                        weights_path += ".bin";
+                    }
+                    if (ov::util::file_exists(weights_path)) {
+                        update_config[ov::weights_path.name()] = weights_path;
+                    }
+                }
                 compiled_model = context ? plugin.import_model(networkStream, context, update_config)
                                          : plugin.import_model(networkStream, update_config);
             });

src/plugins/intel_gpu/src/plugin/plugin.cpp

@@ -600,7 +600,8 @@ std::vector<ov::PropertyName> Plugin::get_supported_properties() const {
         ov::PropertyName{ov::hint::enable_cpu_pinning.name(), PropertyMutability::RW},
         ov::PropertyName{ov::device::id.name(), PropertyMutability::RW},
         ov::PropertyName{ov::hint::dynamic_quantization_group_size.name(), PropertyMutability::RW},
-        ov::PropertyName{ov::hint::activations_scale_factor.name(), PropertyMutability::RW}
+        ov::PropertyName{ov::hint::activations_scale_factor.name(), PropertyMutability::RW},
+        ov::PropertyName{ov::weights_path.name(), PropertyMutability::RW},
     };
 
     return supported_properties;

src/plugins/intel_gpu/tests/functional/behavior/model_cache.cpp

@@ -34,12 +34,22 @@
 #include "openvino/pass/serialize.hpp"
 
 namespace {
-class CheckWeightlessCacheAccuracy : public ::testing::Test {
+class CheckWeightlessCacheAccuracy : public ::testing::Test,
+                                     public ::testing::WithParamInterface<bool> {
+public:
+    static std::string get_test_case_name(::testing::TestParamInfo<bool> obj) {
+        bool use_compile_model_api = obj.param;
+
+        std::ostringstream result;
+        result << "use_compile_model_api=" << use_compile_model_api;
+        return result.str();
+    }
 protected:
     std::shared_ptr<ov::Model> model;
     std::string xml_path;
     std::string bin_path;
     std::string cache_path;
+    bool use_compile_model_api; // for loading from cache
 
     void SetUp() override;
     void TearDown() override;
@@ -51,6 +61,7 @@ void CheckWeightlessCacheAccuracy::SetUp() {
     xml_path = filePrefix + ".xml";
     bin_path = filePrefix + ".bin";
     cache_path = filePrefix + ".blob";
+    use_compile_model_api = GetParam();
 }
 
 void CheckWeightlessCacheAccuracy::TearDown() {
@@ -74,7 +85,13 @@ void CheckWeightlessCacheAccuracy::run() {
 
     auto ifstr = std::ifstream(cache_path, std::ifstream::binary);
     ov::CompiledModel imported_model;
-    OV_ASSERT_NO_THROW(imported_model = core->import_model(ifstr, ov::test::utils::DEVICE_GPU, config_with_weights_path));
+    if (use_compile_model_api) {
+        OV_ASSERT_NO_THROW(imported_model =
+                               core->compile_model(xml_path, ov::test::utils::DEVICE_GPU, config));
+    } else {
+        OV_ASSERT_NO_THROW(imported_model =
+                               core->import_model(ifstr, ov::test::utils::DEVICE_GPU, config_with_weights_path));
+    }
     ifstr.close();
 
     auto orig_req = compiled_model.create_infer_request();
@@ -98,19 +115,23 @@ void CheckWeightlessCacheAccuracy::run() {
     }
 }
 
-TEST_F(CheckWeightlessCacheAccuracy, ReadConcatSplitAssign) {
+TEST_P(CheckWeightlessCacheAccuracy, ReadConcatSplitAssign) {
     model = ov::test::utils::make_read_concat_split_assign({1, 1, 2, 4}, ov::element::f16);
     run();
 }
 
-TEST_F(CheckWeightlessCacheAccuracy, SingleConcatWithConstant) {
+TEST_P(CheckWeightlessCacheAccuracy, SingleConcatWithConstant) {
     model = ov::test::utils::make_single_concat_with_constant({1, 1, 2, 4}, ov::element::f16);
     run();
 }
 
-TEST_F(CheckWeightlessCacheAccuracy, TiWithLstmCell) {
+TEST_P(CheckWeightlessCacheAccuracy, TiWithLstmCell) {
     model = ov::test::utils::make_ti_with_lstm_cell(ov::element::f16);
     run();
 }
 
+INSTANTIATE_TEST_SUITE_P(smoke_CheckWeightlessCacheAccuracy, CheckWeightlessCacheAccuracy,
+                         ::testing::Bool(),
+                         CheckWeightlessCacheAccuracy::get_test_case_name);
+
 }  // namespace