[GPU] Read RT info for compile_model with context

Signed-off-by: Vladimir Paramuzov <[email protected]>
vladimir-paramuzov · Jan 22, 2025 · 580f48f · 580f48f
1 parent 0ceac58
commit 580f48f
Show file tree

Hide file tree

Showing 3 changed files with 7 additions and 3 deletions.
diff --git a/src/plugins/intel_gpu/src/graph/program.cpp b/src/plugins/intel_gpu/src/graph/program.cpp
@@ -530,6 +530,7 @@ void program::init_graph() {
             node->get_output_layouts();
         if (node->is_type<lstm_seq>()) {
             _config.set_property(ov::intel_gpu::use_onednn(true));
+            _config.set_property(ov::intel_gpu::queue_type(QueueTypes::in_order));
         }
     }
     // Perform initial shape_of subgraphs markup

diff --git a/src/plugins/intel_gpu/src/plugin/plugin.cpp b/src/plugins/intel_gpu/src/plugin/plugin.cpp
@@ -238,10 +238,12 @@ std::shared_ptr<ov::ICompiledModel> Plugin::compile_model(const std::shared_ptr<
     auto context_impl = get_context_impl(context);
     auto device_id = ov::DeviceIDParser{context_impl->get_device_name()}.get_device_id();
 
-    OPENVINO_ASSERT(m_configs_map.find(device_id) != m_configs_map.end(), "[GPU] LoadExeNetworkImpl: Couldn't find config for GPU with id ", device_id);
+    OPENVINO_ASSERT(m_configs_map.find(device_id) != m_configs_map.end(), "[GPU] compile_model: Couldn't find config for GPU with id ", device_id);
 
     ExecutionConfig config = m_configs_map.at(device_id);
     config.set_user_property(orig_config);
+    if (model->has_rt_info("runtime_options"))
+        config.apply_rt_info(context_impl->get_engine().get_device_info(), model->get_rt_info<ov::AnyMap>("runtime_options"), is_llm(model));
     config.apply_user_properties(context_impl->get_engine().get_device_info());
 
     set_cache_info(model, config);

diff --git a/src/plugins/intel_gpu/src/runtime/execution_config.cpp b/src/plugins/intel_gpu/src/runtime/execution_config.cpp
@@ -264,7 +264,7 @@ void ExecutionConfig::apply_user_properties(const cldnn::device_info& info) {
         }
     }
 
-    if (!is_set_by_user(ov::hint::kv_cache_precision) || get_property(ov::hint::kv_cache_precision) == ov::element::undefined) {
+    if (!is_set_by_user(ov::hint::kv_cache_precision)) {
         if (info.supports_immad) {  // MFDNN-11755
             set_property(ov::hint::kv_cache_precision(get_property(ov::hint::inference_precision)));
         } else {
@@ -274,7 +274,8 @@ void ExecutionConfig::apply_user_properties(const cldnn::device_info& info) {
     }
 
     // Enable dynamic quantization by default for non-systolic platforms
-    if (get_property(ov::hint::dynamic_quantization_group_size) == 0 && !info.supports_immad) {
+    if (!is_set_by_user(ov::hint::dynamic_quantization_group_size) &&
+         get_property(ov::hint::dynamic_quantization_group_size) == 0 && !info.supports_immad) {
         set_property(ov::hint::dynamic_quantization_group_size(32));
     }