Release captured graph resources when generator is destroyed

src/models/model.cpp

@@ -63,7 +63,8 @@ State::State(const GeneratorParams& params, const Model& model)
     std::random_device rd;
     std::mt19937 gen(rd());
     std::uniform_int_distribution<> dis(1, INT_MAX);
-    graph_id_ = std::to_string(dis(gen));
+    graph_id_value_ = dis(gen);
+    graph_id_ = std::to_string(graph_id_value_);
   }
 }
 
@@ -93,6 +94,7 @@ void State::Run(OrtSession& session, bool graph_capture_this_run) {
   DurationTrace trace{"State::Run"};
 
   if (params_->use_graph_capture) {
+    graph_capture_session_ = &session;
     if (graph_capture_this_run) {
       run_options_->AddConfigEntry("gpu_graph_id", graph_id_.c_str());
     } else {
@@ -222,6 +224,21 @@ void State::SetActiveAdapter(Adapters* adapters, const std::string& adapter_name
 }
 
 State::~State() {
+  // Release captured graph resources in the EP.
+  // Note: the EP contract is to no-op when the id was never captured (e.g., when
+  // graph_capture_this_run was always false for this State). The call is wrapped
+  // in try/catch because destructors must not throw -- a throw during unwinding
+  // would call std::terminate.
+#if ORT_API_VERSION >= 27
+  if (graph_capture_session_ && graph_id_value_ > 0) {
+    try {
+      graph_capture_session_->ReleaseCapturedGraph(graph_id_value_);
+    } catch (...) {
+      // Best-effort cleanup; swallow to keep the destructor non-throwing.
+    }
+  }
+#endif
+
   if (adapters_) {
     for (const auto& adapter_name : adapter_names_) {
       adapters_->ReleaseAdapter(adapter_name);

src/models/model.h

@@ -63,6 +63,11 @@ struct State {
 
  private:
   std::string graph_id_{};
+  int graph_id_value_{0};  // integer form of graph_id_, used to avoid re-parsing in the destructor
+  // Session used for graph capture; not owned. Lifetime invariant: the OrtSession
+  // outlives this State because State is owned by Generator, and Generator is
+  // destroyed before the Model (and its session) that produced it.
+  OrtSession* graph_capture_session_{nullptr};
   std::shared_ptr<Adapters> adapters_;
   ExtraOutputs extra_outputs_;
 };

src/models/onnxruntime_api.h

@@ -786,6 +786,10 @@ struct OrtSession {
 
   void SetEpDynamicOptions(_In_opt_ const char* const* keys, const char* const* values, size_t kv_len);
 
+#if ORT_API_VERSION >= 27
+  void ReleaseCapturedGraph(int graph_annotation_id);  ///< Wraps OrtApi::SessionReleaseCapturedGraph (ORT 1.27+)
+#endif
+
   static void operator delete(void* p) { Ort::api->ReleaseSession(reinterpret_cast<OrtSession*>(p)); }
   Ort::Abstract make_abstract;
 };

src/models/onnxruntime_inline.h

@@ -935,6 +935,12 @@ inline void OrtSession::SetEpDynamicOptions(const char* const* keys, const char*
   Ort::ThrowOnError(Ort::api->SetEpDynamicOptions(this, keys, values, kv_len));
 }
 
+#if ORT_API_VERSION >= 27
+inline void OrtSession::ReleaseCapturedGraph(int graph_annotation_id) {
+  Ort::ThrowOnError(Ort::api->SessionReleaseCapturedGraph(this, graph_annotation_id));
+}
+#endif
+
 inline std::string OrtModelMetadata::GetProducerName() const {
   Ort::StringAllocator string_allocator;
   Ort::ThrowOnError(Ort::api->ModelMetadataGetProducerName(this, &string_allocator, &string_allocator.out));