fix engine cache path with EP context

onnxruntime/core/providers/nv_tensorrt_rtx/nv_execution_provider.cc

@@ -2611,8 +2611,6 @@ Status NvExecutionProvider::CreateNodeComputeInfoFromGraph(const GraphViewer& gr
         cache_path = GetCachePath(cache_path_, fused_node.Name()) + ".engine";
         ;
       }
-      auto cache_file_name = std::filesystem::path(cache_path).filename();
-      cache_path = std::filesystem::path(engine_cache_relative_path_to_context_model_dir).append(cache_file_name.string()).string();
       // NV TRT EP per default generates hardware compatible engines for any RTX device with compute capability > 80
       std::string compute_capability_hw_compat = "80+";
       if (!ep_context_model_) {

onnxruntime/core/providers/nv_tensorrt_rtx/onnx_ctx_model_helper.cc

@@ -111,7 +111,8 @@ Status CreateCtxNode(const GraphViewer& graph_viewer,
     }
     attr_ep_cache_context->set_s(engine_data_str);
   } else {
-    attr_ep_cache_context->set_s(engine_cache_path);
+    std::string engine_cache_filename = std::filesystem::path(engine_cache_path).filename().string();
+    attr_ep_cache_context->set_s(engine_cache_filename);
     std::fstream engine_cache_file(engine_cache_path, std::ios::binary | std::ios::out);
     if (engine_cache_file.is_open()) {
       engine_cache_file.write(engine_data, size);