change the default value for session option ep.context_embed_mode to …

…0 since it avoid the model loading memory overhead
microsoft · Dec 10, 2024 · 2bf03ae · 2bf03ae
1 parent bf4d3e1
commit 2bf03ae
Show file tree

Hide file tree

Showing 8 changed files with 12 additions and 9 deletions.
diff --git a/include/onnxruntime/core/session/onnxruntime_session_options_config_keys.h b/include/onnxruntime/core/session/onnxruntime_session_options_config_keys.h
@@ -261,8 +261,8 @@ static const char* const kOrtSessionOptionEpContextEnable = "ep.context_enable";
 static const char* const kOrtSessionOptionEpContextFilePath = "ep.context_file_path";
 
 // Flag to specify whether to dump the EP context into the Onnx model.
-// "0": dump the EP context into separate file, keep the file name in the Onnx model.
-// "1": dump the EP context into the Onnx model. (default).
+// "0": dump the EP context into separate file, keep the file name in the Onnx model. (default).
+// "1": dump the EP context into the Onnx model.
 static const char* const kOrtSessionOptionEpContextEmbedMode = "ep.context_embed_mode";
 
 // Specify the EPContext node name prefix to make it unique

diff --git a/onnxruntime/core/providers/openvino/contexts.h b/onnxruntime/core/providers/openvino/contexts.h
@@ -18,7 +18,7 @@ struct GlobalContext {
   bool is_wholly_supported_graph = false;
   bool enable_opencl_throttling = false;
   bool disable_dynamic_shapes = false;
-  bool ep_context_embed_mode = true;
+  bool ep_context_embed_mode = false;
   bool export_ep_ctx_blob = false;
   bool enable_qdq_optimizer = false;
   bool disable_cpu_fallback = false;

diff --git a/onnxruntime/core/providers/openvino/openvino_execution_provider.h b/onnxruntime/core/providers/openvino/openvino_execution_provider.h
@@ -90,7 +90,7 @@ struct OpenVINOExecutionProviderInfo {
   bool export_ep_ctx_blob_{false};
   bool enable_qdq_optimizer_{false};
   bool disable_cpu_fallback_{false};
-  bool so_epctx_embed_mode_{true};
+  bool so_epctx_embed_mode_{false};
 
   OpenVINOExecutionProviderInfo() = delete;
 

diff --git a/onnxruntime/core/providers/openvino/openvino_provider_factory.cc b/onnxruntime/core/providers/openvino/openvino_provider_factory.cc
@@ -53,7 +53,7 @@ struct OpenVINOProviderFactory : IExecutionProviderFactory {
 std::unique_ptr<IExecutionProvider> OpenVINOProviderFactory::CreateProvider() {
   bool so_disable_cpu_fallback = config_options_.GetConfigOrDefault(kOrtSessionOptionsDisableCPUEPFallback, "0") == "1";
   bool so_export_ep_ctx_blob = config_options_.GetConfigOrDefault(kOrtSessionOptionEpContextEnable, "0") == "1";
-  bool so_epctx_embed_mode = config_options_.GetConfigOrDefault(kOrtSessionOptionEpContextEmbedMode, "1") == "1";
+  bool so_epctx_embed_mode = config_options_.GetConfigOrDefault(kOrtSessionOptionEpContextEmbedMode, "0") == "1";
   std::string so_cache_path = config_options_.GetConfigOrDefault(kOrtSessionOptionEpContextFilePath, "").c_str();
 
   if (so_export_ep_ctx_blob && !so_cache_path.empty()) {

diff --git a/onnxruntime/core/providers/qnn/qnn_execution_provider.cc b/onnxruntime/core/providers/qnn/qnn_execution_provider.cc
@@ -204,7 +204,7 @@ QNNExecutionProvider::QNNExecutionProvider(const ProviderOptions& provider_optio
     LOGS_DEFAULT(VERBOSE) << "Context cache enable: " << context_cache_enabled_;
 
     std::string embed_mode = session_options->config_options.GetConfigOrDefault(
-        kOrtSessionOptionEpContextEmbedMode, "1");
+        kOrtSessionOptionEpContextEmbedMode, "0");
     if ("1" == embed_mode) {
       qnn_context_embed_mode_ = true;
     } else if ("0" == embed_mode) {

diff --git a/onnxruntime/core/providers/vitisai/vitisai_execution_provider.h b/onnxruntime/core/providers/vitisai/vitisai_execution_provider.h
@@ -52,7 +52,7 @@ class VitisAIExecutionProvider : public IExecutionProvider {
   std::shared_ptr<KernelRegistry> registry_;
   // EP context related.
   bool ep_ctx_enabled_ = false;
-  bool ep_ctx_embed_mode_ = true;
+  bool ep_ctx_embed_mode_ = false;
   std::string ep_ctx_model_path_cfg_{""};
   mutable PathString ep_ctx_model_file_loc_{};
   // It might need to be called before loading

diff --git a/onnxruntime/core/session/provider_bridge_ort.cc b/onnxruntime/core/session/provider_bridge_ort.cc
@@ -2242,7 +2242,7 @@ ORT_API_STATUS_IMPL(OrtApis::SessionOptionsAppendExecutionProvider_TensorRT_V2,
       new_tensorrt_options.trt_ep_context_file_path = (context_cache_path.size() == 0) ? nullptr : context_cache_path.c_str();
       LOGS_DEFAULT(VERBOSE) << "User specified context cache path: " << context_cache_path;
 
-      embed_mode = (options->value).config_options.GetConfigOrDefault(kOrtSessionOptionEpContextEmbedMode, "1");
+      embed_mode = (options->value).config_options.GetConfigOrDefault(kOrtSessionOptionEpContextEmbedMode, "0");
       if ("1" == embed_mode) {
         new_tensorrt_options.trt_ep_context_embed_mode = 1;
       } else if ("0" == embed_mode) {

diff --git a/onnxruntime/test/onnx/main.cc b/onnxruntime/test/onnx/main.cc
@@ -454,8 +454,11 @@ int real_main(int argc, char* argv[], Ort::Env& env) {
 
     if (ep_context_enable)
       sf.AddConfigEntry(kOrtSessionOptionEpContextEnable, "1");
-    if (disable_ep_context_embed_mode)
+    if (disable_ep_context_embed_mode) {
       sf.AddConfigEntry(kOrtSessionOptionEpContextEmbedMode, "0");
+    } else {
+      sf.AddConfigEntry(kOrtSessionOptionEpContextEmbedMode, "1");
+    }
 
     for (auto& it : session_config_entries) {
       sf.AddConfigEntry(it.first.c_str(), it.second.c_str());