Enable serialization of predict response as tensor content.

PiperOrigin-RevId: 689691887

Author: tensorflower-gardener

tensorflow_serving/model_servers/BUILD

@@ -497,6 +497,7 @@ cc_library(
         "//tensorflow_serving/servables/tensorflow:get_model_metadata_impl",
         "//tensorflow_serving/servables/tensorflow:multi_inference",
         "//tensorflow_serving/servables/tensorflow:predict_impl",
+        "//tensorflow_serving/servables/tensorflow:predict_response_tensor_serialization_option",
         "//tensorflow_serving/servables/tensorflow:regression_service",
         "//tensorflow_serving/servables/tensorflow:saved_model_bundle_source_adapter",
         "//tensorflow_serving/servables/tensorflow:session_bundle_config_cc_proto",

tensorflow_serving/model_servers/main.cc

@@ -307,7 +307,11 @@ int main(int argc, char** argv) {
                        "Whether to skip auto initializing TPU."),
       tensorflow::Flag("enable_grpc_healthcheck_service",
                        &options.enable_grpc_healthcheck_service,
-                       "Enable the standard gRPC healthcheck service.")};
+                       "Enable the standard gRPC healthcheck service."),
+      tensorflow::Flag(
+          "enable_serialization_as_tensor_content",
+          &options.enable_serialization_as_tensor_content,
+          "Enable serialization of predict response as tensor content.")};
 
   const auto& usage = tensorflow::Flags::Usage(argv[0], flag_list);
   if (!tensorflow::Flags::Parse(&argc, argv, flag_list)) {

tensorflow_serving/model_servers/server.cc

@@ -52,6 +52,7 @@ limitations under the License.
 #include "tensorflow_serving/model_servers/model_platform_types.h"
 #include "tensorflow_serving/model_servers/server_core.h"
 #include "tensorflow_serving/model_servers/server_init.h"
+#include "tensorflow_serving/servables/tensorflow/predict_response_tensor_serialization_option.h"
 #include "tensorflow_serving/servables/tensorflow/session_bundle_config.pb.h"
 #include "tensorflow_serving/servables/tensorflow/thread_pool_factory_config.pb.h"
 #include "tensorflow_serving/servables/tensorflow/util.h"
@@ -321,6 +322,10 @@ Status Server::BuildAndStart(const Options& server_options) {
   options.force_allow_any_version_labels_for_unavailable_models =
       server_options.force_allow_any_version_labels_for_unavailable_models;
   options.enable_cors_support = server_options.enable_cors_support;
+  if (server_options.enable_serialization_as_tensor_content) {
+    options.predict_response_tensor_serialization_option =
+        internal::PredictResponseTensorSerializationOption::kAsProtoContent;
+  }
 
   TF_RETURN_IF_ERROR(ServerCore::Create(std::move(options), &server_core_));
 

tensorflow_serving/model_servers/server.h

@@ -105,6 +105,8 @@ class Server {
     bool skip_initialize_tpu = false;
     // Misc GRPC features
     bool enable_grpc_healthcheck_service = false;
+    // Control whether to serialize predict response as tensor content.
+    bool enable_serialization_as_tensor_content = false;
     Options();
   };
 

tensorflow_serving/servables/tensorflow/BUILD

@@ -501,6 +501,9 @@ cc_test(
 cc_library(
     name = "predict_response_tensor_serialization_option",
     hdrs = ["predict_response_tensor_serialization_option.h"],
+    visibility = [
+        "//visibility:public",
+    ],
 )
 
 cc_library(