Add disable_host_compilation option to allow disabling host compilation after load.

PiperOrigin-RevId: 764103003

Author: christisg

tensorflow_serving/servables/tensorflow/run_options_base.h

@@ -36,6 +36,11 @@ struct RunOptionsBase {
   // Requests with higher priority always get prioritized for latency over
   // requests with lower priority. 0 is the lowest latency priority.
   int32_t latency_priority = 0;
+
+  // If true, just-in-time host compilation is disabled, and then if the
+  // specified graph is not compiled, request execution will return an error.
+  // Supported only in TFRT servable. Ignored in non-TFRT servable.
+  bool disable_host_compilation = false;
 };
 
 }  // namespace servables

tensorflow_serving/servables/tensorflow/tfrt_servable.cc

@@ -98,6 +98,9 @@ TfrtSavedModelServable::GetTFRTSavedModelRunOptions(
   }
   options.validate_input_specs = config_.validate_input_specs();
   options.validate_input_specs_dry_run = config_.validate_input_specs_dry_run();
+  if (run_options.disable_host_compilation) {
+    options.disable_compilation = true;
+  }
   return options;
 }