Fix Autocomplete To Use CPU When No GPU Found (#118)

* fix autocomplete to use cpu instance when no gpu found

* default to kind cpu in autocomplete; took out KIND_AUTO check in autocomplete

* simplified logic when setting the instance group kind in autocomplete

Author: nv-kmcgill53

src/onnxruntime.cc

@@ -682,10 +682,36 @@ ModelState::AutoCompleteConfig()
   OrtAllocator* default_allocator;
   std::string model_path;
   {
+    TRITONSERVER_InstanceGroupKind kind = TRITONSERVER_INSTANCEGROUPKIND_CPU;
+
+#ifdef TRITON_ENABLE_GPU
+    triton::common::TritonJson::Value instance_group;
+    ModelConfig().Find("instance_group", &instance_group);
+
+    // Earlier in the model lifecycle, device checks for the instance group
+    // have already occurred. If at least one instance group with
+    // "kind" = "KIND_GPU" then allow model to use GPU else autocomplete to
+    // "KIND_CPU"
+    for (size_t i = 0; i < instance_group.ArraySize(); ++i) {
+      triton::common::TritonJson::Value instance_obj;
+      instance_group.IndexAsObject(i, &instance_obj);
+
+      triton::common::TritonJson::Value instance_group_kind;
+      instance_obj.Find("kind", &instance_group_kind);
+      std::string kind_str;
+      RETURN_IF_ERROR(instance_group_kind.AsString(&kind_str));
+
+      if (kind_str == "KIND_GPU") {
+        kind = TRITONSERVER_INSTANCEGROUPKIND_GPU;
+        break;
+      }
+    }
+#endif  // TRITON_ENABLE_GPU
+
     OrtSession* sptr = nullptr;
     RETURN_IF_ERROR(LoadModel(
-        artifact_name, TRITONSERVER_INSTANCEGROUPKIND_AUTO, 0, &model_path,
-        &sptr, &default_allocator, nullptr));
+        artifact_name, kind, 0, &model_path, &sptr, &default_allocator,
+        nullptr));
     session.reset(sptr);
   }
   OnnxTensorInfoMap input_tensor_infos;