diff --git a/scripts/generate_wrappers.py b/scripts/generate_wrappers.py
index 2ac41ceca..79b3798db 100644
--- a/scripts/generate_wrappers.py
+++ b/scripts/generate_wrappers.py
@@ -1534,18 +1534,36 @@ def _append_unique(declaration, definition):
         declarations.append(declaration)
         definitions.append(definition)
 
-    _append_unique(
-        f"extern template std::vector<std::size_t> "
-        f"Operator<{op_type}>::active_implementation_indices(Device::Type);",
-        f"template std::vector<std::size_t> "
-        f"Operator<{op_type}>::active_implementation_indices(Device::Type);",
+    active_implementation_declaration = (
+        "template <>\n"
+        "std::vector<std::size_t>\n"
+        f"Operator<{op_type}>::active_implementation_indices(Device::Type);"
+    )
+    active_implementation_definition = (
+        "template <>\n"
+        "std::vector<std::size_t>\n"
+        f"Operator<{op_type}>::active_implementation_indices("
+        "Device::Type dev_type) {\n"
+        f"  return detail::ActiveImplementationIndices<{op_type}>(dev_type);\n"
+        "}"
     )
+    _append_unique(active_implementation_declaration, active_implementation_definition)
 
+    default_implementation_declaration = (
+        "template <>\n"
+        "std::size_t\n"
+        f"Operator<{op_type}>::DefaultImplementationIndex(Device::Type);"
+    )
+    default_implementation_definition = (
+        "template <>\n"
+        "std::size_t\n"
+        f"Operator<{op_type}>::DefaultImplementationIndex("
+        "Device::Type dev_type) {\n"
+        f"  return detail::DefaultImplementationIndex<{op_type}>(dev_type);\n"
+        "}"
+    )
     _append_unique(
-        f"extern template std::size_t "
-        f"Operator<{op_type}>::DefaultImplementationIndex(Device::Type);",
-        f"template std::size_t "
-        f"Operator<{op_type}>::DefaultImplementationIndex(Device::Type);",
+        default_implementation_declaration, default_implementation_definition
     )
 
     for call in operator.calls:
diff --git a/src/operator.h b/src/operator.h
index 18f242fa7..093becb71 100644
--- a/src/operator.h
+++ b/src/operator.h
@@ -157,6 +157,15 @@ struct CacheKeyBuilder {
 template <typename Key, Device::Type kDev>
 struct ActiveImplementations;
 
+namespace detail {
+
+template <typename Key>
+std::vector<std::size_t> ActiveImplementationIndices(Device::Type dev_type);
+template <typename Key>
+std::size_t DefaultImplementationIndex(Device::Type dev_type);
+
+}  // namespace detail
+
 class OperatorBase {
  public:
   virtual ~OperatorBase() = default;
@@ -297,20 +306,7 @@ class Operator : public OperatorBase {
 
   static std::vector<std::size_t> active_implementation_indices(
       Device::Type dev_type) {
-    if (!detail::ListContains(dev_type, ActiveDevices<Key>{})) {
-      return {};
-    }
-
-    std::vector<std::size_t> result;
-    DispatchFunc<ActiveDevices<Key>>(
-        dev_type,
-        [&](auto device_tag) {
-          constexpr Device::Type kDev = decltype(device_tag)::value;
-          result = detail::ListToVector(
-              typename ActiveImplementations<Key, kDev>::type{});
-        },
-        "Operator::active_implementation_indices");
-    return result;
+    return detail::ActiveImplementationIndices<Key>(dev_type);
   }
 
   template <typename... Args>
@@ -334,30 +330,8 @@ class Operator : public OperatorBase {
   static constexpr std::size_t implementation_index_{implementation_index};
 
  private:
-  template <auto first, auto... rest>
-  static constexpr std::size_t FirstActiveImplementationIndex(
-      List<first, rest...>) {
-    return static_cast<std::size_t>(first);
-  }
-
-  static std::size_t FirstActiveImplementationIndex(List<>) {
-    assert(false && "operator has no active implementation for this device");
-    std::abort();
-  }
-
   static std::size_t DefaultImplementationIndex(Device::Type dev_type) {
-    std::size_t default_index{0};
-
-    DispatchFunc<ActiveDevices<Key>>(
-        dev_type,
-        [&](auto device_tag) {
-          constexpr Device::Type kDev = decltype(device_tag)::value;
-          default_index = FirstActiveImplementationIndex(
-              typename ActiveImplementations<Key, kDev>::type{});
-        },
-        "Operator::DefaultImplementationIndex");
-
-    return default_index;
+    return detail::DefaultImplementationIndex<Key>(dev_type);
   }
 
   static Config DefaultConfig(Device::Type dev_type) {
@@ -458,6 +432,54 @@ struct ActiveImplementations {
       Key, kDev, std::make_index_sequence<kMaxImplementations>>::type;
 };
 
+namespace detail {
+
+template <auto first, auto... rest>
+constexpr std::size_t FirstActiveImplementationIndex(List<first, rest...>) {
+  return static_cast<std::size_t>(first);
+}
+
+inline std::size_t FirstActiveImplementationIndex(List<>) {
+  assert(false && "operator has no active implementation for this device");
+  std::abort();
+}
+
+template <typename Key>
+std::size_t DefaultImplementationIndex(Device::Type dev_type) {
+  std::size_t default_index{0};
+
+  DispatchFunc<ActiveDevices<Key>>(
+      dev_type,
+      [&](auto device_tag) {
+        constexpr Device::Type kDev = decltype(device_tag)::value;
+        default_index = FirstActiveImplementationIndex(
+            typename ActiveImplementations<Key, kDev>::type{});
+      },
+      "Operator::DefaultImplementationIndex");
+
+  return default_index;
+}
+
+template <typename Key>
+std::vector<std::size_t> ActiveImplementationIndices(Device::Type dev_type) {
+  if (!ListContains(dev_type, ActiveDevices<Key>{})) {
+    return {};
+  }
+
+  std::vector<std::size_t> result;
+  DispatchFunc<ActiveDevices<Key>>(
+      dev_type,
+      [&](auto device_tag) {
+        constexpr Device::Type kDev = decltype(device_tag)::value;
+        result =
+            ListToVector(typename ActiveImplementations<Key, kDev>::type{});
+      },
+      "Operator::active_implementation_indices");
+  return result;
+}
+
+}  // namespace detail
+
 }  // namespace infini::ops
 
 #endif
diff --git a/tests/test_cpp_api.py b/tests/test_cpp_api.py
index 2cbd2962f..594c68408 100644
--- a/tests/test_cpp_api.py
+++ b/tests/test_cpp_api.py
@@ -18,6 +18,7 @@ def test_cpp_operator_call_instantiation_smoke(tmp_path):
         [
             _compiler("CXX", "c++"),
             "-std=c++17",
+            "-O3",
             "-Werror",
             f"-I{include_dir}",
             str(source),
diff --git a/tests/test_generate_wrappers.py b/tests/test_generate_wrappers.py
index 90e3930a7..f977cec0c 100644
--- a/tests/test_generate_wrappers.py
+++ b/tests/test_generate_wrappers.py
@@ -97,12 +97,22 @@ def test_operator_call_instantiations_externalize_default_implementation_lookup(
         operator
     )
 
-    signature = (
-        "std::size_t "
+    declaration = (
+        "template <>\n"
+        "std::size_t\n"
         "Operator<::infini::ops::Abs>::DefaultImplementationIndex(Device::Type);"
     )
-    assert f"extern template {signature}" in declarations
-    assert f"template {signature}" in definitions
+    definition = (
+        "template <>\n"
+        "std::size_t\n"
+        "Operator<::infini::ops::Abs>::DefaultImplementationIndex("
+        "Device::Type dev_type) {\n"
+        "  return detail::DefaultImplementationIndex<::infini::ops::Abs>("
+        "dev_type);\n"
+        "}"
+    )
+    assert declaration in declarations
+    assert definition in definitions
 
 
 def test_operator_call_instantiations_externalize_active_implementation_query():
@@ -113,12 +123,22 @@ def test_operator_call_instantiations_externalize_active_implementation_query():
         operator
     )
 
-    signature = (
-        "std::vector<std::size_t> "
+    declaration = (
+        "template <>\n"
+        "std::vector<std::size_t>\n"
         "Operator<::infini::ops::Add>::active_implementation_indices(Device::Type);"
     )
-    assert f"extern template {signature}" in declarations
-    assert f"template {signature}" in definitions
+    definition = (
+        "template <>\n"
+        "std::vector<std::size_t>\n"
+        "Operator<::infini::ops::Add>::active_implementation_indices("
+        "Device::Type dev_type) {\n"
+        "  return detail::ActiveImplementationIndices<::infini::ops::Add>("
+        "dev_type);\n"
+        "}"
+    )
+    assert declaration in declarations
+    assert definition in definitions
 
 
 def test_operator_call_instantiations_keep_scalar_and_optional_tensor_overloads_distinct(