From 14c8a85c32decfb043a3b2320e6ac4d2c9681ef5 Mon Sep 17 00:00:00 2001
From: "Yu, Zijun" <zijun.yu@intel.com>
Date: Wed, 27 Aug 2025 17:06:35 +0800
Subject: [PATCH] Perf: RMS fused to OV internal RMS op

---
 ggml/src/ggml-openvino/openvino/op/rms_norm.cpp | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/ggml/src/ggml-openvino/openvino/op/rms_norm.cpp b/ggml/src/ggml-openvino/openvino/op/rms_norm.cpp
index 211692a3c7..c9df4c42f3 100644
--- a/ggml/src/ggml-openvino/openvino/op/rms_norm.cpp
+++ b/ggml/src/ggml-openvino/openvino/op/rms_norm.cpp
@@ -3,6 +3,7 @@
 #include <openvino/op/constant.hpp>
 #include <openvino/op/divide.hpp>
 #include <openvino/op/multiply.hpp>
+#include <openvino/op/power.hpp>
 #include <openvino/op/reduce_mean.hpp>
 #include <openvino/op/sqrt.hpp>
 
@@ -19,18 +20,17 @@ OutputVector translate_rms_norm(const NodeContext& context) {
     num_inputs_check(context, 1, 1);
 
     auto input_node = context.get_input(0);
-    auto square = std::make_shared<ov::op::v1::Multiply>(input_node, input_node);
+    auto square = std::make_shared<ov::op::v1::Power>(
+        input_node, ov::op::v0::Constant::create(ov::element::f32, ov::Shape{1}, {2.0f}));
 
-    auto mean =
-        std::make_shared<ov::op::v1::ReduceMean>(square,
-                                                 ov::op::v0::Constant::create(ov::element::i64, ov::Shape{1}, {2}),
-                                                 true);
+    auto mean = std::make_shared<ov::op::v1::ReduceMean>(
+        square, ov::op::v0::Constant::create(ov::element::i64, ov::Shape{1}, {-1}), true);
 
     float eps;
     memcpy(&eps, context.get_output_op_params(0), sizeof(float));
 
     auto rms = std::make_shared<ov::op::v0::Sqrt>(
-        std::make_shared<ov::op::v1::Add>(mean, ov::op::v0::Constant::create(ov::element::f32, ov::Shape{}, {eps})));
+        std::make_shared<ov::op::v1::Add>(mean, ov::op::v0::Constant::create(ov::element::f32, ov::Shape{1}, {eps})));
 
     auto reciprocal =
         std::make_shared<ov::op::v1::Divide>(ov::op::v0::Constant::create(ov::element::f32, ov::Shape{1}, {1.0f}), rms);