[ONNX] Added FastGelu from com.microsoft domain (#28715)

Details:
Microsoft Contrib Operator "FastGelu" for ONNX RT

Tickets:
N/A

---------

Co-authored-by: Georgy Krivoruchko <georgy.krivoruchko@intel.com>

Author: vatsalashanubhag

src/frontends/onnx/frontend/src/op/com.microsoft/fast_gelu.cpp

@@ -0,0 +1,52 @@
+// Copyright (C) 2018-2025 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#include "core/operator_set.hpp"
+#include "exceptions.hpp"
+#include "openvino/frontend/exception.hpp"
+#include "openvino/op/add.hpp"
+#include "openvino/op/constant.hpp"
+#include "openvino/op/gelu.hpp"
+#include "utils/common.hpp"
+
+using namespace ov::op;
+
+namespace ov {
+namespace frontend {
+namespace onnx {
+namespace com_microsoft {
+namespace opset_1 {
+
+ov::OutputVector fast_gelu(const ov::frontend::onnx::Node& node) {
+    // Original documentation:
+    // https://github.com/microsoft/onnxruntime/blob/main/docs/ContribOperators.md#commicrosoftfastgelu
+    common::default_op_checks(node, 1);
+
+    const auto inputs = node.get_ov_inputs();
+    const auto& x = inputs[0];
+
+    CHECK_VALID_NODE(node,
+                     x.get_element_type() == ov::element::f16 || x.get_element_type() == ov::element::f32 ||
+                         x.get_element_type() == ov::element::bf16,
+                     "Unsupported input data type for X, expected FP16, FP32, or BF16 but got: ",
+                     x.get_element_type());
+
+    auto input_with_bias = x;
+
+    if (inputs.size() > 1) {
+        auto& bias = inputs[1];
+        input_with_bias = std::make_shared<v1::Add>(x, bias);
+    }
+
+    const auto approximation_mode = ov::op::GeluApproximationMode::TANH;
+    return {std::make_shared<v7::Gelu>(input_with_bias, approximation_mode)};
+}
+
+ONNX_OP("FastGelu", OPSET_SINCE(1), com_microsoft::opset_1::fast_gelu, MICROSOFT_DOMAIN);
+
+}  // namespace opset_1
+}  // namespace com_microsoft
+}  // namespace onnx
+}  // namespace frontend
+}  // namespace ov

src/frontends/onnx/tests/models/com.microsoft/fast_gelu.prototxt

@@ -0,0 +1,66 @@
+ir_version: 7
+producer_name: "OpenVINO ONNX Frontend"
+graph {
+  node {
+    input: "X"
+    input: "bias"
+    output: "Y"
+    op_type: "FastGelu"
+    domain: "com.microsoft"
+  }
+  name: "test_fast_gelu"
+  input {
+    name: "X"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 4
+          }
+          dim {
+            dim_value: 3
+          }
+        }
+      }
+    }
+  }
+  input {
+    name: "bias"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 3
+          }
+        }
+      }
+    }
+  }
+  output {
+    name: "Y"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 4
+          }
+          dim {
+            dim_value: 3
+          }
+        }
+      }
+    }
+  }
+}
+opset_import {
+  version: 1
+}

src/frontends/onnx/tests/onnx_import_com_microsoft.in.cpp

@@ -1683,6 +1683,26 @@ OPENVINO_TEST(${BACKEND_NAME}, onnx_com_microsoft_qlinear_mul) {
     test_case.run();
 }
 
+OPENVINO_TEST(${BACKEND_NAME}, onnx_com_microsoft_fast_gelu) {
+    const auto model = convert_model("com.microsoft/fast_gelu.onnx");
+    auto test_case = ov::test::TestCase(model, s_device);
+
+    const std::vector<float> X = {0.0f,  0.5f, 1.0f,  -0.5f, -1.0f, 2.0f, 1.5f,  -2.0f, 0.3f,  0.7f, -0.7f, 1.2f,
+                                  -0.3f, 0.8f, -1.5f, 1.1f,  -0.9f, 0.4f, -1.2f, 2.1f,  -0.4f, 0.6f, -1.1f, 1.3f};
+    const std::vector<float> bias = {0.1f, -0.1f, 0.2f};
+
+    const std::vector<float> expected_output = {
+        0.05398275f, 0.2621714f, 1.0615974f,  -0.1378286f, -0.1492447f, 2.1699832f, 1.51072f,     -0.0378f,
+        0.34569725f, 0.6304f,    -0.1696f,    1.2874f,     -0.0841f,    0.53025f,   -0.12675f,    1.0614f,
+        -0.159f,     0.4353f,    -0.1492447f, 1.954764f,   -0.0841f,    0.53025f,   -0.13829723f, 1.39957158f};
+
+    test_case.add_input<float>(Shape{2, 4, 3}, X);
+    test_case.add_input<float>(Shape{3}, bias);
+
+    test_case.add_expected_output<float>(Shape{2, 4, 3}, expected_output);
+    test_case.run_with_tolerance_as_fp(0.0055f);
+}
+
 OPENVINO_TEST(${BACKEND_NAME}, onnx_com_microsoft_bias_add) {
     const auto model = convert_model("com.microsoft/bias_add.onnx");
     auto test_case = ov::test::TestCase(model, s_device);