84 lines
2.6 KiB
C++
84 lines
2.6 KiB
C++
|
// Copyright (C) 2018-2025 Intel Corporation
|
||
|
|
// SPDX-License-Identifier: Apache-2.0
|
||
|
|
//
|
||
|
|
|
||
|
|
#pragma once
|
||
|
|
|
||
|
|
#include "openvino/op/op.hpp"
|
||
|
|
#include "openvino/op/util/unary_elementwise_arithmetic.hpp"
|
||
|
|
|
||
|
|
namespace ov {
|
||
|
|
namespace op {
|
||
|
|
namespace v0 {
|
||
|
|
/// \brief Gaussian Error Linear Unit
|
||
|
|
/// f(x) = 0.5 * x * (1 + erf( x / sqrt(2) )
|
||
|
|
/// \ingroup ov_ops_cpp_api
|
||
|
|
class OPENVINO_API Gelu : public util::UnaryElementwiseArithmetic {
|
||
|
|
public:
|
||
|
|
OPENVINO_OP("Gelu", "opset2", util::UnaryElementwiseArithmetic);
|
||
|
|
|
||
|
|
Gelu();
|
||
|
|
/// \brief Constructs a Gelu operation.
|
||
|
|
///
|
||
|
|
/// \param data Input tensor
|
||
|
|
Gelu(const Output<Node>& data);
|
||
|
|
|
||
|
|
void validate_and_infer_types() override;
|
||
|
|
|
||
|
|
std::shared_ptr<Node> clone_with_new_inputs(const OutputVector& new_args) const override;
|
||
|
|
};
|
||
|
|
} // namespace v0
|
||
|
|
|
||
|
|
/// \brief Specifies the approximation to calculate Gelu
|
||
|
|
enum class GeluApproximationMode { TANH, ERF };
|
||
|
|
OPENVINO_API std::ostream& operator<<(std::ostream& s, const GeluApproximationMode& type);
|
||
|
|
|
||
|
|
namespace v7 {
|
||
|
|
/// \brief Gaussian Error Linear Unit
|
||
|
|
/// f(x) = 0.5 * x * (1 + erf( x / sqrt(2) ) for "approximation" = "erf"
|
||
|
|
/// f(x) = 0.5 * x * (1 + tanh([sqrt(2 / pi)] * [x + 0.044715^3]) for "approximation" =
|
||
|
|
/// "tanh"
|
||
|
|
/// \ingroup ov_ops_cpp_api
|
||
|
|
class OPENVINO_API Gelu : public util::UnaryElementwiseArithmetic {
|
||
|
|
public:
|
||
|
|
OPENVINO_OP("Gelu", "opset7", util::UnaryElementwiseArithmetic);
|
||
|
|
|
||
|
|
Gelu() = default;
|
||
|
|
/// \brief Constructs a Gelu operation.
|
||
|
|
///
|
||
|
|
/// \param data Input tensor
|
||
|
|
/// \param mode Approximation mode
|
||
|
|
Gelu(const Output<Node>& data, GeluApproximationMode mode = GeluApproximationMode::ERF);
|
||
|
|
|
||
|
|
bool visit_attributes(AttributeVisitor& visitor) override;
|
||
|
|
|
||
|
|
void validate_and_infer_types() override;
|
||
|
|
|
||
|
|
bool evaluate(TensorVector& outputs, const TensorVector& inputs) const override;
|
||
|
|
bool has_evaluate() const override;
|
||
|
|
|
||
|
|
std::shared_ptr<Node> clone_with_new_inputs(const OutputVector& new_args) const override;
|
||
|
|
|
||
|
|
GeluApproximationMode get_approximation_mode() const;
|
||
|
|
void set_approximation_mode(const GeluApproximationMode& approximation_mode) {
|
||
|
|
m_approximation_mode = approximation_mode;
|
||
|
|
}
|
||
|
|
|
||
|
|
private:
|
||
|
|
GeluApproximationMode m_approximation_mode = GeluApproximationMode::ERF;
|
||
|
|
};
|
||
|
|
} // namespace v7
|
||
|
|
} // namespace op
|
||
|
|
|
||
|
|
template <>
|
||
|
|
class OPENVINO_API AttributeAdapter<op::GeluApproximationMode>
|
||
|
|
: public EnumAttributeAdapterBase<op::GeluApproximationMode> {
|
||
|
|
public:
|
||
|
|
AttributeAdapter(op::GeluApproximationMode& value) : EnumAttributeAdapterBase<op::GeluApproximationMode>(value) {}
|
||
|
|
|
||
|
|
~AttributeAdapter() override;
|
||
|
|
|
||
|
|
OPENVINO_RTTI("AttributeAdapter<ov::op::GeluApproximationMode>");
|
||
|
|
};
|
||
|
|
} // namespace ov
|