* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
* \file nn_norm_ops.h
* \brief
*/
#ifndef OPS_BUILT_IN_OP_PROTO_INC_NN_NORM1_OPS_H_
#define OPS_BUILT_IN_OP_PROTO_INC_NN_NORM1_OPS_H_
#include "graph/operator_reg.h"
namespace ge {
* @brief LayerNormGrad operator interface implementation \n
* calculating: dy, x, variance, mean, gamma \n
* rstd = 1.0 / sqrt(variance + eps) \n
* pd_xl = data_dy*data_gamma \n
* pd_var = np.sum(((-0.5)*pd_xl*(data_x - data_mean)
* np.power(rstd, 3)),
* reduce_axis, keepdims=True) \n
* pd_mean = np.sum(((-1.0)*pd_xl*rstd), reduce_axis, keepdims=True)
* + pd_var*(1.0/m)
* np.sum(((-2.0)*(data_x - data_mean)), reduce_axis, keepdims=True) \n
* pd_x = pd_xl*rstd +
* pd_var*(2.0/m)*(data_x - data_mean) + pd_mean*(1.0/m) \n
* pd_gamma = np.sum((data_dy*(data_x - data_mean)*rstd), param_axis, keepdims=True) \n
* pd_beta = np.sum(data_dy, param_axis, keepdims=True)
*@par Inputs:
*Five inputs, including:
* @li dy: A tensor. The gradient tensor that represents the reverse calculation.
* Must be one of the following types: float16, float32, bfloat16. The format must be ND.
* The shape is equal to the shape of x, that is, [A1, ...,Ai,R1, ...,Rj].
* Has the same type and format as x.
* @li x: A tensor. First input of forward propagation. Must be one of the following types: float16, float32, bfloat16.
* The shape is equal to the shape of dy, that is, [A1, ...,Ai,R1, ...,Rj].
* Has the same type and format as dy.
* @li variance: A tensor. Third output of forward propagation, indicating the variance value of input.
* Must be one of the following types: float16, float32, bfloat16. The format must be ND.
* Has the same shape as mean, which is [A1,...,Ai,1,...,1], where there are j 1s after Ai,
* and j is the length of the axis that requires normalization.
* Has the same type and format as mean.
* @li mean: A tensor. Second output of forward propagation, indicating the mean value of input.
* Must be one of the following types: float16, float32, bfloat16. The format must be ND.
* Has the same shape as variance, which is [A1,...,Ai,1,...,1], where there are j 1s after Ai,
* and j is the length of the axis that requires normalization.
* Has the same type and format as variance.
* @li gamma: A tensor. Indicates the weight tensor.
* Must be one of the following types: float16, float32, bfloat16.
* The format must be ND. Has the same type as x. The shape is [R1,...,Rj].
*@par Outputs:
*Three outputs, including:
* @li pd_x: A tensor. Indicates the first output of the forward calculation, which is the output tensor of the x derivative.
* Must be one of the following types: float16, float32, bfloat16. The format must be ND.
* Has the same type, shape and format as x.
* @li pd_gamma: A tensor. Indicates the output tensor of gamma derivative.
* Must be one of the following types: float16, float32, bfloat16. The format must be ND.
* Has the same type, shape and format as gamma.
* @li pd_beta: A tensor. The bias tensor of forward output.
* Must be one of the following types: float16, float32, bfloat16. The format must be ND.
* Has the same type, shape and format as gamma.
*@par Restrictions:
*Warning: THIS FUNCTION IS EXPERIMENTAL. Please do not use.
*/
REG_OP(LayerNormGrad)
.INPUT(dy, TensorType({DT_FLOAT, DT_FLOAT16, DT_BF16}))
.INPUT(x, TensorType({DT_FLOAT, DT_FLOAT16, DT_BF16}))
.INPUT(variance, TensorType({DT_FLOAT, DT_FLOAT16, DT_BF16}))
.INPUT(mean, TensorType({DT_FLOAT, DT_FLOAT16, DT_BF16}))
.INPUT(gamma, TensorType({DT_FLOAT, DT_FLOAT16, DT_BF16}))
.OUTPUT(pd_x, TensorType({DT_FLOAT, DT_FLOAT16, DT_BF16}))
.OUTPUT(pd_gamma, TensorType({DT_FLOAT, DT_FLOAT16, DT_BF16}))
.OUTPUT(pd_beta, TensorType({DT_FLOAT, DT_FLOAT16, DT_BF16}))
.OP_END_FACTORY_REG(LayerNormGrad)
}
#endif