multihead_attention.hpp

Go to the documentation of this file.

#ifndef MLPACK_METHODS_ANN_LAYER_MULTIHEAD_ATTENTION_HPP
#define MLPACK_METHODS_ANN_LAYER_MULTIHEAD_ATTENTION_HPP

#include <mlpack/prereqs.hpp>
#include <mlpack/methods/ann/layer/softmax.hpp>
#include <mlpack/methods/ann/layer/dropout.hpp>
#include <mlpack/methods/ann/init_rules/glorot_init.hpp>
#include <mlpack/methods/ann/regularizer/no_regularizer.hpp>

namespace mlpack {
namespace ann  {

template <
    typename InputDataType = arma::mat,
    typename OutputDataType = arma::mat,
    typename RegularizerType = NoRegularizer
>
class MultiheadAttention
{
 public:
  MultiheadAttention();

  MultiheadAttention(const size_t tgtSeqLen,
                     const size_t srcSeqLen,
                     const size_t embedDim,
                     const size_t numHeads);

  void Reset();

  template<typename eT>
  void Forward(const arma::Mat<eT>& input, arma::Mat<eT>& output);

  template<typename eT>
  void Backward(const arma::Mat<eT>& /* input */,
                const arma::Mat<eT>& gy,
                arma::Mat<eT>& g);

  template<typename eT>
  void Gradient(const arma::Mat<eT>& input,
                const arma::Mat<eT>& error,
                arma::Mat<eT>& gradient);

  size_t WeightSize() const { return 4 * (embedDim + 1) * embedDim; }

  template<typename Archive>
  void serialize(Archive& ar, const uint32_t /* version */);

  size_t TgtSeqLen() const { return tgtSeqLen; }
  size_t& TgtSeqLen() { return tgtSeqLen; }

  size_t SrcSeqLen() const { return srcSeqLen; }
  size_t& SrcSeqLen() { return srcSeqLen; }

  size_t EmbedDim() const { return embedDim; }
  size_t& EmbedDim() { return embedDim; }

  size_t NumHeads() const { return numHeads; }
  size_t& NumHeads() { return numHeads; }

  OutputDataType const& AttentionMask() const { return attnMask; }
  OutputDataType& AttentionMask() { return attnMask; }

  OutputDataType const& KeyPaddingMask() const { return keyPaddingMask; }
  OutputDataType& KeyPaddingMask() { return keyPaddingMask; }

  OutputDataType const& OutputParameter() const { return outputParameter; }
  OutputDataType& OutputParameter() { return outputParameter; }

  OutputDataType const& Delta() const { return delta; }
  OutputDataType& Delta() { return delta; }

  OutputDataType const& Gradient() const { return grad; }
  OutputDataType& Gradient() { return grad; }

  OutputDataType const& Parameters() const { return weights; }
  OutputDataType& Parameters() { return weights; }

  size_t InputShape() const
  {
    return embedDim * (tgtSeqLen + 2 * srcSeqLen);
  }

 private:
  typedef typename OutputDataType::elem_type ElemType;

  size_t tgtSeqLen;

  size_t srcSeqLen;

  size_t embedDim;

  size_t numHeads;

  size_t headDim;

  OutputDataType attnMask;

  OutputDataType keyPaddingMask;

  OutputDataType queryWt;

  OutputDataType keyWt;

  OutputDataType valueWt;

  OutputDataType outWt;

  OutputDataType qBias;

  OutputDataType kBias;

  OutputDataType vBias;

  OutputDataType outBias;

  OutputDataType weights;

  arma::Cube<ElemType> qProj;

  arma::Cube<ElemType> kProj;

  arma::Cube<ElemType> vProj;

  arma::Cube<ElemType> scores;

  arma::Cube<ElemType> attnOut;

  Softmax<InputDataType, OutputDataType> softmax;

  OutputDataType delta;

  OutputDataType grad;

  OutputDataType outputParameter;

  RegularizerType regularizer;
}; // class MultiheadAttention
} // namespace ann
} // namespace mlpack

// Include implementation.
#include "multihead_attention_impl.hpp"

#endif