StuMYY/笔记/transformer.md

https://zhuanlan.zhihu.com/p/308301901 ##单头Attention的实现

class ScaledDotProductAttention(nn.Module):
    ''' Scaled Dot-Product Attention '''

    def __init__(self, temperature, attn_dropout=0.1):
        super().__init__()
        self.temperature = temperature
        self.dropout = nn.Dropout(attn_dropout)

    def forward(self, q, k, v, mask=None):
        # self.temperature是论文中的d_k ** 0.5，防止梯度过大
        # QxK/sqrt(dk)
        attn = torch.matmul(q / self.temperature, k.transpose(2, 3))

        if mask is not None:
            # 屏蔽不想要的输出
            attn = attn.masked_fill(mask == 0, -1e9)
        # softmax+dropout
        attn = self.dropout(F.softmax(attn, dim=-1))
        # 概率分布xV
        output = torch.matmul(attn, v)

        return output, attn

###多头 Attention

class MultiHeadAttention(nn.Module):
    ''' Multi-Head Attention module '''

    # n_head头的个数，默认是8
    # d_model编码向量长度，例如本文说的512
    # d_k, d_v的值一般会设置为 n_head * d_k=d_model，
    # 此时concat后正好和原始输入一样，当然不相同也可以，因为后面有fc层
    # 相当于将可学习矩阵分成独立的n_head份
    def __init__(self, n_head, d_model, d_k, d_v, dropout=0.1):
        super().__init__()
        # 假设n_head=8，d_k=64
        self.n_head = n_head
        self.d_k = d_k
        self.d_v = d_v
        # d_model输入向量，n_head * d_k输出向量
        # 可学习W^Q，W^K,W^V矩阵参数初始化
        self.w_qs = nn.Linear(d_model, n_head * d_k, bias=False)
        self.w_ks = nn.Linear(d_model, n_head * d_k, bias=False)
        self.w_vs = nn.Linear(d_model, n_head * d_v, bias=False)
        # 最后的输出维度变换操作
        self.fc = nn.Linear(n_head * d_v, d_model, bias=False)
        # 单头自注意力
        self.attention = ScaledDotProductAttention(temperature=d_k ** 0.5)
        self.dropout = nn.Dropout(dropout)
        # 层归一化
        self.layer_norm = nn.LayerNorm(d_model, eps=1e-6)

    def forward(self, q, k, v, mask=None):
        # 假设qkv输入是(b,100,512),100是训练每个样本最大单词个数
        # 一般qkv相等，即自注意力
        residual = q
        # 将输入x和可学习矩阵相乘，得到(b,100,512)输出
        # 其中512的含义其实是8x64，8个head，每个head的可学习矩阵为64维度
        # q的输出是(b,100,8,64),kv也是一样
        q = self.w_qs(q).view(sz_b, len_q, n_head, d_k)
        k = self.w_ks(k).view(sz_b, len_k, n_head, d_k)
        v = self.w_vs(v).view(sz_b, len_v, n_head, d_v)

        # 变成(b,8,100,64)，方便后面计算，也就是8个头单独计算
        q, k, v = q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2)

        if mask is not None:
            mask = mask.unsqueeze(1)   # For head axis broadcasting.
        # 输出q是(b,8,100,64),维持不变,内部计算流程是：
        # q*k转置，除以d_k ** 0.5，输出维度是b,8,100,100即单词和单词直接的相似性
        # 对最后一个维度进行softmax操作得到b,8,100,100
        # 最后乘上V，得到b,8,100,64输出
        q, attn = self.attention(q, k, v, mask=mask)

        # b,100,8,64-->b,100,512
        q = q.transpose(1, 2).contiguous().view(sz_b, len_q, -1)
        q = self.dropout(self.fc(q))
        # 残差计算
        q += residual
        # 层归一化，在512维度计算均值和方差，进行层归一化
        q = self.layer_norm(q)

        return q, attn

class PositionwiseFeedForward(nn.Module):
    ''' A two-feed-forward-layer module '''

    def __init__(self, d_in, d_hid, dropout=0.1):
        super().__init__()
        # 两个fc层，对最后的512维度进行变换
        self.w_1 = nn.Linear(d_in, d_hid) # position-wise
        self.w_2 = nn.Linear(d_hid, d_in) # position-wise
        self.layer_norm = nn.LayerNorm(d_in, eps=1e-6)
        self.dropout = nn.Dropout(dropout)

    def forward(self, x):
        residual = x

        x = self.w_2(F.relu(self.w_1(x)))
        x = self.dropout(x)
        x += residual

        x = self.layer_norm(x)

        return x

EncoderLayer

class EncoderLayer(nn.Module):
    def __init__(self, d_model, d_inner, n_head, d_k, d_v, dropout=0.1):
        super(EncoderLayer, self).__init__()
        self.slf_attn = MultiHeadAttention(n_head, d_model, d_k, d_v, dropout=dropout)
        self.pos_ffn = PositionwiseFeedForward(d_model, d_inner, dropout=dropout)

    def forward(self, enc_input, slf_attn_mask=None):
        # Q K V是同一个，自注意力
        # enc_input来自源单词嵌入向量或者前一个编码器输出
        enc_output, enc_slf_attn = self.slf_attn(
            enc_input, enc_input, enc_input, mask=slf_attn_mask)
        enc_output = self.pos_ffn(enc_output)
        return enc_output, enc_slf_attn

###Encoder层

class Encoder(nn.Module):
    def __init__(
            self, n_src_vocab, d_word_vec, n_layers, n_head, d_k, d_v,
            d_model, d_inner, pad_idx, dropout=0.1, n_position=200):
        # nlp领域的词嵌入向量生成过程(单词在词表里面的索引idx-->d_word_vec长度的向量)
        self.src_word_emb = nn.Embedding(n_src_vocab, d_word_vec, padding_idx=pad_idx)
        # 位置编码
        self.position_enc = PositionalEncoding(d_word_vec, n_position=n_position)
        self.dropout = nn.Dropout(p=dropout)
        # n个编码器层
        self.layer_stack = nn.ModuleList([
            EncoderLayer(d_model, d_inner, n_head, d_k, d_v, dropout=dropout)
            for _ in range(n_layers)])
        # 层归一化
        self.layer_norm = nn.LayerNorm(d_model, eps=1e-6)

    def forward(self, src_seq, src_mask, return_attns=False):
        # 对输入序列进行词嵌入，加上位置编码
        enc_output = self.dropout(self.position_enc(self.src_word_emb(src_seq)))
        enc_output = self.layer_norm(enc_output)
        # 作为编码器层输入
        for enc_layer in self.layer_stack:
            enc_output, _ = enc_layer(enc_output, slf_attn_mask=src_mask)
        return enc_output

###解码的mask

class ScaledDotProductAttention(nn.Module):
    ''' Scaled Dot-Product Attention '''

    def __init__(self, temperature, attn_dropout=0.1):
        super().__init__()
        self.temperature = temperature
        self.dropout = nn.Dropout(attn_dropout)

    def forward(self, q, k, v, mask=None):
        # 假设q是b,8,10,64(b是batch，8是head个数，10是样本最大单词长度，
        # 64是每个单词的编码向量)
        # attn输出维度是b,8,10,10
        attn = torch.matmul(q / self.temperature, k.transpose(2, 3))
        # 故mask维度也是b,8,10,10
        # 忽略b,8，只关注10x10的矩阵，其是下三角矩阵，下三角位置全1，其余位置全0
        if mask is not None:
            # 提前算出mask，将为0的地方变成极小值-1e9，把这些位置的值设置为忽略
            # 目的是避免解码过程中利用到未来信息
            attn = attn.masked_fill(mask == 0, -1e9)
        # softmax+dropout
        attn = self.dropout(F.softmax(attn, dim=-1))
        output = torch.matmul(attn, v)

        return output, attn

decoder layer

class DecoderLayer(nn.Module):
    ''' Compose with three layers '''

    def __init__(self, d_model, d_inner, n_head, d_k, d_v, dropout=0.1):
        super(DecoderLayer, self).__init__()
        self.slf_attn = MultiHeadAttention(n_head, d_model, d_k, d_v, dropout=dropout)
        self.enc_attn = MultiHeadAttention(n_head, d_model, d_k, d_v, dropout=dropout)
        self.pos_ffn = PositionwiseFeedForward(d_model, d_inner, dropout=dropout)

    def forward(
            self, dec_input, enc_output,
            slf_attn_mask=None, dec_enc_attn_mask=None):
        # 标准的自注意力，QKV=dec_input来自目标单词嵌入或者前一个解码器输出
        dec_output, dec_slf_attn = self.slf_attn(
            dec_input, dec_input, dec_input, mask=slf_attn_mask)
        # KV来自最后一个编码层输出enc_output，Q来自带有mask的self.slf_attn输出
        dec_output, dec_enc_attn = self.enc_attn(
            dec_output, enc_output, enc_output, mask=dec_enc_attn_mask)
        dec_output = self.pos_ffn(dec_output)
        return dec_output, dec_slf_attn, dec_enc_attn

decoder 流程

class Decoder(nn.Module):
    def __init__(
            self, n_trg_vocab, d_word_vec, n_layers, n_head, d_k, d_v,
            d_model, d_inner, pad_idx, n_position=200, dropout=0.1):
        # 目标单词嵌入
        self.trg_word_emb = nn.Embedding(n_trg_vocab, d_word_vec, padding_idx=pad_idx)
        # 位置嵌入向量
        self.position_enc = PositionalEncoding(d_word_vec, n_position=n_position)
        self.dropout = nn.Dropout(p=dropout)
        # n个解码器
        self.layer_stack = nn.ModuleList([
            DecoderLayer(d_model, d_inner, n_head, d_k, d_v, dropout=dropout)
            for _ in range(n_layers)])
        # 层归一化
        self.layer_norm = nn.LayerNorm(d_model, eps=1e-6)

    def forward(self, trg_seq, trg_mask, enc_output, src_mask, return_attns=False):
        # 目标单词嵌入+位置编码
        dec_output = self.dropout(self.position_enc(self.trg_word_emb(trg_seq)))
        dec_output = self.layer_norm(dec_output)
        # 遍历每个解码器
        for dec_layer in self.layer_stack:  
            # 需要输入3个信息：目标单词嵌入+位置编码、最后一个编码器输出enc_output
            # 和dec_enc_attn_mask，解码时候不能看到未来单词信息
            dec_output, dec_slf_attn, dec_enc_attn = dec_layer(
                dec_output, enc_output, slf_attn_mask=trg_mask, dec_enc_attn_mask=src_mask)
        return dec_output

主要区别

decoder 模块比encoder 模块 中多了一个 decodermask decoder会mask掉后面的信息