Transformer代码：适用于语音识别（streaming-mode）

1. 简介

1. 本代码是可运行的transformer代码块，只是基于一组数据（随机生成的数据）来调试，具体业务需要自己封装数据。
2. 对Encoder/Decoder都做了mask处理（上三角矩阵），适用于streaming模式。（比如语音识别）
3. mask未采用chunk-mode，不能偷窥到后面输入，所以性能应该不高（需要自己调chunk-mask）
4. lost-func(损失函数)采用的是KLDIVCross。

2. 网络结构

• Trasnformer
• Encoder
  • MultiHeadAttention
  • poise-embedding
  • mask-matrix
• Decoder
  • MultiHeadAttention
  • poise-embedding
  • mask-matrix
• KL_LOSS

3. 源码

下面展示一些 内联代码片。

"""
  By fangfuping
  2022/7/30
"""
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
import math
import logging
## 10
def get_attn_subsequent_mask(seq):
"""
    seq: [batch_size, tgt_len]
"""
    attn_shape = [seq.size(0), seq.size(1), seq.size(1)]
    # attn_shape: [batch_size, tgt_len, tgt_len]
    subsequence_mask = np.triu(np.ones(attn_shape), k=1)  # 生成一个上三角矩阵
    subsequence_mask = torch.from_numpy(subsequence_mask).byte()
    return subsequence_mask  # [batch_size, tgt_len, tgt_len]
## 7. ScaledDotProductAttention
class ScaledDotProductAttention(nn.Module):
    def __init__(self, d_k):
        super(ScaledDotProductAttention, self).__init__()
        self.d_k = d_k
    def forward(self, Q, K, V):
        ## 输入进来的维度分别是 [batch_size x n_heads x len_q x d_k]  K： [batch_size x n_heads x len_k x d_k]  V: [batch_size x n_heads x len_k x d_v]
        ##首先经过matmul函数得到的scores形状是 : [batch_size x n_heads x len_q x len_k]
        scores = torch.matmul(Q, K.transpose(-1, -2)) / np.sqrt(self.d_k)

        ## 然后关键词地方来了，下面这个就是用到了我们之前重点讲的attn_mask，把被mask的地方置为无限小，softmax之后基本就是0，对q的单词不起作用
        #scores.masked_fill_(attn_mask, -1e9) # Fills elements of self tensor with value where mask is one.

        attn = nn.Softmax(dim=-1)(scores)
        context = torch.matmul(attn, V)
        return context, attn

## 6. MultiHeadAttention
class MultiHeadAttention(nn.Module):
    def __init__(self, d_model, d_k, d_v, n_heads):
        super(MultiHeadAttention, self).__init__()
        self.d_model = d_model
        self.d_k = d_k
        self.d_v = d_v
        self.n_heads = n_heads
        ## 输入进来的QKV是相等的，我们会使用映射linear做一个映射得到参数矩阵Wq, Wk,Wv
        self.W_Q = nn.Linear(d_model, d_k * n_heads)
        self.W_K = nn.Linear(d_model, d_k * n_heads)
        self.W_V = nn.Linear(d_model, d_v * n_heads)
        self.linear = nn.Linear(n_heads * d_v, d_model)
        self.layer_norm = nn.LayerNorm(d_model)

    def forward(self, Q, K, V):

        ## 这个多头分为这几个步骤，首先映射分头，然后计算atten_scores，然后计算atten_value;
        ##输入进来的数据形状： Q: [batch_size x len_q x d_model], K: [batch_size x len_k x d_model], V: [batch_size x len_k x d_model]
        residual, batch_size = Q, Q.size(0)
        # (B, S, D) -proj-> (B, S, D) -split-> (B, S, H, W) -trans-> (B, H, S, W)

        ##下面这个就是先映射，后分头；一定要注意的是q和k分头之后维度是一致额，所以一看这里都是dk
        q_s = self.W_Q(Q).view(batch_size, -1, self.n_heads, self.d_k).transpose(1,2)  # q_s: [batch_size x n_heads x len_q x d_k]
        k_s = self.W_K(K).view(batch_size, -1, self.n_heads, self.d_k).transpose(1,2)  # k_s: [batch_size x n_heads x len_k x d_k]
        v_s = self.W_V(V).view(batch_size, -1, self.n_heads, self.d_v).transpose(1,2)  # v_s: [batch_size x n_heads x len_k x d_v]

        ## 输入进行的attn_mask形状是 batch_size x len_q x len_k，然后经过下面这个代码得到 新的attn_mask : [batch_size x n_heads x len_q x len_k]，就是把pad信息重复了n个头上
        #attn_mask = attn_mask.unsqueeze(1).repeat(1, n_heads, 1, 1)

        ##然后我们计算 ScaledDotProductAttention 这个函数，去7.看一下
        ## 得到的结果有两个：context: [batch_size x n_heads x len_q x d_v], attn: [batch_size x n_heads x len_q x len_k]
        context, attn = ScaledDotProductAttention(self.d_k)(q_s, k_s, v_s)
        context = context.transpose(1, 2).contiguous().view(batch_size, -1, self.n_heads * self.d_v) # context: [batch_size x len_q x n_heads * d_v]
        output = self.linear(context)
        return self.layer_norm(output + residual), attn # output: [batch_size x len_q x d_model]

## 8. PoswiseFeedForwardNet
class PoswiseFeedForwardNet(nn.Module):
    def __init__(self, d_model, d_ff):
        super(PoswiseFeedForwardNet, self).__init__()
        self.conv1 = nn.Conv1d(in_channels=d_model, out_channels=d_ff, kernel_size=1)
        self.conv2 = nn.Conv1d(in_channels=d_ff, out_channels=d_model, kernel_size=1)
        self.layer_norm = nn.LayerNorm(d_model)

    def forward(self, inputs):
        residual = inputs # inputs : [batch_size, len_q, d_model]
        output = nn.ReLU()(self.conv1(inputs.transpose(1, 2)))
        output = self.conv2(output).transpose(1, 2)
        return self.layer_norm(output + residual)

## 4. get_attn_pad_mask

## len_input * len*input  代表每个单词对其余包含自己的单词的影响力
## 没用上，因为没有补0
def get_attn_pad_mask(seq_q, seq_k):
    batch_size, len_q = seq_q.size()
    batch_size, len_k = seq_k.size()
    # eq(zero) is PAD token
    pad_attn_mask = seq_k.data.eq(0).unsqueeze(1)  # batch_size x 1 x len_k, one is masking
    return pad_attn_mask.expand(batch_size, len_q, len_k)  # batch_size x len_q x len_k

## 3. PositionalEncoding 代码实现
class PositionalEncoding(nn.Module):
    def __init__(self, d_model, dropout=0.1, max_len=5000):
        super(PositionalEncoding, self).__init__()

        ## 位置编码直接对照着公式去敲代码就可以；
        self.dropout = nn.Dropout(p=dropout)
        pe = torch.zeros(max_len, d_model)
        position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)
        div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-math.log(10000.0) / d_model))
        pe[:, 0::2] = torch.sin(position * div_term)## 这里需要注意的是pe[:, 0::2]这个用法，就是从0开始到最后面，补长为2，其实代表的就是偶数位置
        pe[:, 1::2] = torch.cos(position * div_term)##这里需要注意的是pe[:, 1::2]这个用法，就是从1开始到最后面，补长为2，其实代表的就是奇数位置
        ## 上面代码获取之后得到的pe:[max_len*d_model]

        ## 下面这个代码之后，我们得到的pe形状是：[max_len*1*d_model]
        pe = pe.unsqueeze(0).transpose(0, 1)
        self.register_buffer('pe', pe)  ## 定一个缓冲区，其实简单理解为这个参数不更新就可以
    def forward(self, x):
"""
        x: [seq_len, batch_size, d_model]
"""
        x = x + self.pe[:x.size(0), :]
        return self.dropout(x)

## 5. EncoderLayer ：包含两个部分，多头注意力机制和前馈神经网络
class EncoderLayer(nn.Module):
    def __init__(self,d_model,d_k,d_v,n_heads,d_ff):
        super(EncoderLayer, self).__init__()
        self.enc_self_attn = MultiHeadAttention(d_model,d_k,d_v,n_heads)
        self.pos_ffn = PoswiseFeedForwardNet(d_model,d_ff)

    def forward(self, enc_inputs):
        ## 下面这个就是做自注意力层，输入是enc_inputs，形状是[batch_size x seq_len_q x d_model] 需要注意的是最初始的QKV矩阵是等同于这个输入的，去看一下enc_self_attn函数 6.

        enc_outputs, attn = self.enc_self_attn(enc_inputs, enc_inputs, enc_inputs) # enc_inputs to same Q,K,V
        enc_outputs = self.pos_ffn(enc_outputs) # enc_outputs: [batch_size x len_q x d_model]
        return enc_outputs, attn
## 2. Encoder 部分包含三个部分：词向量embedding，位置编码部分，注意力层及后续的前馈神经网络

class Encoder(nn.Module):
    def __init__(self, src_vocab_size,d_model,n_layers,d_k,d_v,n_heads,d_ff):
        super(Encoder, self).__init__()
        #self.src_emb = nn.Embedding(src_vocab_size, 80)  ## 这个其实就是去定义生成一个矩阵，大小是 src_vocab_size * d_model
        self.linear = nn.Linear(src_vocab_size,d_model)
        self.pos_emb = PositionalEncoding(d_model) ## 位置编码情况，这里是固定的正余弦函数，也可以使用类似词向量的nn.Embedding获得一个可以更新学习的位置编码
        self.layers = nn.ModuleList([EncoderLayer(d_model,d_k,d_v,n_heads,d_ff) for _ in range(n_layers)]) ## 使用ModuleList对多个encoder进行堆叠，因为后续的encoder并没有使用词向量和位置编码，所以抽离出来；

    def forward(self, enc_inputs):
        enc_outputs = self.linear(enc_inputs)
        ## 这里就是位置编码，把两者相加放入到了这个函数里面，从这里可以去看一下位置编码函数的实现；3.

        enc_outputs = self.pos_emb(enc_outputs.transpose(0, 1)).transpose(0, 1)

        ##get_attn_pad_mask是为了得到句子中pad的位置信息，给到模型后面，在计算自注意力和交互注意力的时候去掉pad符号的影响，去看一下这个函数 4.

        #enc_self_attn_mask = get_attn_pad_mask(enc_inputs, enc_inputs)
        enc_self_attns = []
        for layer in self.layers:
            ## 去看EncoderLayer 层函数 5.

            enc_outputs, enc_self_attn = layer(enc_outputs)
            enc_self_attns.append(enc_self_attn)
        return enc_outputs, enc_self_attns

## 12. ScaledDotProductAttentionMask
class ScaledDotProductAttentionMask(nn.Module):
    def __init__(self, d_k):
        super(ScaledDotProductAttentionMask, self).__init__()
        self.d_k = d_k
    def forward(self, Q, K, V, attn_mask):
        ## 输入进来的维度分别是 [batch_size x n_heads x len_q x d_k]  K： [batch_size x n_heads x len_k x d_k]  V: [batch_size x n_heads x len_k x d_v]
        ##首先经过matmul函数得到的scores形状是 : [batch_size x n_heads x len_q x len_k]
        scores = torch.matmul(Q, K.transpose(-1, -2)) / np.sqrt(self.d_k)

        ## 然后关键词地方来了，下面这个就是用到了我们之前重点讲的attn_mask，把被mask的地方置为无限小，softmax之后基本就是0，对q的单词不起作用
        scores.masked_fill_(attn_mask, -1e9) # Fills elements of self tensor with value where mask is one.

        attn = nn.Softmax(dim=-1)(scores)
        context = torch.matmul(attn, V)
        return context, attn

## 11 mask-atten
class MultiHeadAttentionMask(nn.Module):
    def __init__(self, d_model, d_k, d_v, n_heads):
        super(MultiHeadAttentionMask, self).__init__()
        self.d_model = d_model
        self.d_k = d_k
        self.d_v = d_v
        self.n_heads = n_heads
        ## 输入进来的QKV是相等的，我们会使用映射linear做一个映射得到参数矩阵Wq, Wk,Wv
        self.W_Q = nn.Linear(d_model, d_k * n_heads)
        self.W_K = nn.Linear(d_model, d_k * n_heads)
        self.W_V = nn.Linear(d_model, d_v * n_heads)
        self.linear = nn.Linear(n_heads * d_v, d_model)
        self.layer_norm = nn.LayerNorm(d_model)
    def forward(self, Q, K, V, attn_mask):

        ## 这个多头分为这几个步骤，首先映射分头，然后计算atten_scores，然后计算atten_value;
        ##输入进来的数据形状： Q: [batch_size x len_q x d_model], K: [batch_size x len_k x d_model], V: [batch_size x len_k x d_model]
        residual, batch_size = Q, Q.size(0)

        ##下面这个就是先映射，后分头；这里都是dk
        q_s = self.W_Q(Q).view(batch_size, -1, self.n_heads, self.d_k).transpose(1,2)  # q_s: [batch_size x n_heads x len_q x d_k]
        k_s = self.W_K(K).view(batch_size, -1, self.n_heads, self.d_k).transpose(1,2)  # k_s: [batch_size x n_heads x len_k x d_k]
        v_s = self.W_V(V).view(batch_size, -1, self.n_heads, self.d_v).transpose(1,2)  # v_s: [batch_size x n_heads x len_k x d_v]

        ## 输入进行的attn_mask形状是 batch_size x len_q x len_k，然后经过下面这个代码得到 新的attn_mask : [batch_size x n_heads x len_q x len_k]，就是把pad信息重复了n个头上
        attn_mask = attn_mask.unsqueeze(1).repeat(1, self.n_heads, 1, 1)

        ## context: [batch_size x n_heads x len_q x d_v], attn: [batch_size x n_heads x len_q x len_k]
        context, attn = ScaledDotProductAttentionMask(self.d_k)(q_s, k_s, v_s, attn_mask)
        context = context.transpose(1, 2).contiguous().view(batch_size, -1, self.n_heads * self.d_v) # context: [batch_size x len_q x n_heads * d_v]
        output = self.linear(context)
        return self.layer_norm(output + residual), attn # output: [batch_size x len_q x d_model]

## 10.

class DecoderLayer(nn.Module):
    def __init__(self,d_model,d_k,d_v,n_heads,d_ff):
        super(DecoderLayer, self).__init__()
        self.dec_self_attn = MultiHeadAttentionMask(d_model,d_k,d_v,n_heads)
        self.dec_enc_attn = MultiHeadAttention(d_model,d_k,d_v,n_heads)
        self.pos_ffn = PoswiseFeedForwardNet(d_model,d_ff)

    def forward(self, dec_inputs, enc_outputs, mask):
        dec_outputs, dec_self_attn = self.dec_self_attn(dec_inputs, dec_inputs, dec_inputs, mask)
        dec_outputs, dec_enc_attn = self.dec_enc_attn(dec_outputs, enc_outputs, enc_outputs)
        dec_outputs = self.pos_ffn(dec_outputs)
        return dec_outputs, dec_self_attn, dec_enc_attn

## 9. Decoder

class Decoder(nn.Module):
    def __init__(self, tgt_vocab_size, d_model, n_layers,d_k,d_v,n_heads,d_ff):
        super(Decoder, self).__init__()
        self.tgt_emb = nn.Embedding(tgt_vocab_size, d_model)
        self.pos_emb = PositionalEncoding(d_model,0.1,5000)
        self.layers = nn.ModuleList([DecoderLayer(d_model,d_k,d_v,n_heads,d_ff) for _ in range(n_layers)])

    def forward(self, dec_inputs, enc_inputs, enc_outputs): # dec_inputs : [batch_size x target_len]
        dec_outputs = self.tgt_emb(dec_inputs)  # [batch_size, tgt_len, d_model]
        dec_outputs = self.pos_emb(dec_outputs.transpose(0, 1)).transpose(0, 1) # [batch_size, tgt_len, d_model]

        dec_self_attn_subsequent_mask = get_attn_subsequent_mask(dec_inputs)

        dec_self_attns, dec_enc_attns = [], []
        for layer in self.layers:
            dec_outputs, dec_self_attn, dec_enc_attn = layer(dec_outputs, enc_outputs,dec_self_attn_subsequent_mask)
            dec_self_attns.append(dec_self_attn)
            dec_enc_attns.append(dec_enc_attn)
        return dec_outputs, dec_self_attns, dec_enc_attns

## 1. 从整体网路结构来看，分为三个部分：编码层，解码层，输出层
class Transformer(nn.Module):
    def __init__(self, src_vocab_size,tgt_vocab_size, d_model, n_layers,d_k,d_v,n_heads,d_ff):
        super(Transformer, self).__init__()
        self.d_model = d_model
        self.src_vocab_size = src_vocab_size
        self.tgt_vocab_size = tgt_vocab_size
        self.encoder = Encoder(src_vocab_size,d_model,n_layers,d_k,d_v,n_heads,d_ff)  ## 编码层
        self.decoder = Decoder(tgt_vocab_size, d_model, n_layers,d_k,d_v,n_heads,d_ff)  ## 解码层
        self.crition = LOSS(tgt_vocab_size)
        self.projection = nn.Linear(d_model, tgt_vocab_size, bias=False) ## 输出层 d_model 是我们解码层每个token输出的维度大小，之后会做一个 tgt_vocab_size 大小的softmax
    def forward(self, enc_inputs, dec_inputs):
        ## 这里有两个数据进行输入，一个是enc_inputs 形状为[batch_size, src_len]，主要是作为编码段的输入，一个dec_inputs，形状为[batch_size, tgt_len]，主要是作为解码端的输入

        ## enc_inputs作为输入 形状为[batch_size, src_len]，输出由自己的函数内部指定，想要什么指定输出什么，可以是全部tokens的输出，可以是特定每一层的输出；也可以是中间某些参数的输出；
        ## enc_outputs就是主要的输出，enc_self_attns这里没记错的是QK转置相乘之后softmax之后的矩阵值，代表的是每个单词和其他单词相关性；
        enc_outputs, enc_self_attns = self.encoder(enc_inputs)

        ## dec_outputs 是decoder主要输出，用于后续的linear映射； dec_self_attns类比于enc_self_attns 是查看每个单词对decoder中输入的其余单词的相关性；dec_enc_attns是decoder中每个单词对encoder中每个单词的相关性；
        dec_outputs, dec_self_attns, dec_enc_attns = self.decoder(dec_inputs, enc_inputs, enc_outputs)

        ## dec_outputs做映射到词表大小
        dec_logits = self.projection(dec_outputs) # dec_logits : [batch_size x src_vocab_size x tgt_vocab_size]
        dec_logits = dec_logits.view(-1,dec_logits.size(-1))
        loss1 = self.crition(dec_logits,dec_inputs)
        return loss1

13、KILDIV
class LOSS(nn.Module):
    def __init__(self, tgt_vocab_size):
        super(LOSS, self).__init__()
        self.criterion = nn.KLDivLoss()
        self.tgt_vocab_size = tgt_vocab_size
    def forward(self,outputs, target):
        batch_size = outputs.size(0)
        outputs = outputs.view(-1,self.tgt_vocab_size)
        target = target.view(-1)

        true_dist = torch.zeros_like(outputs)
        true_dist.fill_(0.01 / (self.tgt_vocab_size - 1))
        true_dist.scatter_(1, target.unsqueeze(1), 1 - 0.01)
        outputs = torch.softmax(outputs, dim=0)
        return self.criterion(outputs.log(),true_dist)

if __name__ == '__main__':
    #src_vocab_size = 80 # fbank
    #tgt_vocab_size = 5000 # bpe
    src_vocab_size=40
    tgt_vocab_size=1000
    d_model=512
    n_layers=6
    d_k=64
    d_v=64
    n_heads=8
    d_ff=2048
    model = Transformer(src_vocab_size,tgt_vocab_size, d_model, n_layers,d_k,d_v,n_heads,d_ff)
    optimizer = optim.Adam(model.parameters(), lr=0.001)

    enc_inputs = torch.ones((1,3,40),dtype=torch.float32)
    dec_inputs = torch.ones((1,50),dtype=torch.int64)
    for epoch in range(20):
        optimizer.zero_grad()
        loss = model(enc_inputs, dec_inputs)
        print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.6f}'.format(loss))
        loss.backward()
        optimizer.step()
        logging.info('loss'.format(loss))

Original: https://blog.csdn.net/qq_37258753/article/details/126329654
Author: 方付平
Title: Transformer代码：适用于语音识别（streaming-mode）