Transformer代码学习

三类应用：

1. 机器翻译类应用 - Encoder和Decoder共同使用
2. 只使用Encoder端-文本分类BERT和图片分类VIT
3. 只使用Decoder端-生成类模型

源码总览

# code by Tae Hwan Jung(Jeff Jung) @graykode, Derek Miller @dmmiller612
# Reference : https://github.com/jadore801120/attention-is-all-you-need-pytorch
#           https://github.com/JayParks/transformer
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
import matplotlib.pyplot as plt

# S: Symbol that shows starting of decoding input
# E: Symbol that shows starting of decoding output
# P: Symbol that will fill in blank sequence if current batch data size is short than time steps

def make_batch(sentences):
    input_batch = [[src_vocab[n] for n in sentences[0].split()]]
    output_batch = [[tgt_vocab[n] for n in sentences[1].split()]]
    target_batch = [[tgt_vocab[n] for n in sentences[2].split()]]
    return torch.LongTensor(input_batch), torch.LongTensor(output_batch), torch.LongTensor(target_batch)

def get_sinusoid_encoding_table(n_position, d_model):
    def cal_angle(position, hid_idx):
        return position / np.power(10000, 2 * (hid_idx // 2) / d_model)
    def get_posi_angle_vec(position):
        return [cal_angle(position, hid_j) for hid_j in range(d_model)]

    sinusoid_table = np.array([get_posi_angle_vec(pos_i) for pos_i in range(n_position)])
    sinusoid_table[:, 0::2] = np.sin(sinusoid_table[:, 0::2])  # dim 2i
    sinusoid_table[:, 1::2] = np.cos(sinusoid_table[:, 1::2])  # dim 2i+1
    return torch.FloatTensor(sinusoid_table)

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(=len_q), one is masking
    return pad_attn_mask.expand(batch_size, len_q, len_k)  # batch_size x len_q x len_k

def get_attn_subsequent_mask(seq):
    attn_shape = [seq.size(0), seq.size(1), seq.size(1)]
    subsequent_mask = np.triu(np.ones(attn_shape), k=1)
    subsequent_mask = torch.from_numpy(subsequent_mask).byte()
    return subsequent_mask

class ScaledDotProductAttention(nn.Module):
    def __init__(self):
        super(ScaledDotProductAttention, self).__init__()

    def forward(self, Q, K, V, attn_mask):
        scores = torch.matmul(Q, K.transpose(-1, -2)) / np.sqrt(d_k) # scores : [batch_size x n_heads x len_q(=len_k) x len_k(=len_q)]
        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

class MultiHeadAttention(nn.Module):
    def __init__(self):
        super(MultiHeadAttention, self).__init__()
        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):
        # 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)
        q_s = self.W_Q(Q).view(batch_size, -1, n_heads, 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, n_heads, 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, n_heads, d_v).transpose(1,2)  # v_s: [batch_size x n_heads x len_k x d_v]

        attn_mask = attn_mask.unsqueeze(1).repeat(1, n_heads, 1, 1) # attn_mask : [batch_size x n_heads x len_q x len_k]

        # context: [batch_size x n_heads x len_q x d_v], attn: [batch_size x n_heads x len_q(=len_k) x len_k(=len_q)]
        context, attn = ScaledDotProductAttention()(q_s, k_s, v_s, attn_mask)
        context = context.transpose(1, 2).contiguous().view(batch_size, -1, n_heads * 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]

class PoswiseFeedForwardNet(nn.Module):
    def __init__(self):
        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)

class EncoderLayer(nn.Module):
    def __init__(self):
        super(EncoderLayer, self).__init__()
        self.enc_self_attn = MultiHeadAttention()
        self.pos_ffn = PoswiseFeedForwardNet()

    def forward(self, enc_inputs, enc_self_attn_mask):
        enc_outputs, attn = self.enc_self_attn(enc_inputs, enc_inputs, enc_inputs, enc_self_attn_mask) # 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

class DecoderLayer(nn.Module):
    def __init__(self):
        super(DecoderLayer, self).__init__()
        self.dec_self_attn = MultiHeadAttention()
        self.dec_enc_attn = MultiHeadAttention()
        self.pos_ffn = PoswiseFeedForwardNet()

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

class Encoder(nn.Module):
    def __init__(self):
        super(Encoder, self).__init__()
        self.src_emb = nn.Embedding(src_vocab_size, d_model)
        self.pos_emb = nn.Embedding.from_pretrained(get_sinusoid_encoding_table(src_len+1, d_model),freeze=True)
        self.layers = nn.ModuleList([EncoderLayer() for _ in range(n_layers)])

    def forward(self, enc_inputs): # enc_inputs : [batch_size x source_len]
        enc_outputs = self.src_emb(enc_inputs) + self.pos_emb(torch.LongTensor([[1,2,3,4,0]]))
        enc_self_attn_mask = get_attn_pad_mask(enc_inputs, enc_inputs)
        enc_self_attns = []
        for layer in self.layers:
            enc_outputs, enc_self_attn = layer(enc_outputs, enc_self_attn_mask)
            enc_self_attns.append(enc_self_attn)
        return enc_outputs, enc_self_attns

class Decoder(nn.Module):
    def __init__(self):
        super(Decoder, self).__init__()
        self.tgt_emb = nn.Embedding(tgt_vocab_size, d_model)
        self.pos_emb = nn.Embedding.from_pretrained(get_sinusoid_encoding_table(tgt_len+1, d_model),freeze=True)
        self.layers = nn.ModuleList([DecoderLayer() 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) + self.pos_emb(torch.LongTensor([[5,1,2,3,4]]))
        dec_self_attn_pad_mask = get_attn_pad_mask(dec_inputs, dec_inputs)
        dec_self_attn_subsequent_mask = get_attn_subsequent_mask(dec_inputs)
        dec_self_attn_mask = torch.gt((dec_self_attn_pad_mask + dec_self_attn_subsequent_mask), 0)

        dec_enc_attn_mask = get_attn_pad_mask(dec_inputs, enc_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_mask, dec_enc_attn_mask)
            dec_self_attns.append(dec_self_attn)
            dec_enc_attns.append(dec_enc_attn)
        return dec_outputs, dec_self_attns, dec_enc_attns

class Transformer(nn.Module):
    def __init__(self):
        super(Transformer, self).__init__()
        self.encoder = Encoder()
        self.decoder = Decoder()
        self.projection = nn.Linear(d_model, tgt_vocab_size, bias=False)
    def forward(self, enc_inputs, dec_inputs):
        enc_outputs, enc_self_attns = self.encoder(enc_inputs)
        dec_outputs, dec_self_attns, dec_enc_attns = self.decoder(dec_inputs, enc_inputs, enc_outputs)
        dec_logits = self.projection(dec_outputs) # dec_logits : [batch_size x src_vocab_size x tgt_vocab_size]
        return dec_logits.view(-1, dec_logits.size(-1)), enc_self_attns, dec_self_attns, dec_enc_attns

def showgraph(attn):
    attn = attn[-1].squeeze(0)[0]
    attn = attn.squeeze(0).data.numpy()
    fig = plt.figure(figsize=(n_heads, n_heads)) # [n_heads, n_heads]
    ax = fig.add_subplot(1, 1, 1)
    ax.matshow(attn, cmap='viridis')
    ax.set_xticklabels(['']+sentences[0].split(), fontdict={'fontsize': 14}, rotation=90)
    ax.set_yticklabels(['']+sentences[2].split(), fontdict={'fontsize': 14})
    plt.show()

if __name__ == '__main__':
    sentences = ['ich mochte ein bier P', 'S i want a beer', 'i want a beer E']

    # Transformer Parameters
    # Padding Should be Zero
    src_vocab = {'P': 0, 'ich': 1, 'mochte': 2, 'ein': 3, 'bier': 4}
    src_vocab_size = len(src_vocab)

    tgt_vocab = {'P': 0, 'i': 1, 'want': 2, 'a': 3, 'beer': 4, 'S': 5, 'E': 6}
    number_dict = {i: w for i, w in enumerate(tgt_vocab)}
    tgt_vocab_size = len(tgt_vocab)

    src_len = 5 # length of source
    tgt_len = 5 # length of target

    d_model = 512  # Embedding Size
    d_ff = 2048  # FeedForward dimension
    d_k = d_v = 64  # dimension of K(=Q), V
    n_layers = 6  # number of Encoder of Decoder Layer
    n_heads = 8  # number of heads in Multi-Head Attention

    model = Transformer()

    criterion = nn.CrossEntropyLoss()
    optimizer = optim.Adam(model.parameters(), lr=0.001)

    enc_inputs, dec_inputs, target_batch = make_batch(sentences)

    for epoch in range(20):
        optimizer.zero_grad()
        outputs, enc_self_attns, dec_self_attns, dec_enc_attns = model(enc_inputs, dec_inputs)
        loss = criterion(outputs, target_batch.contiguous().view(-1))
        print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.6f}'.format(loss))
        loss.backward()
        optimizer.step()

    # Test
    predict, _, _, _ = model(enc_inputs, dec_inputs)
    predict = predict.data.max(1, keepdim=True)[1]
    print(sentences[0], '->', [number_dict[n.item()] for n in predict.squeeze()])

    print('first head of last state enc_self_attns')
    showgraph(enc_self_attns)

    print('first head of last state dec_self_attns')
    showgraph(dec_self_attns)

    print('first head of last state dec_enc_attns')
    showgraph(dec_enc_attns)

main部分

if __name__ == '__main__':
	## 句子输入部分，这三个句子是一组句子，属于一个样本（机器翻译例子），batch_size = 1
    sentences = ['ich mochte ein bier P', 'S i want a beer', 'i want a beer E']

编码端输入：ich mochte ein bier P
解码端输入：S i want a beer
与解码段输出对比计算损失的真实标签（分类问题计算损失）：i want a beer E
三个特殊字符：S——start，E——end，P——pad字符填充字符

在训练中batch size往往不是1（不止一个样本）。 eg. batch
size设置为4，有四个样本（4组句子），每组句子包含如上3个句子。 下面的图片代表每组句子中的第一个

一个batch在被模型处理时，采用矩阵化运算，若一个batch中句子长度不一致就无法组成有效矩阵
解决方法：设置一个最大长度max_len，大于maxLenth的部分截掉，小于maxLenth的部分用pad字符补齐

encoder解码，从解码端的输入到输出，再把输出拿到解码端作为下一册的输入，该过程无法并行（下一时刻的输入取决于上一时刻的输出）；
为加快收敛和训练速度，采用Teacher forcing方法：直接将真实标签作为一种输入，使用mask把后面的单词全部mask住（不让看到当前时刻后面的单词）

    # Transformer Parameters
    # Padding Should be Zero
    ## 构建词表
    ## 编码端词表
    src_vocab = {'P': 0, 'ich': 1, 'mochte': 2, 'ein': 3, 'bier': 4}
    src_vocab_size = len(src_vocab)
	
	## 解码端词表
    tgt_vocab = {'P': 0, 'i': 1, 'want': 2, 'a': 3, 'beer': 4, 'S': 5, 'E': 6}
    number_dict = {i: w for i, w in enumerate(tgt_vocab)}
    tgt_vocab_size = len(tgt_vocab)
	
    src_len = 5 # length of source 编码端输入长度
    tgt_len = 5 # length of target 解码端输入长度

	## 模型参数
    d_model = 512  # Embedding Size 每个字符转换为embedding时的大小
    d_ff = 2048  # FeedForward dimension 前馈神经网络中linear层 映射的维度
    d_k = d_v = 64  # dimension of K(=Q), V
    n_layers = 6  # number of Encoder of Decoder Layer encoder和decoder堆叠的数量
    n_heads = 8  # number of heads in Multi-Head Attention 多头注意力机制有几个头
	## 模型部分 最关键部分
    model = Transformer()

    criterion = nn.CrossEntropyLoss()
    optimizer = optim.Adam(model.parameters(), lr=0.001)

    enc_inputs, dec_inputs, target_batch = make_batch(sentences)

    for epoch in range(20):
        optimizer.zero_grad()
        outputs, enc_self_attns, dec_self_attns, dec_enc_attns = model(enc_inputs, dec_inputs)
        loss = criterion(outputs, target_batch.contiguous().view(-1))
        print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.6f}'.format(loss))
        loss.backward()
        optimizer.step()

    # Test
    predict, _, _, _ = model(enc_inputs, dec_inputs)
    predict = predict.data.max(1, keepdim=True)[1]
    print(sentences[0], '->', [number_dict[n.item()] for n in predict.squeeze()])

    print('first head of last state enc_self_attns')
    showgraph(enc_self_attns)

    print('first head of last state dec_self_attns')
    showgraph(dec_self_attns)

    print('first head of last state dec_enc_attns')
    showgraph(dec_enc_attns)

transformer整体结构

从整体网络结构来看，分为三个部分：编码层，解码层，输出层

class Transformer(nn.Module):
	#初始化函数：将这三部分列出来
    def __init__(self):
        super(Transformer, self).__init__()
        self.encoder = Encoder() ## 编码层
        self.decoder = Decoder() ## 解码层
        self.projection = nn.Linear(d_model, tgt_vocab_size, bias=False) ## 输出层 d_model 是解码层每个token输出的维度大小，之后会做一个tat_vocab_size大小的softmax
    def forward(self, enc_inputs, dec_inputs):
    ## 两个数据输入， enc_inputs形状为[batch_size, src_len]，作为编码端输入，一个dec_input，形状为[batch_size, tgt_len]，作为解码端输入
		
		##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_output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]
        return dec_logits.view(-1, dec_logits.size(-1)), enc_self_attns, dec_self_attns, dec_enc_attns

Encoder结构

分为三个部分：词向量embedding，位置编码部分，注意力层及后续前馈神经网络

class Encoder(nn.Module):
    def __init__(self):
        super(Encoder, self).__init__()
        self.src_emb = nn.Embedding(src_vocab_size, d_model) ##词向量层 生成一个矩阵 src_vocab_size * d_model
        self.pos_emb = nn.Embedding.from_pretrained(get_sinusoid_encoding_table(src_len+1, d_model),freeze=True) ##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() 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_inputs : [batch_size x source_len]
    ## 对于encoder接受一个输入，即enc_input 形状是
    
    	## 下面这个代码通过src_emb，进行索引定位，enc_outputs输出形状是[batch_size, src_len, d_model]
    	## 位置编码，把两者相加放入到了这个函数里面
        enc_outputs = self.src_emb(enc_inputs) + self.pos_emb(torch.LongTensor([[1,2,3,4,0]]))
        
        ##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
        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
            enc_outputs, enc_self_attn = layer(enc_outputs, enc_self_attn_mask)
            enc_self_attns.append(enc_self_attn)
        return enc_outputs, enc_self_attns