提示:文章写完后,目录可以自动生成,如何生成可参考右边的帮助文档
文章目录
前言
参考资料
提示:以下是本篇文章正文内容,下面案例可供参考
1.Transformer简介
Transformer 和 LSTM 的最大区别,就是 LSTM 的训练是迭代的、串行的,必须要等当前字处理完,才可以处理下一个字。而 Transformer 的训练时并行的,即所有字是同时训练的,这样就大大增加了计算效率。
Transformer是一个encoder-decoder结构,编码器和解码器都是n层的,整个过程是一个线性的结构,一个encoder的输出作为下一个encoder的输入;当到了最后一个encoder的时候,再把最后一个encoder的输出传给decoder。
Transformer的结构与seq2seq相似,但是它没有用到任何循环神经网络。
2.Encoder
Positional Encoding
因为transformer没有用到任何循环神经网络,而RNN是一种天然的时间序列(一个字接一个字的顺序进行),如果直接编码之后便进行输入,则没有获取到句子当中各个词的位置关系,因此我们引入位置编码。
输入的X在embedding层进行维度变换,由[batch_size,sequence length]变换为[batch_size,sequence length,embedding dimension]词向量可以由训练得到,或者是调用预训练的词向量都是可以的,而位置编码,在transformer模型中,是不训练的【在Bert中是可以训练的】是由一个公式来定义好的。
需要注意的是,位置编码的维度一定是和词向量的维度是相同的,否则无法进行向量相加的操作。
随着维度i的增大,变化越来越平缓。
位置编码的详细讲解:Transformer 中的 Positional Encoding
Self-Attention
通过embedding和position encoding之后得到x1,x2…xm(向量维度[batch_size,sequence length,embedding dimension])对于每个x,通过与三个矩阵Wk, Wq, Wv相乘,衍生出三个向量。
对于所有的字来说,乘的三个矩阵是相同的,即权重是共享的。相乘的操作相当于是做了一个线性变换,即nn.linear
- 首先用第一个字的查询向量q1分别与所有的字的键向量k1…km相乘,通过softmax操作,得到第一个字与所有字的相关程度。
- 同样的操作得到a2,a3…
- 将得到的相关程度矩阵a与v1,v2…相乘相加得到context vector
- c1即为x1通过self-attention后的输出。
这种方法需要一个循环遍历所有的字xi,因此我们可以把上面的向量计算转化为矩阵计算,从而一次计算出所有时刻的输出。
Multi-Head Attention
定义多组Wq,Wk,Wv,得到多组Q,K,V来分别关注不同的上下文。
对于输入矩阵X,每一组QKV都可以得到一个输出矩阵Z。
残差连接和 Layer Normalization
残差连接
X在通过多头注意力机制后不是直接传入到全连接层,而是再次与原来的输入X相加后再传入全连接层。
这样做的好处:解决了退化问题,阻止隐模型深度太深而造成的数据特征消失的问题。
2.Decoder
decoder的输入与encoder类似,通过两个编码的操作之后,进入一个masked multi-head attention,之后经过add&norm,将输出作为Q,而encoder经过多层计算得到的输出作为K,V(K=V)以新的KQV做multi-head attention,最后经过两层的神经网络,add&norm,再经过linear改变维度,经过softmax得到最终的输出。
注意,在神经网络之后得到的输出(而不是linear后的输出)会作为下一次的输入。
Masked Multi-head Attention
代码
# %%
# 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从0开始,每隔两个,即0,2,4,6.。。双数位
sinusoid_table[:, 1::2] = np.cos(sinusoid_table[:, 1::2]) # dim 2i+1从1开始,每隔两个,即1,3,5.。。单数位
return torch.FloatTensor(sinusoid_table)
def get_attn_pad_mask(seq_q, seq_k):#这个函数主要是将input_ids中补0的位置掩盖掉,让他们不参与后面的运算,加快运算速度。
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)]# attn_shape: [batch_size, tgt_len, tgt_len]
subsequent_mask = np.triu(np.ones(attn_shape), k=1)#生成一个上三角矩阵
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):
## 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]
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)]
#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)#softmax函数按横行来做
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)
#当view()参数不止一个,且其中一个为-1时,则-1的作用相当于占位符,其大小由torch根据矩阵元素个数和其他维大小来自动计算
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]
## 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) # 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)
## 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):
super(EncoderLayer, self).__init__()
self.enc_self_attn = MultiHeadAttention()
self.pos_ffn = PoswiseFeedForwardNet()#前馈神经网络就是全连接层组成的
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此处enc_inputs传了三次,分别与wq,wk,wv做运算的带
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)#因为位置编码是公式计算的,不参与反向传播,因此在这里是“freeze”
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()#编码层
self.decoder = Decoder()#解码层
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.SGD(model.parameters(), lr=0.001)#源代码使用的是Adam,运行之后感觉效果不好,故改用SGD
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)
ich mochte ein bier P -> ['i', 'want', 'a', 'beer', 'E']#翻译结果
first head of last state enc_self_attns:
first head of last state dec_self_attns:
first head of last state dec_enc_attns: