Pytorch Transformer

Pytorch Transformer

0. 环境介绍

环境使用 Kaggle 里免费建立的 Notebook

教程使用李沐老师的 动手学深度学习 网站和 视频讲解

小技巧：当遇到函数看不懂的时候可以按 Shift+Tab 查看函数详解。

1. Transformer

1.1 架构

• 基于Encoder-Decoder 架构来处理序列对
• 跟使用注意力的 seq2seq 不同，Transformer 是纯基于注意力

1.2 多头注意力

• 对同一 Key，Value，Query，希望抽取不同的信息
  • 例如短距离关系和长距离关系
• 多头注意力使用 $h$ 个独立的注意力 Pooling
  • 合并各个头（head）输出得到最终输出

数学语言表达：

• 给定 Query $\mathbf{q}\in {\mathbb{R}}^{d_q}$，Key $\mathbf{k}\in {\mathbb{R}}^{d_k}$，Value $\mathbf{v}\in {\mathbb{R}}^{d_v}$
• 头 $i$ 的可学习参数：${\mathbf{W}}_i^{(q)}\in {\mathbb{R}}^{p_q\times d_q}$，${\mathbf{W}}_i^{(k)}\in {\mathbb{R}}^{p_k\times d_k}$，${\mathbf{W}}_i^{(v)}\in {\mathbb{R}}^{p_v\times d_v}$
• 头 $i$ 的输出：${\mathbf{h}}_i=f({\mathbf{W}}_i^{(q)}\mathbf{q},{\mathbf{W}}_i^{(k)}\mathbf{k},{\mathbf{W}}_i^{(v)}\mathbf{v})\in {\mathbb{R}}^{p_v}$
• 输出的可学习参数 ：${\mathbf{W}}_o\in {\mathbb{R}}^{p_o\times h{p}_v}$
• 多头注意力输出：
  $${\mathbf{W}}_o\left[\begin{array}{c}{\mathbf{h}}_1\\ ⋮\\ {\mathbf{h}}_h\end{array}\right]\in {\mathbb{R}}^{p_o}$$

1.3 有掩码的多头注意力

• Decoder 对序列重一个元素输出时，不应该考虑该元素之后的元素（因为从时间序列上还没看到）
• 可以通过掩码来实现
  • 也就是计算 ${\mathbf{x}}_i$ 输出时，假装当前的序列长度为 $i$

1.4 基于位置的前馈网络

$b$：表示一个 batch 有多少句子，$n$：表示句子有多少个单词，$d$ 表示每个单词向量的维度

• 将输入形状由 $(b,n,d)$ 变换成 $(bn,d)$
• 作用两个全连接层
• 输出形状由 $(bn,d)$ 变化回 $(b,n,d)$
• 等价于两层核窗口为 $1$ 的一维卷积层

1.5 层归一化（LayerNorm）

• 批量归一化（BatchNorm）对每个特征/通道里元素进行归一化
  • 不适合序列长度会变的 NLP 应用（每个句子的长度不一样，不能用 BN）
• 层归一化（LayerNorm）对每个样本里的元素进行归一化

1.6 信息传递

• Encoder 中的输出 ${\mathbf{y}}_1,...,{\mathbf{y}}_n$
• 将其作为第 $i$ 个 Transformer 块中多头注意力的 Key 和 Value
  • 它的 Query 来自目标序列
• 意味着 Encoder 和 Decoder 中块的个数和输出维度都是一样的

1.7 预测

• 预测第 $t+1$ 个输出时
• Decoder 中输入前 $t$ 个预测值
  • 在自注意力中，前 $t$ 个预测值作为 Key 和 Value，第 $t$ 个预测值还作为 Query

训练时：Decoder 第一个带掩码的多头注意力的 K，V，来自本身的 Q，第二个多头注意力的 K，V 来自 Encoder
预测时：K，V 来自 Decoder 的上一时刻的输出作为 K，V

2. 代码

2.0 导包

!pip install -U d2l
import math
import pandas as pd
import torch
from torch import nn
from d2l import torch as d2l

2.1 多头注意力

class MultiHeadAttention(nn.Module):
    """多头注意力"""
    def __init__(self, key_size, query_size, value_size, num_hiddens,
                 num_heads, dropout, bias=False, **kwargs):
        super(MultiHeadAttention, self).__init__(**kwargs)
        self.num_heads = num_heads
        self.attention = d2l.DotProductAttention(dropout)
        self.W_q = nn.Linear(query_size, num_hiddens, bias=bias)
        self.W_k = nn.Linear(key_size, num_hiddens, bias=bias)
        self.W_v = nn.Linear(value_size, num_hiddens, bias=bias)
        self.W_o = nn.Linear(num_hiddens, num_hiddens, bias=bias)

    def forward(self, queries, keys, values, valid_lens):
        # queries，keys，values的形状:
        # (batch_size，查询或者“键－值”对的个数，num_hiddens)
        # valid_lens　的形状:
        # (batch_size，)或(batch_size，查询的个数)
        # 经过变换后，输出的queries，keys，values　的形状:
        # (batch_size*num_heads，查询或者“键－值”对的个数，
        # num_hiddens/num_heads)
        queries = transpose_qkv(self.W_q(queries), self.num_heads)
        keys = transpose_qkv(self.W_k(keys), self.num_heads)
        values = transpose_qkv(self.W_v(values), self.num_heads)

        if valid_lens is not None:
            # 在轴0，将第一项（标量或者矢量）复制num_heads次，
            # 然后如此复制第二项，然后诸如此类。
            valid_lens = torch.repeat_interleave(
                valid_lens, repeats=self.num_heads, dim=0)

        # output的形状:(batch_size*num_heads，查询的个数，
        # num_hiddens/num_heads)
        output = self.attention(queries, keys, values, valid_lens)

        # output_concat的形状:(batch_size，查询的个数，num_hiddens)
        output_concat = transpose_output(output, self.num_heads)
        return self.W_o(output_concat)

使多个头并行（为了省去循环操作，所以使用大量的维度变换，节省效率）：

def transpose_qkv(X, num_heads):
    """为了多注意力头的并行计算而变换形状"""
    # 输入X的形状:(batch_size，查询或者“键－值”对的个数，num_hiddens)
    # 输出X的形状:(batch_size，查询或者“键－值”对的个数，num_heads，
    # num_hiddens/num_heads)
    X = X.reshape(X.shape[0], X.shape[1], num_heads, -1)

    # 输出X的形状:(batch_size，num_heads，查询或者“键－值”对的个数,
    # num_hiddens/num_heads)
    X = X.permute(0, 2, 1, 3)

    # 最终输出的形状:(batch_size*num_heads,查询或者“键－值”对的个数,
    # num_hiddens/num_heads)
    return X.reshape(-1, X.shape[2], X.shape[3])


def transpose_output(X, num_heads):
    """逆转transpose_qkv函数的操作"""
    X = X.reshape(-1, num_heads, X.shape[1], X.shape[2])
    X = X.permute(0, 2, 1, 3)
    return X.reshape(X.shape[0], X.shape[1], -1)

测试：

num_hiddens, num_heads = 100, 5
attention = MultiHeadAttention(num_hiddens, num_hiddens, num_hiddens,
                               num_hiddens, num_heads, 0.5)
attention.eval()

batch_size, num_queries = 2, 4
num_kvpairs, valid_lens =  6, torch.tensor([3, 2])
# X 维度(2, 4, 100)
X = torch.ones((batch_size, num_queries, num_hiddens))
# Y 维度(2, 6, 100)
Y = torch.ones((batch_size, num_kvpairs, num_hiddens))
attention(X, Y, Y, valid_lens).shape

2.2 基于位置的前馈网络

实际上就是 MLP：

class PositionWiseFFN(nn.Module):
    """基于位置的前馈网络"""
    def __init__(self, ffn_num_input, ffn_num_hiddens, ffn_num_outputs,
                 **kwargs):
        super(PositionWiseFFN, self).__init__(**kwargs)
        self.dense1 = nn.Linear(ffn_num_input, ffn_num_hiddens)
        self.relu = nn.ReLU()
        self.dense2 = nn.Linear(ffn_num_hiddens, ffn_num_outputs)

    def forward(self, X):
        return self.dense2(self.relu(self.dense1(X)))

测试：

ffn = PositionWiseFFN(4, 4, 8)
ffn.eval()
ffn(torch.ones((2, 3, 4))).shape

2.3 层归一化（LayerNorm）

对比 BN 和 LN：

ln = nn.LayerNorm(3)
bn = nn.BatchNorm1d(3)
X = torch.tensor([[1, 3, 5], 
                  [1, 2, 3]], dtype=torch.float32)
# 在训练模式下计算X的均值和方差
print('layer norm:', ln(X), '\nbatch norm:', bn(X))

2.4 加入残差连接

class AddNorm(nn.Module):
    """残差连接后进行层规范化"""
    def __init__(self, normalized_shape, dropout, **kwargs):
        super(AddNorm, self).__init__(**kwargs)
        self.dropout = nn.Dropout(dropout)
        self.ln = nn.LayerNorm(normalized_shape)

    def forward(self, X, Y):
        return self.ln(self.dropout(Y) + X)

add_norm = AddNorm([3, 4], 0.5)
add_norm.eval()
add_norm(torch.ones((2, 3, 4)), torch.ones((2, 3, 4))).shape

2.5 EncoderBlock

class EncoderBlock(nn.Module):
    """transformer编码器块"""
    def __init__(self, key_size, query_size, value_size, num_hiddens,
                 norm_shape, ffn_num_input, ffn_num_hiddens, num_heads,
                 dropout, use_bias=False, **kwargs):
        super(EncoderBlock, self).__init__(**kwargs)
        self.attention = d2l.MultiHeadAttention(
            key_size, query_size, value_size, num_hiddens, num_heads, dropout,
            use_bias)
        self.addnorm1 = AddNorm(norm_shape, dropout)
        self.ffn = PositionWiseFFN(
            ffn_num_input, ffn_num_hiddens, num_hiddens)
        self.addnorm2 = AddNorm(norm_shape, dropout)

    def forward(self, X, valid_lens):
        Y = self.addnorm1(X, self.attention(X, X, X, valid_lens))
        return self.addnorm2(Y, self.ffn(Y))

Transformer 编码器中的任何层都不会改变其输入的形状：

X = torch.ones((2, 100, 24))
valid_lens = torch.tensor([3, 2])
encoder_blk = EncoderBlock(24, 24, 24, 24, [100, 24], 24, 48, 8, 0.5)
encoder_blk.eval()
encoder_blk(X, valid_lens).shape

2.6 TransformerEncoder

class TransformerEncoder(d2l.Encoder):
    """transformer编码器"""
    def __init__(self, vocab_size, key_size, query_size, value_size,
                 num_hiddens, norm_shape, ffn_num_input, ffn_num_hiddens,
                 num_heads, num_layers, dropout, use_bias=False, **kwargs):
        super(TransformerEncoder, self).__init__(**kwargs)
        self.num_hiddens = num_hiddens
        self.embedding = nn.Embedding(vocab_size, num_hiddens)
        self.pos_encoding = d2l.PositionalEncoding(num_hiddens, dropout)
        self.blks = nn.Sequential()
        for i in range(num_layers):
            self.blks.add_module("block"+str(i),
                EncoderBlock(key_size, query_size, value_size, num_hiddens,
                             norm_shape, ffn_num_input, ffn_num_hiddens,
                             num_heads, dropout, use_bias))

    def forward(self, X, valid_lens, *args):
        # 因为位置编码值在-1和1之间，
        # 因此嵌入值乘以嵌入维度的平方根进行缩放，
        # 然后再与位置编码相加。
        X = self.pos_encoding(self.embedding(X) * math.sqrt(self.num_hiddens))
        self.attention_weights = [None] * len(self.blks)
        for i, blk in enumerate(self.blks):
            X = blk(X, valid_lens)
            self.attention_weights[i] = blk.attention.attention.attention_weights
        return X

创建一个两层的 Transformer 编码器：

encoder = TransformerEncoder(
    200, 24, 24, 24, 24, [100, 24], 24, 48, 8, 2, 0.5)
encoder.eval()
encoder(torch.ones((2, 100), dtype=torch.long), valid_lens).shape

2.7 DecoderBlock

class DecoderBlock(nn.Module):
    """解码器中第i个块"""
    def __init__(self, key_size, query_size, value_size, num_hiddens,
                 norm_shape, ffn_num_input, ffn_num_hiddens, num_heads,
                 dropout, i, **kwargs):
        super(DecoderBlock, self).__init__(**kwargs)
        self.i = i
        self.attention1 = d2l.MultiHeadAttention(
            key_size, query_size, value_size, num_hiddens, num_heads, dropout)
        self.addnorm1 = AddNorm(norm_shape, dropout)
        self.attention2 = d2l.MultiHeadAttention(
            key_size, query_size, value_size, num_hiddens, num_heads, dropout)
        self.addnorm2 = AddNorm(norm_shape, dropout)
        self.ffn = PositionWiseFFN(ffn_num_input, ffn_num_hiddens,
                                   num_hiddens)
        self.addnorm3 = AddNorm(norm_shape, dropout)

    def forward(self, X, state):
        enc_outputs, enc_valid_lens = state[0], state[1]
        # 训练阶段，输出序列的所有词元都在同一时间处理，
        # 因此state[2][self.i]初始化为None。
        # 预测阶段，输出序列是通过词元一个接着一个解码的，
        # 因此state[2][self.i]包含着直到当前时间步第i个块解码的输出表示
        if state[2][self.i] is None:
            key_values = X
        else:
            key_values = torch.cat((state[2][self.i], X), axis=1)
        state[2][self.i] = key_values
        if self.training:
            batch_size, num_steps, _ = X.shape
            # dec_valid_lens的开头:(batch_size,num_steps),
            # 其中每一行是[1,2,...,num_steps]
            dec_valid_lens = torch.arange(
                1, num_steps + 1, device=X.device).repeat(batch_size, 1)
        else:
            dec_valid_lens = None

        # 自注意力
        X2 = self.attention1(X, key_values, key_values, dec_valid_lens)
        Y = self.addnorm1(X, X2)
        # 编码器－解码器注意力。
        # enc_outputs的开头:(batch_size,num_steps,num_hiddens)
        Y2 = self.attention2(Y, enc_outputs, enc_outputs, enc_valid_lens)
        Z = self.addnorm2(Y, Y2)
        return self.addnorm3(Z, self.ffn(Z)), state

decoder_blk = DecoderBlock(24, 24, 24, 24, [100, 24], 24, 48, 8, 0.5, 0)
decoder_blk.eval()
X = torch.ones((2, 100, 24))
state = [encoder_blk(X, valid_lens), valid_lens, [None]]
decoder_blk(X, state)[0].shape

2.8 TransformerDecoder

class TransformerDecoder(d2l.AttentionDecoder):
    def __init__(self, vocab_size, key_size, query_size, value_size,
                 num_hiddens, norm_shape, ffn_num_input, ffn_num_hiddens,
                 num_heads, num_layers, dropout, **kwargs):
        super(TransformerDecoder, self).__init__(**kwargs)
        self.num_hiddens = num_hiddens
        self.num_layers = num_layers
        self.embedding = nn.Embedding(vocab_size, num_hiddens)
        self.pos_encoding = d2l.PositionalEncoding(num_hiddens, dropout)
        self.blks = nn.Sequential()
        for i in range(num_layers):
            self.blks.add_module("block"+str(i),
                DecoderBlock(key_size, query_size, value_size, num_hiddens,
                             norm_shape, ffn_num_input, ffn_num_hiddens,
                             num_heads, dropout, i))
        self.dense = nn.Linear(num_hiddens, vocab_size)

    def init_state(self, enc_outputs, enc_valid_lens, *args):
        return [enc_outputs, enc_valid_lens, [None] * self.num_layers]

    def forward(self, X, state):
        X = self.pos_encoding(self.embedding(X) * math.sqrt(self.num_hiddens))
        self._attention_weights = [[None] * len(self.blks) for _ in range (2)]
        for i, blk in enumerate(self.blks):
            X, state = blk(X, state)
            # 解码器自注意力权重
            self._attention_weights[0][i] = blk.attention1.attention.attention_weights
            # “编码器－解码器”自注意力权重
            self._attention_weights[1][i] = blk.attention2.attention.attention_weights
        return self.dense(X), state

    @property
    def attention_weights(self):
        return self._attention_weights

2.9 训练

Q，K，V 长度都是 $32$，多头注意力头数为 $4$，编码器和解码器个数都是 $4$。

num_hiddens, num_layers, dropout, batch_size, num_steps = 32, 2, 0.1, 64, 10
lr, num_epochs, device = 0.005, 200, d2l.try_gpu()
ffn_num_input, ffn_num_hiddens, num_heads = 32, 64, 4
key_size, query_size, value_size = 32, 32, 32
norm_shape = [32]

train_iter, src_vocab, tgt_vocab = d2l.load_data_nmt(batch_size, num_steps)

encoder = TransformerEncoder(
    len(src_vocab), key_size, query_size, value_size, num_hiddens,
    norm_shape, ffn_num_input, ffn_num_hiddens, num_heads,
    num_layers, dropout)
decoder = TransformerDecoder(
    len(tgt_vocab), key_size, query_size, value_size, num_hiddens,
    norm_shape, ffn_num_input, ffn_num_hiddens, num_heads,
    num_layers, dropout)
net = d2l.EncoderDecoder(encoder, decoder)
d2l.train_seq2seq(net, train_iter, lr, num_epochs, tgt_vocab, device)

2.10 预测

engs = ['go .', "i lost .", 'he\'s calm .', 'i\'m home .']
fras = ['va !', 'j\'ai perdu .', 'il est calme .', 'je suis chez moi .']
for eng, fra in zip(engs, fras):
    translation, dec_attention_weight_seq = d2l.predict_seq2seq(
        net, eng, src_vocab, tgt_vocab, num_steps, device, True)
    print(f'{eng} => {translation}, ',
          f'bleu {d2l.bleu(translation, fra, k=2):.3f}')

2.11 可视化 Transformer 的注意力权重

enc_attention_weights = torch.cat(net.encoder.attention_weights, 0).reshape((num_layers, num_heads,
    -1, num_steps))
enc_attention_weights.shape

纵坐标表示 Query，横坐标表示 Key，每个点表示 Query 看到的 Key。

d2l.show_heatmaps(
    enc_attention_weights.cpu(), xlabel='Key positions',
    ylabel='Query positions', titles=['Head %d' % i for i in range(1, 5)],
    figsize=(7, 3.5))

dec_attention_weights_2d = [head[0].tolist()
                            for step in dec_attention_weight_seq
                            for attn in step for blk in attn for head in blk]
dec_attention_weights_filled = torch.tensor(
    pd.DataFrame(dec_attention_weights_2d).fillna(0.0).values)
dec_attention_weights = dec_attention_weights_filled.reshape((-1, 2, num_layers, num_heads, num_steps))
dec_self_attention_weights, dec_inter_attention_weights = \
    dec_attention_weights.permute(1, 2, 3, 0, 4)
dec_self_attention_weights.shape, dec_inter_attention_weights.shape

解码偏后的词会去看编码偏后的词：

# Plusonetoincludethebeginning-of-sequencetoken
d2l.show_heatmaps(
    dec_self_attention_weights[:, :, :, :len(translation.split()) + 1],
    xlabel='Key positions', ylabel='Query positions',
    titles=['Head %d' % i for i in range(1, 5)], figsize=(7, 3.5))

与编码器的自注意力的情况类似，通过指定输入序列的有效长度，输出序列的查询不会与输入序列中填充位置的词元进行注意力计算：

d2l.show_heatmaps(
    dec_inter_attention_weights, xlabel='Key positions',
    ylabel='Query positions', titles=['Head %d' % i for i in range(1, 5)],
    figsize=(7, 3.5))