#

变压器编码器和解码器模型

13import math
14
15import torch
16import torch.nn as nn
17
18from labml_nn.utils import clone_module_list
19from .feed_forward import FeedForward
20from .mha import MultiHeadAttention
21from .positional_encoding import get_positional_encoding

#

嵌入令牌并添加固定位置编码

24class EmbeddingsWithPositionalEncoding(nn.Module):

#

31    def __init__(self, d_model: int, n_vocab: int, max_len: int = 5000):
32        super().__init__()
33        self.linear = nn.Embedding(n_vocab, d_model)
34        self.d_model = d_model
35        self.register_buffer('positional_encodings', get_positional_encoding(d_model, max_len))

#

37    def forward(self, x: torch.Tensor):
38        pe = self.positional_encodings[:x.shape[0]].requires_grad_(False)
39        return self.linear(x) * math.sqrt(self.d_model) + pe

#

嵌入令牌并添加参数化的位置编码

42class EmbeddingsWithLearnedPositionalEncoding(nn.Module):

#

49    def __init__(self, d_model: int, n_vocab: int, max_len: int = 5000):
50        super().__init__()
51        self.linear = nn.Embedding(n_vocab, d_model)
52        self.d_model = d_model
53        self.positional_encodings = nn.Parameter(torch.zeros(max_len, 1, d_model), requires_grad=True)

#

55    def forward(self, x: torch.Tensor):
56        pe = self.positional_encodings[:x.shape[0]]
57        return self.linear(x) * math.sqrt(self.d_model) + pe

#

Transformer Layer

This can act as an encoder layer or a decoder layer. We use pre-norm.

60class TransformerLayer(nn.Module):

#

d_model 是令牌嵌入的大小
self_attn 是自我关注模块
src_attn 是源关注模块（当它在解码器中使用时）
feed_forward 是前馈模块
dropout_prob 是自我关注和 FFN 后退学的概率

69    def __init__(self, *,
70                 d_model: int,
71                 self_attn: MultiHeadAttention,
72                 src_attn: MultiHeadAttention = None,
73                 feed_forward: FeedForward,
74                 dropout_prob: float):

#

82        super().__init__()
83        self.size = d_model
84        self.self_attn = self_attn
85        self.src_attn = src_attn
86        self.feed_forward = feed_forward
87        self.dropout = nn.Dropout(dropout_prob)
88        self.norm_self_attn = nn.LayerNorm([d_model])
89        if self.src_attn is not None:
90            self.norm_src_attn = nn.LayerNorm([d_model])
91        self.norm_ff = nn.LayerNorm([d_model])

#

是否将输入保存到前馈层

93        self.is_save_ff_input = False

#

95    def forward(self, *,
96                x: torch.Tensor,
97                mask: torch.Tensor,
98                src: torch.Tensor = None,
99                src_mask: torch.Tensor = None):

#

在进行自我注意之前对向量进行归一化

101        z = self.norm_self_attn(x)

#

通过自我关注，即关键和价值来自自我

103        self_attn = self.self_attn(query=z, key=z, value=z, mask=mask)

#

添加自我关注的结果

105        x = x + self.dropout(self_attn)

#

如果提供了来源，则从关注源获取结果。这是当你有一个关注编码器输出的解码器层

时

110        if src is not None:

#

归一化向量

112            z = self.norm_src_attn(x)

#

注意源。即键和值来自源

114            attn_src = self.src_attn(query=z, key=src, value=src, mask=src_mask)

#

添加来源关注结果

116            x = x + self.dropout(attn_src)

#

标准化以进行前馈

119        z = self.norm_ff(x)

#

如果已指定，则将输入保存到前馈图层

121        if self.is_save_ff_input:
122            self.ff_input = z.clone()

#

通过前馈网络

124        ff = self.feed_forward(z)

#

将前馈结果添加回来

126        x = x + self.dropout(ff)
127
128        return x

#

变压器编码

131class Encoder(nn.Module):

#

138    def __init__(self, layer: TransformerLayer, n_layers: int):
139        super().__init__()

#

制作变压器层的副本

141        self.layers = clone_module_list(layer, n_layers)

#

最终归一化层

143        self.norm = nn.LayerNorm([layer.size])

#

145    def forward(self, x: torch.Tensor, mask: torch.Tensor):

#

穿过每个变压器层

147        for layer in self.layers:
148            x = layer(x=x, mask=mask)

#

最后，对向量进行归一化

150        return self.norm(x)

#

变压器解码器

153class Decoder(nn.Module):

#

160    def __init__(self, layer: TransformerLayer, n_layers: int):
161        super().__init__()

#

制作变压器层的副本

163        self.layers = clone_module_list(layer, n_layers)

#

最终归一化层

165        self.norm = nn.LayerNorm([layer.size])

#

167    def forward(self, x: torch.Tensor, memory: torch.Tensor, src_mask: torch.Tensor, tgt_mask: torch.Tensor):

#

穿过每个变压器层

169        for layer in self.layers:
170            x = layer(x=x, mask=tgt_mask, src=memory, src_mask=src_mask)

#

最后，对向量进行归一化

172        return self.norm(x)

#

发电机

这可以预测令牌并给出其中的lof softmax。如果你正在使用，你不需要这个nn.CrossEntropyLoss 。

175class Generator(nn.Module):

#

185    def __init__(self, n_vocab: int, d_model: int):
186        super().__init__()
187        self.projection = nn.Linear(d_model, n_vocab)

#

189    def forward(self, x):
190        return self.projection(x)

#

组合式编码器-解码器

193class EncoderDecoder(nn.Module):

#

200    def __init__(self, encoder: Encoder, decoder: Decoder, src_embed: nn.Module, tgt_embed: nn.Module, generator: nn.Module):
201        super().__init__()
202        self.encoder = encoder
203        self.decoder = decoder
204        self.src_embed = src_embed
205        self.tgt_embed = tgt_embed
206        self.generator = generator

#

从他们的代码来看，这很重要。使用 Glorot/fan_avg 初始化参数。

210        for p in self.parameters():
211            if p.dim() > 1:
212                nn.init.xavier_uniform_(p)

#

214    def forward(self, src: torch.Tensor, tgt: torch.Tensor, src_mask: torch.Tensor, tgt_mask: torch.Tensor):

#

通过编码器运行源码

216        enc = self.encode(src, src_mask)

#

通过解码器运行编码和目标

218        return self.decode(enc, src_mask, tgt, tgt_mask)

#

220    def encode(self, src: torch.Tensor, src_mask: torch.Tensor):
221        return self.encoder(self.src_embed(src), src_mask)

#

223    def decode(self, memory: torch.Tensor, src_mask: torch.Tensor, tgt: torch.Tensor, tgt_mask: torch.Tensor):
224        return self.decoder(self.tgt_embed(tgt), memory, src_mask, tgt_mask)