資源來自：DataWhale?學(xué)習(xí)資料

????????最近看了DataWhale 的Transformer圖解，突然對Transformer的結(jié)構(gòu)圖有了更加清晰的理解，特此記錄。

1、大框架

Transformer是由6個encoder和6個decoder組成，模型的具體實現(xiàn)是model變量里邊，參數(shù)有Encoder[編碼器]、Decoder[解碼器]、Embedding(src_vocab)[輸入文本進(jìn)行詞向量化]、Embedding(tgt_vocab)[目標(biāo)文本進(jìn)行詞向量化]，Generator[生成器]。

def make_model(src_vocab, tgt_vocab, N=6, d_model=512, d_ff=2048, h=8, dropout=0.1):"Helper: Construct a model from hyperparameters."c = copy.deepcopy#多頭注意力attn = MultiHeadedAttention(h, d_model)#前饋神經(jīng)網(wǎng)絡(luò)ff = PositionwiseFeedForward(d_model, d_ff, dropout)#位置編碼position = PositionalEncoding(d_model, dropout)#模型定義model = EncoderDecoder(Encoder(EncoderLayer(d_model, c(attn), c(ff), dropout), N),Decoder(DecoderLayer(d_model, c(attn), c(attn), c(ff), dropout), N),nn.Sequential(Embeddings(d_model, src_vocab), c(position)),nn.Sequential(Embeddings(d_model, tgt_vocab), c(position)),Generator(d_model, tgt_vocab))# This was important from their code. # Initialize parameters with Glorot / fan_avg.for p in model.parameters():if p.dim() > 1:nn.init.xavier_uniform(p)return model

查看EncoderDecoder函數(shù)，搭建了一個seq2seq框架，即包含encoder和decoder，在EncoderDecoder函數(shù)中，變量src是輸入文本，tgt是輸出文本，src_mask是輸入文本的掩碼，tgt_mask是輸出文本的掩碼，memory是encoder的最終輸出。

class EncoderDecoder(nn.Module):"""基礎(chǔ)的Encoder-Decoder結(jié)構(gòu)。A standard Encoder-Decoder architecture. Base for this and many other models."""def __init__(self, encoder, decoder, src_embed, tgt_embed, generator):super(EncoderDecoder, self).__init__()self.encoder = encoderself.decoder = decoderself.src_embed = src_embedself.tgt_embed = tgt_embedself.generator = generatordef forward(self, src, tgt, src_mask, tgt_mask):"Take in and process masked src and target sequences."return self.decode(self.encode(src, src_mask), src_mask,tgt, tgt_mask)def encode(self, src, src_mask):return self.encoder(self.src_embed(src), src_mask)def decode(self, memory, src_mask, tgt, tgt_mask):return self.decoder(self.tgt_embed(tgt), memory, src_mask, tgt_mask)

2、Encoder

（1）clone

由于Transformer是有6個encoder組成，則使用clone函數(shù)進(jìn)行EncodeLayer層的復(fù)制：

def clones(module, N):"產(chǎn)生N個完全相同的網(wǎng)絡(luò)層""Produce N identical layers."return nn.ModuleList([copy.deepcopy(module) for _ in range(N)])

（解釋：nn.ModuleList 函數(shù)是保存子模塊列表通過for循環(huán)，建立6個Encoder）

（2）Encoder

class Encoder(nn.Module):"完整的Encoder包含N層"def __init__(self, layer, N):super(Encoder, self).__init__()self.layers = clones(layer, N)self.norm = LayerNorm(layer.size)def forward(self, x, mask):"每一層的輸入是x和mask"for layer in self.layers:x = layer(x, mask)return self.norm(x)

Encoder需要進(jìn)行“層歸一化”，因此是在encoder建立之后進(jìn)行了LayerNorm操作。

?（3）EncoderLayer

先介紹EncoderLayer層（一個編碼器encoder），編碼器的構(gòu)成部分是self_Attention->SubLayerConnection(層歸一化和殘差連接)->FFNN->SubLayerConnection(層歸一化和殘差連接).

代碼中，對SubLayerConnection復(fù)制兩份，分別加入在self-Attention和FFNN之后。

class EncoderLayer(nn.Module):"Encoder is made up of self-attn and feed forward (defined below)"def __init__(self, size, self_attn, feed_forward, dropout):super(EncoderLayer, self).__init__()self.self_attn = self_attnself.feed_forward = feed_forwardself.sublayer = clones(SublayerConnection(size, dropout), 2)self.size = sizedef forward(self, x, mask):"Follow Figure 1 (left) for connections."x = self.sublayer[0](x, lambda x: self.self_attn(x, x, x, mask))return self.sublayer[1](x, self.feed_forward)

（4）LayerNorm

‘層歸一化’是該層的輸入值進(jìn)行對歸一化處理，公式為,層歸一化分別在Encoder中的Attention（自身注意力）和FFNN(前饋神經(jīng)網(wǎng)絡(luò))模塊后。

class LayerNorm(nn.Module):"Construct a layernorm module (See citation for details)."def __init__(self, features, eps=1e-6):super(LayerNorm, self).__init__()self.a_2 = nn.Parameter(torch.ones(features))self.b_2 = nn.Parameter(torch.zeros(features))self.eps = epsdef forward(self, x):mean = x.mean(-1, keepdim=True)std = x.std(-1, keepdim=True)return self.a_2 * (x - mean) / (std + self.eps) + self.b_2

（5）殘差連接

所進(jìn)行的操作時，對輸入數(shù)據(jù)進(jìn)行層歸一化，然后進(jìn)行sublayer操作，此時sublayer傳入的操作是self.attn和self.feed_forward.

class SublayerConnection(nn.Module):"""A residual connection followed by a layer norm.Note for code simplicity the norm is first as opposed to last."""def __init__(self, size, dropout):super(SublayerConnection, self).__init__()self.norm = LayerNorm(size)self.dropout = nn.Dropout(dropout)def forward(self, x, sublayer):"Apply residual connection to any sublayer with the same size."return x + self.dropout(sublayer(self.norm(x)))

因為self-Attention和FFNN在encoder和decoder有異同，下邊進(jìn)行集中梳理。

3、Decoder

（1）Decoder

在Decoder中，也進(jìn)行了clone操作，此處相較于encoder，多了一個memory和src、tgt的掩碼mask。

class Decoder(nn.Module):"Generic N layer decoder with masking."def __init__(self, layer, N):super(Decoder, self).__init__()self.layers = clones(layer, N)self.norm = LayerNorm(layer.size)def forward(self, x, memory, src_mask, tgt_mask):for layer in self.layers:x = layer(x, memory, src_mask, tgt_mask)return self.norm(x)

（2）DecoderLayer

相較于EncoderLayer層，多了一個attention操作，即self_attn是在decoder的注意力機(jī)制，即增加了mask機(jī)制，src_attn是encoder的輸出結(jié)果，q是decoder層，k,v是encoder的輸出。

?（模塊1是self_attn,模塊2是src_attn）

由于新增一個attention模塊，SubLayerConnection就有三層，解碼器的構(gòu)成部分是self_Attention->SubLayerConnection(層歸一化和殘差連接)->src_Attention->SubLayerConnection(層歸一化和殘差連接)->FFNN->SubLayerConnection(層歸一化和殘差連接).

class DecoderLayer(nn.Module):"Decoder is made of self-attn, src-attn, and feed forward (defined below)"def __init__(self, size, self_attn, src_attn, feed_forward, dropout):super(DecoderLayer, self).__init__()self.size = sizeself.self_attn = self_attnself.src_attn = src_attnself.feed_forward = feed_forwardself.sublayer = clones(SublayerConnection(size, dropout), 3)def forward(self, x, memory, src_mask, tgt_mask):"Follow Figure 1 (right) for connections."m = memoryx = self.sublayer[0](x, lambda x: self.self_attn(x, x, x, tgt_mask))x = self.sublayer[1](x, lambda x: self.src_attn(x, m, m, src_mask))return self.sublayer[2](x, self.feed_forward)

4、Embedding? src_vocab & tgt_vocab

Embedding是對文本進(jìn)行詞向量轉(zhuǎn)換，調(diào)用函數(shù)為nn.Embedding,且進(jìn)行了math.sqrt(self.d_model)操作。

class Embeddings(nn.Module):def __init__(self, d_model, vocab):super(Embeddings, self).__init__()self.lut = nn.Embedding(vocab, d_model)self.d_model = d_modeldef forward(self, x):return self.lut(x) * math.sqrt(self.d_model)

5、額外實現(xiàn)

（1）self-Attention

• Attention計算

目前，Atention機(jī)制的演變過程是加法和點(diǎn)積計算，加法計算是計算q,k的相似度，點(diǎn)積是計算q,k的點(diǎn)積，公式為點(diǎn)積計算。

在進(jìn)行Attention計算時，特別注意mask參數(shù)， 當(dāng)mask不為None時，則加入了Mask機(jī)制

def attention(query, key, value, mask=None, dropout=None):"Compute 'Scaled Dot Product Attention'"d_k = query.size(-1)scores = torch.matmul(query, key.transpose(-2, -1)) \/ math.sqrt(d_k)if mask is not None:scores = scores.masked_fill(mask == 0, -1e9)p_attn = F.softmax(scores, dim = -1)if dropout is not None:p_attn = dropout(p_attn)return torch.matmul(p_attn, value), p_attn

• Multi-Head

只計算單個Attention很難捕捉輸入句中所有空間的訊息，為了優(yōu)化模型，論文提出了一個multi head的概念，把key,value,query線性映射到不同空間h次，但是在傳入Scaled-Dot-Product Attention中時，需要固定的長度，因此再對head進(jìn)行concat。

?代碼如下：

class MultiHeadedAttention(nn.Module):def __init__(self, h, d_model, dropout=0.1):"Take in model size and number of heads."super(MultiHeadedAttention, self).__init__()assert d_model % h == 0# We assume d_v always equals d_kself.d_k = d_model // hself.h = hself.linears = clones(nn.Linear(d_model, d_model), 4)self.attn = Noneself.dropout = nn.Dropout(p=dropout)def forward(self, query, key, value, mask=None):"Implements Figure 2"if mask is not None:# Same mask applied to all h heads.mask = mask.unsqueeze(1)nbatches = query.size(0)# 1) Do all the linear projections in batch from d_model => h x d_k query, key, value = \[l(x).view(nbatches, -1, self.h, self.d_k).transpose(1, 2)for l, x in zip(self.linears, (query, key, value))]# 2) Apply attention on all the projected vectors in batch. x, self.attn = attention(query, key, value, mask=mask, dropout=self.dropout)# 3) "Concat" using a view and apply a final linear. x = x.transpose(1, 2).contiguous() \.view(nbatches, -1, self.h * self.d_k)return self.linears[-1](x)

?定義了4個linear層，前三個分別對q,v,k進(jìn)行分解，維度是(h,d_k,關(guān)系是d_model = h*d_k，h是head的數(shù)量)，最后一個linear層是對多頭的連接之后的數(shù)據(jù)進(jìn)行線性變換。

• mask機(jī)制

mask機(jī)制就是防止在訓(xùn)練的時候使用未來的輸出的單詞，確保對位置i的預(yù)測僅依賴于已知的位置i之前的輸出，而不會依賴于位置i之后的輸出。 比如訓(xùn)練時， 第一個單詞是不能參考第二個單詞的生成結(jié)果的。 mask就會把這個信息變成0， 用來保證預(yù)測位置 i 的信息只能基于比 i 小的輸出；

def subsequent_mask(size):"Mask out subsequent positions."attn_shape = (1, size, size)subsequent_mask = np.triu(np.ones(attn_shape), k=1).astype('uint8')return torch.from_numpy(subsequent_mask) == 0

生成一個上三角矩陣，令size=3，測試結(jié)果為

（2）FFNN

FFNN有兩層線性變換，結(jié)構(gòu)是linear->relu->dropout->linear。

class PositionwiseFeedForward(nn.Module):"Implements FFN equation."def __init__(self, d_model, d_ff, dropout=0.1):super(PositionwiseFeedForward, self).__init__()self.w_1 = nn.Linear(d_model, d_ff)self.w_2 = nn.Linear(d_ff, d_model)self.dropout = nn.Dropout(dropout)def forward(self, x):return self.w_2(self.dropout(F.relu(self.w_1(x))))

（3）位置編碼

????????encoder的輸入層和decoder的輸入層是一樣的結(jié)構(gòu)，都是token embedding（詞向量）+ positional embedding（位置向量），得到最終的輸入向量。之所以引入positional embedding主要是解決單單使用token embedding（類似于詞袋子），并沒有詞序的概念的問題。因為該模型并不包括任何的循環(huán)或卷積神經(jīng)網(wǎng)絡(luò)，所以模型添加了位置編碼，為模型提供了關(guān)于單詞再句子中相對位置的信息。這個向量能決定當(dāng)前詞的位置，或者說在一個句子中不同的詞之間的距離。計算方法如下：

pos表示單詞的位置，i是指單詞的維度，偶數(shù)位置用正弦，奇數(shù)位置用余弦。

class PositionalEncoding(nn.Module):"Implement the PE function."def __init__(self, d_model, dropout, max_len=5000):super(PositionalEncoding, self).__init__()self.dropout = nn.Dropout(p=dropout)# Compute the positional encodings once in log space.pe = torch.zeros(max_len, d_model)position = torch.arange(0, max_len).unsqueeze(1)div_term = torch.exp(torch.arange(0, d_model, 2) *-(math.log(10000.0) / d_model))pe[:, 0::2] = torch.sin(position * div_term)pe[:, 1::2] = torch.cos(position * div_term)pe = pe.unsqueeze(0)self.register_buffer('pe', pe)def forward(self, x):x = x + Variable(self.pe[:, :x.size(1)], requires_grad=False)return self.dropout(x)

squeeze和unsqueeze函數(shù)：對張量Tensor的維度進(jìn)行壓縮或者擴(kuò)充！！！

6、實現(xiàn)順序

（1）模擬數(shù)據(jù)

def data_gen(V, batch, nbatches):"Generate random data for a src-tgt copy task."for i in range(nbatches):data = torch.from_numpy(np.random.randint(1, V, size=(batch, 10)))data[:, 0] = 1src = Variable(data, requires_grad=False)tgt = Variable(data, requires_grad=False)yield Batch(src, tgt, 0)

（2）批處理和掩碼

class Batch:"Object for holding a batch of data with mask during training."def __init__(self, src, trg=None, pad=0):self.src = srcself.src_mask = (src != pad).unsqueeze(-2)if trg is not None:self.trg = trg[:, :-1]self.trg_y = trg[:, 1:]self.trg_mask = \self.make_std_mask(self.trg, pad)self.ntokens = (self.trg_y != pad).data.sum()@staticmethoddef make_std_mask(tgt, pad):"Create a mask to hide padding and future words."tgt_mask = (tgt != pad).unsqueeze(-2)tgt_mask = tgt_mask & Variable(subsequent_mask(tgt.size(-1)).type_as(tgt_mask.data))return tgt_mask

（3）模型優(yōu)化

class NoamOpt:"Optim wrapper that implements rate."def __init__(self, model_size, factor, warmup, optimizer):self.optimizer = optimizerself._step = 0self.warmup = warmupself.factor = factorself.model_size = model_sizeself._rate = 0def step(self):"Update parameters and rate"self._step += 1rate = self.rate()for p in self.optimizer.param_groups:p['lr'] = rateself._rate = rateself.optimizer.step()def rate(self, step = None):"Implement `lrate` above"if step is None:step = self._stepreturn self.factor * \(self.model_size ** (-0.5) *min(step ** (-0.5), step * self.warmup ** (-1.5)))def get_std_opt(model):return NoamOpt(model.src_embed[0].d_model, 2, 4000,torch.optim.Adam(model.parameters(), lr=0, betas=(0.9, 0.98), eps=1e-9))

（4）標(biāo)簽平滑

class LabelSmoothing(nn.Module):"Implement label smoothing."def __init__(self, size, padding_idx, smoothing=0.0):super(LabelSmoothing, self).__init__()self.criterion = nn.KLDivLoss(size_average=False)self.padding_idx = padding_idxself.confidence = 1.0 - smoothingself.smoothing = smoothingself.size = sizeself.true_dist = Nonedef forward(self, x, target):assert x.size(1) == self.sizetrue_dist = x.data.clone()true_dist.fill_(self.smoothing / (self.size - 2))true_dist.scatter_(1, target.data.unsqueeze(1), self.confidence)true_dist[:, self.padding_idx] = 0mask = torch.nonzero(target.data == self.padding_idx)if mask.dim() > 0:true_dist.index_fill_(0, mask.squeeze(), 0.0)self.true_dist = true_distreturn self.criterion(x, Variable(true_dist, requires_grad=False))

（5）損失函數(shù)計算

class SimpleLossCompute:"A simple loss compute and train function."def __init__(self, generator, criterion, opt=None):self.generator = generatorself.criterion = criterionself.opt = optdef __call__(self, x, y, norm):x = self.generator(x)loss = self.criterion(x.contiguous().view(-1, x.size(-1)), y.contiguous().view(-1)) / normloss.backward()if self.opt is not None:self.opt.step()self.opt.optimizer.zero_grad()return loss.item() * norm

（6）批次運(yùn)行

def run_epoch(data_iter, model, loss_compute):"Standard Training and Logging Function"start = time.time()total_tokens = 0total_loss = 0tokens = 0for i, batch in enumerate(data_iter):out = model.forward(batch.src, batch.trg, batch.src_mask, batch.trg_mask)loss = loss_compute(out, batch.trg_y, batch.ntokens)total_loss += losstotal_tokens += batch.ntokenstokens += batch.ntokensif i % 50 == 1:elapsed = time.time() - startprint("Epoch Step: %d Loss: %f Tokens per Sec: %f" %(i, loss / batch.ntokens, tokens / elapsed))start = time.time()tokens = 0return total_loss / total_tokens

（7）調(diào)用

# Train the simple copy task. V = 11 #標(biāo)簽平滑 criterion = LabelSmoothing(size=V, padding_idx=0, smoothing=0.0) #定義模型 model = make_model(V, V, N=2) #模型優(yōu)化，采用Adam model_opt = NoamOpt(model.src_embed[0].d_model, 1, 400,torch.optim.Adam(model.parameters(), lr=0, betas=(0.9, 0.98), eps=1e-9))#訓(xùn)練10次，并進(jìn)行損失函數(shù)計算 for epoch in range(10):model.train()run_epoch(data_gen(V, 30, 20), model, SimpleLossCompute(model.generator, criterion, model_opt))model.eval()print(run_epoch(data_gen(V, 30, 5), model, SimpleLossCompute(model.generator, criterion, None)))

參考教程：

1、learn-nlp-with-transformers/2.2.1-Pytorch編寫Transformer.md at main · datawhalechina/learn-nlp-with-transformers · GitHub

（原文鏈接）：The Annotated Transformer
2、Datawhale-零基礎(chǔ)入門NLP-新聞文本分類Task06_櫻緣之夢-CSDN博客

總結(jié)

以上是生活随笔為你收集整理的Transformer模型拆解分析的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。