理解transformer源码

        本文参考哈佛开源代码，对代码进行了一些修改，结合论文学习笔记加深对transformer算法整体的理解。

import copy
import math
import time
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable

##########################################################
# 生成模型
##########################################################
class EncoderDecoder(nn.Module):
    """标准的Encoder-Decoder架构"""

    def __init__(self, encoder, decoder, src_embed, tgt_embed, generator):
        super(EncoderDecoder, self).__init__()
        self.encoder = encoder
        self.decoder = decoder
        self.src_embed = src_embed  # 源序列embedding
        self.tgt_embed = tgt_embed  # 目标序列embedding
        self.generator = generator  # 生成目标单词的概率

    def forward(self, src, tgt, src_mask, tgt_mask):
        """接收和处理原序列,目标序列,以及他们的mask"""
        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)


class Generator(nn.Module):
    """定义标准的linear+softmax生成步骤"""

    def __init__(self, d_model, vocab):
        super(Generator, self).__init__()
        self.proj = nn.Linear(d_model, vocab)

    def forward(self, x):
        return F.log_softmax(self.proj(x), dim=-1)


# Encoder部分
def clones(module, N):
    """产生N个相同的层"""
    return nn.ModuleList([copy.deepcopy(module) for _ in range(N)])


class Encoder(nn.Module):
    """N层堆叠的Encoder"""

    def __init__(self, layer, N):
        super(Encoder, self).__init__()
        self.layers = clones(layer, N)
        self.norm = LayerNorm(layer.size)

    def forward(self, x, mask):
        """每层layer依次通过输入序列与mask"""
        for layer in self.layers:
            x = layer(x, mask)
        return self.norm(x)


class LayerNorm(nn.Module):
    """构造一个layernorm模块"""

    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 = eps

    def forward(self, x):
        """Norm"""
        mean = x.mean(-1, keepdim=True)
        std = x.std(-1, keepdim=True)
        return self.a_2 * (x - mean) / (std + self.eps) + self.b_2


class SublayerConnection(nn.Module):
    """Add+Norm"""

    def __init__(self, size, dropout):
        super(SublayerConnection, self).__init__()
        self.norm = LayerNorm(size)
        self.dropout = nn.Dropout(dropout)

    def forward(self, x, sublayer):
        """add norm"""
        norm = self.dropout(sublayer(self.norm(x)))
        return x + norm


class EncoderLayer(nn.Module):
    """Encoder分为两层Self-Attn和Feed Forward"""

    def __init__(self, size, self_attn, feed_forward, dropout):
        super(EncoderLayer, self).__init__()
        self.self_attn = self_attn
        self.feed_forward = feed_forward
        self.sublayer = clones(SublayerConnection(size, dropout), 2)
        self.size = size

    def forward(self, x, mask):
        """Self-Attn和Feed Forward"""
        x = self.sublayer[0](x, lambda x: self.self_attn(x, x, x, mask))
        return self.sublayer[1](x, self.feed_forward)


# Decoder部分
class Decoder(nn.Module):
    """带mask功能的通用Decoder结构"""

    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)


class DecoderLayer(nn.Module):
    """Decoder is made of self-attn, src-attn, and feed forward"""

    def __init__(self, size, self_attn, src_attn, feed_forward, dropout):
        super(DecoderLayer, self).__init__()
        self.size = size
        self.self_attn = self_attn
        self.src_attn = src_attn
        self.feed_forward = feed_forward
        self.sublayer = clones(SublayerConnection(size, dropout), 3)

    def forward(self, x, memory, src_mask, tgt_mask):
        """将decoder的三个Sublayer串联起来"""
        m = memory
        x = 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)


def subsequent_mask(size):
    """
    mask后续的位置，返回[size, size]尺寸下三角Tensor
    对角线及其左下角全是1，右上角全是0
    """
    attn_shape = (1, size, size)
    subsequent_mask = np.triu(np.ones(attn_shape), k=1).astype('uint8')
    return torch.from_numpy(subsequent_mask) == 0


# Attention
def attention(query, key, value, mask=None, dropout=None):
    """计算Attention即点乘V"""
    d_k = query.size(-1)
    # [B, h, L, L]
    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


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
        self.d_k = d_model // h
        self.h = h
        self.linears = clones(nn.Linear(d_model, d_model), 4)
        self.attn = None
        self.dropout = nn.Dropout(p=dropout)

    def forward(self, query, key, value, mask=None):
        """
        实现MultiHeadedAttention。
           输入的q，k，v是形状 [batch, L, d_model]。
           输出的x 的形状同上。
        """
        if mask is not None:
            # Same mask applied to all h heads.
            mask = mask.unsqueeze(1)
        nbatches = query.size(0)

        # 1) 这一步qkv变化:[batch, L, d_model] ->[batch, h, L, d_model/h]
        # 因为初始化不同，所以最后输出的k,q,v矩阵不同
        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) 计算注意力attn 得到attn*v 与attn
        # qkv :[batch, h, L, d_model/h] -->x:[b, h, L, d_model/h], attn[b, h, L, L]
        x, self.attn = attention(query, key, value, mask=mask, dropout=self.dropout)
        # 3) 上一步的结果合并在一起还原成原始输入序列的形状
        x = x.transpose(1, 2).contiguous().view(nbatches, -1, self.h * self.d_k)
        # 最后再过一个线性层
        return self.linears[-1](x)


# Position-wise Feed-Forward Networks
class PositionwiseFeedForward(nn.Module):
    """实现FFN函数"""

    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))))


# Embeddings
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_model  # 表示embedding的维度

    def forward(self, x):
        embed = self.lut(x)
        dk = math.sqrt(self.d_model)
        return embed * dk


# Positional Encoding
class PositionalEncoding(nn.Module):
    """实现PE功能"""

    def __init__(self, d_model, dropout, max_len=5000):
        super(PositionalEncoding, self).__init__()
        self.dropout = nn.Dropout(p=dropout)

        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)  # 偶数列
        pe[:, 1::2] = torch.cos(position * div_term)  # 奇数列
        pe = pe.unsqueeze(0)  # [1, max_len, d_model]
        self.register_buffer('pe', pe)

    def forward(self, x):
        x = x + Variable(self.pe[:, :x.size(1)], requires_grad=False)
        return self.dropout(x)


# 定义一个接受超参数并生成完整模型的函数
def make_model(src_vocab, tgt_vocab, N=6, d_model=512, d_ff=2048, h=8, dropout=0.1):
    """根据输入的超参数构建一个模型"""
    c = copy.deepcopy
    attn = MultiHeadedAttention(h, d_model)
    ff = PositionwiseFeedForward(d_model, d_ff, dropout)

    position = PositionalEncoding(d_model, dropout)
    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))

    # 使用xavier初始化参数
    for p in model.parameters():
        if p.dim() > 1:
            nn.init.xavier_uniform_(p)
    return model

之后生成模拟数据进行训练过程

# 定义个一个Batch对象
class Batch(object):
    """定义一个训练时需要的批次数据对象，封装了用于训练的src和tgt句子，以及mask"""

    def __init__(self, src, trg=None, pad=0):
        self.src = src  # B 个序列[1,5,3, 0]
        self.src_mask = (src != pad).unsqueeze(-2)  # [[1,1,1,0]]
        if trg is not None:
            self.trg = trg[:, :-1]  #
            self.trg_y = trg[:, 1:]  # 后挪一个位置开始
            self.trg_mask = \
                self.make_std_mask(self.trg, pad)
            self.ntokens = (self.trg_y != pad).data.sum()

    @staticmethod
    def 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


# 定义一个训练函数用于训练和计算损失、更新梯度
def run_epoch(data_iter, model, loss_compute, device):
    """提供训练和日志功能"""
    start = time.time()
    total_tokens = 0
    total_loss = 0
    tokens = 0
    for i, batch in enumerate(data_iter):
        src = batch.src.to(device)
        trg = batch.trg.to(device)
        src_mask = batch.src_mask.to(device)
        trg_mask = batch.trg_mask.to(device)
        trg_y = batch.trg_y.to(device)
        ntokens = batch.ntokens.to(device)

        out = model.forward(src, trg, src_mask, trg_mask)
        loss = loss_compute(out, trg_y, ntokens)
        total_loss += loss.detach().cpu().numpy()
        total_tokens += ntokens.cpu().numpy()
        tokens += ntokens.cpu().numpy()
        if i % 50 == 1:
            elapsed = time.time() - start
            print("Epoch Step: %d Loss: %f Tokens per Sec: %f" %
                  (i, loss.detach().cpu().numpy() / ntokens.cpu().numpy(), tokens / elapsed))
            start = time.time()
            tokens = 0
    return total_loss / total_tokens


class NoamOpt(object):
    """Optim wrapper that implements rate."""

    def __init__(self, model_size, factor, warmup, optimizer):
        self.optimizer = optimizer
        self._step = 0
        self.warmup = warmup
        self.factor = factor
        self.model_size = model_size
        self._rate = 0

    def step(self):
        """Update parameters and rate"""
        self._step += 1
        rate = self.rate()
        for p in self.optimizer.param_groups:
            p['lr'] = rate
        self._rate = rate
        self.optimizer.step()

    def rate(self, step=None):
        """Implement `lrate` above"""
        if step is None:
            step = self._step
        return self.factor * \
               (self.model_size ** (-0.5) *
                min(step ** (-0.5), step * self.warmup ** (-1.5)))


class LabelSmoothing(nn.Module):
    """实现labelsmoothing."""

    def __init__(self, size, padding_idx, smoothing=0.0):
        super(LabelSmoothing, self).__init__()
        self.criterion = nn.KLDivLoss(reduction='sum')
        self.padding_idx = padding_idx
        self.confidence = 1.0 - smoothing
        self.smoothing = smoothing
        self.size = size
        self.true_dist = None

    def forward(self, x, target):
        assert x.size(1) == self.size
        true_dist = x.data.clone()
        true_dist.fill_(self.smoothing / (self.size - 2))
        true_dist.scatter_(1, target.data.type(torch.int64).unsqueeze(1), self.confidence)
        # true_dist.scatter_(1, target.data.unsqueeze(1), self.confidence)
        true_dist[:, self.padding_idx] = 0
        mask = torch.nonzero(target.data == self.padding_idx)
        if mask.dim() > 0:
            true_dist.index_fill_(0, mask.squeeze(), 0.0)
        self.true_dist = true_dist
        return self.criterion(x, Variable(true_dist, requires_grad=False))


# 数据生成
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] = 1.
        src = Variable(data, requires_grad=False)
        tgt = Variable(data, requires_grad=False)
        yield Batch(src, tgt, 0)

class SimpleLossCompute(object):
    """A simple loss compute and train function."""
    def __init__(self, generator, criterion, opt=None):
        self.generator = generator
        self.criterion = criterion
        self.opt = opt

    def __call__(self, x, y, norm):
        x = self.generator(x)
        loss = self.criterion(x.contiguous().view(-1, x.size(-1)),
                              y.contiguous().view(-1)) / norm
        loss.backward()
        if self.opt is not None:
            self.opt.step()
            self.opt.optimizer.zero_grad()
        return loss.item() * norm.float()


V = 11
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
criterion = LabelSmoothing(size=V, padding_idx=0, smoothing=0.1)

model = make_model(V, V, N=2)
model = model.to(device)
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))

for epoch in range(5):
    model.train()
    loss_func = SimpleLossCompute(model.generator, criterion, model_opt)
    run_epoch(data_gen(V, 30, 20), model, loss_func, device)
    model.eval()
    print(run_epoch(data_gen(V, 30, 5), model,
                  SimpleLossCompute(model.generator, criterion, None), device))

        将以上代码放在同一个py文件中，通过断点调试可以加深对transformer算法整体框架的理解。