序列对序列建模nn.Transformer和TorchText

译者：dabney777

校验：dabney777

本教程将会使用 nn.Transformer 模块训练一个序列到序列模型。

PyTorch 1.2 版本依据论文 Attention is All You Need 发布了标准的 transformer 模型。Transformer 模型已被证明在解决序列到序列问题时效果优异。

nn.Transformer 模块通过注意力机制( nn.MultiheadAttention )来取得输入与输出之间的全局相关性。nn.Transformer 模块现已高度模块化，可以直接用于构建其他模型(如 nn.TransformerEncoder)。

img/transformer_architecture.jpg

定义模型

在本教程中，我们训练 nn.TransformerEncoder 用于构建语言模型。语言模型的目标是对给定字/词序列打分，判断该字/词序列出现在文本中的概率。字符序列首先会被传进 embedding 层转化为向量，然后被传入位置编码层 (详见下段）。 nn.TransformerEncoder 由多个编码层nn.TransformerEncoderLayer组成。对输入序列的每一维需要施加一个自注意力权重影响。nn.TransformerEncoder 的自注意力权重只影响序列中靠前的数据，不修改之后位置的数据。在本任务中，nn.TransformerEncoder 的输出将会被送至最终的线性层，该层为一个 log-Softmax 层。

  1. import math
  2. import torch
  3. import torch.nn as nn
  4. import torch.nn.functional as F
  6. class TransformerModel(nn.Module):
  8.     def __init__(self, ntoken, ninp, nhead, nhid, nlayers, dropout=0.5):
  9.         super(TransformerModel, self).__init__()
  10.         from torch.nn import TransformerEncoder, TransformerEncoderLayer
  11.         self.model_type = 'Transformer'
  12.         self.src_mask = None
  13.         self.pos_encoder = PositionalEncoding(ninp, dropout)
  14.         encoder_layers = TransformerEncoderLayer(ninp, nhead, nhid, dropout)
  15.         self.transformer_encoder = TransformerEncoder(encoder_layers, nlayers)
  16.         self.encoder = nn.Embedding(ntoken, ninp)
  17.         self.ninp = ninp
  18.         self.decoder = nn.Linear(ninp, ntoken)
  20.         self.init_weights()
  22.     def _generate_square_subsequent_mask(self, sz):
  23.         mask = (torch.triu(torch.ones(sz, sz)) == 1).transpose(0, 1)
  24.         mask = mask.float().masked_fill(mask == 0, float('-inf')).masked_fill(mask == 1, float(0.0))
  25.         return mask
  27.     def init_weights(self):
  28.         initrange = 0.1
  29.         self.encoder.weight.data.uniform_(-initrange, initrange)
  30.         self.decoder.bias.data.zero_()
  31.         self.decoder.weight.data.uniform_(-initrange, initrange)
  33.     def forward(self, src):
  34.         if self.src_mask is None or self.src_mask.size(0) != len(src):
  35.             device = src.device
  36.             mask = self._generate_square_subsequent_mask(len(src)).to(device)
  37.             self.src_mask = mask
  39.         src = self.encoder(src) * math.sqrt(self.ninp)
  40.         src = self.pos_encoder(src)
  41.         output = self.transformer_encoder(src, self.src_mask)
  42.         output = self.decoder(output)
  43.         return F.log_softmax(output, dim=-1)

PositionalEncoding 模块将字/词在序列中的绝对位置或相对位置信息编码。 位置编码与嵌入层具有相同的维度，这样位置信息向量和嵌入向量可以直接相加。 这里，我们使用 sin 和 cos 函数在不同位置的值来作为位置编码的值。具体计算公式见下方代码。

  1. class PositionalEncoding(nn.Module):
  3.     def __init__(self, d_model, dropout=0.1, max_len=5000):
  4.         super(PositionalEncoding, self).__init__()
  5.         self.dropout = nn.Dropout(p=dropout)
  7.         pe = torch.zeros(max_len, d_model)
  8.         position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)
  9.         div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-math.log(10000.0) / d_model))
  10.         pe[:, 0::2] = torch.sin(position * div_term)
  11.         pe[:, 1::2] = torch.cos(position * div_term)
  12.         pe = pe.unsqueeze(0).transpose(0, 1)
  13.         self.register_buffer('pe', pe)
  15.     def forward(self, x):
  16.         x = x + self.pe[:x.size(0), :]
  17.         return self.dropout(x)

加载和整合数据

训练过程中使用的数据机是从 torchtext 中得到的wikitext的-2数据集。词典对象基于训练数据集进行构建。batchify(） 函数把数据集中的数据排到多个列中，在划分成多个大小为 batch_size 的集合后，剩下的少于 batch_size 个数据会被丢弃。例如，对于字母序列(长度为26, batch_size 为4），将按照以下方法划分：

\[\begin{split}\begin{bmatrix} \text{A} & \text{B} & \text{C} & \ldots & \text{X} & \text{Y} & \text{Z} \end{bmatrix} \Rightarrow \begin{bmatrix} \begin{bmatrix}\text{A} \\ \text{B} \\ \text{C} \\ \text{D} \\ \text{E} \\ \text{F}\end{bmatrix} & \begin{bmatrix}\text{G} \\ \text{H} \\ \text{I} \\ \text{J} \\ \text{K} \\ \text{L}\end{bmatrix} & \begin{bmatrix}\text{M} \\ \text{N} \\ \text{O} \\ \text{P} \\ \text{Q} \\ \text{R}\end{bmatrix} & \begin{bmatrix}\text{S} \\ \text{T} \\ \text{U} \\ \text{V} \\ \text{W} \\ \text{X}\end{bmatrix} \end{bmatrix}\end{split}\]

对于我们的模型来说，只学习同一列中的数据的关系，不同的列各自独立。即我们的模型无法学习到 G 和 F 之间的联系，这样可以增加模型的并行度，增加学习效率。

  1. import torchtext
  2. from torchtext.data.utils import get_tokenizer
  3. TEXT = torchtext.data.Field(tokenize=get_tokenizer("basic_english"),
  4.                             init_token='<sos>',
  5.                             eos_token='<eos>',
  6.                             lower=True)
  7. train_txt, val_txt, test_txt = torchtext.datasets.WikiText2.splits(TEXT)
  8. TEXT.build_vocab(train_txt)
  9. device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
  11. def batchify(data, bsz):
  12.     data = TEXT.numericalize([data.examples[0].text])
  13.     # Divide the dataset into bsz parts.
  14.     nbatch = data.size(0) // bsz
  15.     # Trim off any extra elements that wouldn't cleanly fit (remainders).
  16.     data = data.narrow(0, 0, nbatch * bsz)
  17.     # Evenly divide the data across the bsz batches.
  18.     data = data.view(bsz, -1).t().contiguous()
  19.     return data.to(device)
  21. batch_size = 20
  22. eval_batch_size = 10
  23. train_data = batchify(train_txt, batch_size)
  24. val_data = batchify(val_txt, eval_batch_size)
  25. test_data = batchify(test_txt, eval_batch_size)

输出：

  1. downloading wikitext-2-v1.zip
  2. extracting

生成训练数据(输入和目标输出)的函数

get_batch() 函数生成用于 transformer 模型的输入和目标序列。它把源数据细分为长度为 bptt 的块。对于语言模型，需要当前词的下一个词作为目标词。例如当 bptt 为2， i =0 时，该函数会产生以下数据：

img/transformer_input_target.png

张量的第0维是不同的块，块的大小与 Transformer 中的编码层大小一致。张量的第1维大小为 batch 大小。

  1. bptt = 35
  2. def get_batch(source, i):
  3.     seq_len = min(bptt, len(source) - 1 - i)
  4.     data = source[i:i+seq_len]
  5.     target = source[i+1:i+1+seq_len].view(-1)
  6.     return data, target

初始化模型

模型的超参数如下，词典大小为 vocab 数组的长度。

  1. ntokens = len(TEXT.vocab.stoi) # the size of vocabulary
  2. emsize = 200 # embedding dimension
  3. nhid = 200 # the dimension of the feedforward network model in nn.TransformerEncoder
  4. nlayers = 2 # the number of nn.TransformerEncoderLayer in nn.TransformerEncoder
  5. nhead = 2 # the number of heads in the multiheadattention models
  6. dropout = 0.2 # the dropout value
  7. model = TransformerModel(ntokens, emsize, nhead, nhid, nlayers, dropout).to(device)

运行模型

模型使用交叉墒( CrossEntropyLoss )作为损失函数，使用随机梯度下降( SGD )方法更新参数。初始学习率设置为5.0。 StepLR 用于调节学习速率。在训练过程中，使用nn.utils.clipgrad_norm 函数限制梯度大小以防梯度爆炸。

  1. criterion = nn.CrossEntropyLoss()
  2. lr = 5.0 # learning rate
  3. optimizer = torch.optim.SGD(model.parameters(), lr=lr)
  4. scheduler = torch.optim.lr_scheduler.StepLR(optimizer, 1.0, gamma=0.95)
  6. import time
  7. def train():
  8.     model.train() # Turn on the train mode
  9.     total_loss = 0.
  10.     start_time = time.time()
  11.     ntokens = len(TEXT.vocab.stoi)
  12.     for batch, i in enumerate(range(0, train_data.size(0) - 1, bptt)):
  13.         data, targets = get_batch(train_data, i)
  14.         optimizer.zero_grad()
  15.         output = model(data)
  16.         loss = criterion(output.view(-1, ntokens), targets)
  17.         loss.backward()
  18.         torch.nn.utils.clip_grad_norm_(model.parameters(), 0.5)
  19.         optimizer.step()
  21.         total_loss += loss.item()
  22.         log_interval = 200
  23.         if batch % log_interval == 0 and batch > 0:
  24.             cur_loss = total_loss / log_interval
  25.             elapsed = time.time() - start_time
  26.             print('| epoch {:3d} | {:5d}/{:5d} batches | '
  27.                   'lr {:02.2f} | ms/batch {:5.2f} | '
  28.                   'loss {:5.2f} | ppl {:8.2f}'.format(
  29.                     epoch, batch, len(train_data) // bptt, scheduler.get_lr()[0],
  30.                     elapsed * 1000 / log_interval,
  31.                     cur_loss, math.exp(cur_loss)))
  32.             total_loss = 0
  33.             start_time = time.time()
  35. def evaluate(eval_model, data_source):
  36.     eval_model.eval() # Turn on the evaluation mode
  37.     total_loss = 0.
  38.     ntokens = len(TEXT.vocab.stoi)
  39.     with torch.no_grad():
  40.         for i in range(0, data_source.size(0) - 1, bptt):
  41.             data, targets = get_batch(data_source, i)
  42.             output = eval_model(data)
  43.             output_flat = output.view(-1, ntokens)
  44.             total_loss += len(data) * criterion(output_flat, targets).item()
  45.     return total_loss / (len(data_source) - 1)

在每个 epoch 结束时，若验证集的损失函数为最低则会更新一次学习率。

  1. best_val_loss = float("inf")
  2. epochs = 3 # The number of epochs
  3. best_model = None
  5. for epoch in range(1, epochs + 1):
  6.     epoch_start_time = time.time()
  7.     train()
  8.     val_loss = evaluate(model, val_data)
  9.     print('-' * 89)
  10.     print('| end of epoch {:3d} | time: {:5.2f}s | valid loss {:5.2f} | '
  11.           'valid ppl {:8.2f}'.format(epoch, (time.time() - epoch_start_time),
  12.                                      val_loss, math.exp(val_loss)))
  13.     print('-' * 89)
  15.     if val_loss < best_val_loss:
  16.         best_val_loss = val_loss
  17.         best_model = model
  19.     scheduler.step()

输出:

  1. | epoch   1 |   200/ 2981 batches | lr 5.00 | ms/batch 35.59 | loss  8.12 | ppl  3348.51
  2. | epoch   1 |   400/ 2981 batches | lr 5.00 | ms/batch 34.57 | loss  6.82 | ppl   912.80
  3. | epoch   1 |   600/ 2981 batches | lr 5.00 | ms/batch 34.55 | loss  6.39 | ppl   597.41
  4. | epoch   1 |   800/ 2981 batches | lr 5.00 | ms/batch 34.59 | loss  6.25 | ppl   517.17
  5. | epoch   1 |  1000/ 2981 batches | lr 5.00 | ms/batch 34.58 | loss  6.12 | ppl   455.67
  6. | epoch   1 |  1200/ 2981 batches | lr 5.00 | ms/batch 34.59 | loss  6.09 | ppl   442.33
  7. | epoch   1 |  1400/ 2981 batches | lr 5.00 | ms/batch 34.60 | loss  6.04 | ppl   421.27
  8. | epoch   1 |  1600/ 2981 batches | lr 5.00 | ms/batch 34.59 | loss  6.05 | ppl   423.61
  9. | epoch   1 |  1800/ 2981 batches | lr 5.00 | ms/batch 34.60 | loss  5.96 | ppl   386.26
  10. | epoch   1 |  2000/ 2981 batches | lr 5.00 | ms/batch 34.60 | loss  5.96 | ppl   387.13
  11. | epoch   1 |  2200/ 2981 batches | lr 5.00 | ms/batch 34.60 | loss  5.85 | ppl   347.56
  12. | epoch   1 |  2400/ 2981 batches | lr 5.00 | ms/batch 34.60 | loss  5.89 | ppl   362.72
  13. | epoch   1 |  2600/ 2981 batches | lr 5.00 | ms/batch 34.60 | loss  5.90 | ppl   363.70
  14. | epoch   1 |  2800/ 2981 batches | lr 5.00 | ms/batch 34.61 | loss  5.80 | ppl   330.43
  15. -----------------------------------------------------------------------------------------
  16. | end of epoch   1 | time: 107.65s | valid loss  5.77 | valid ppl   321.01
  17. -----------------------------------------------------------------------------------------
  18. | epoch   2 |   200/ 2981 batches | lr 4.75 | ms/batch 34.78 | loss  5.81 | ppl   333.28
  19. | epoch   2 |   400/ 2981 batches | lr 4.75 | ms/batch 34.63 | loss  5.78 | ppl   324.24
  20. | epoch   2 |   600/ 2981 batches | lr 4.75 | ms/batch 34.62 | loss  5.61 | ppl   272.10
  21. | epoch   2 |   800/ 2981 batches | lr 4.75 | ms/batch 34.62 | loss  5.65 | ppl   283.77
  22. | epoch   2 |  1000/ 2981 batches | lr 4.75 | ms/batch 34.61 | loss  5.60 | ppl   269.12
  23. | epoch   2 |  1200/ 2981 batches | lr 4.75 | ms/batch 34.63 | loss  5.62 | ppl   275.40
  24. | epoch   2 |  1400/ 2981 batches | lr 4.75 | ms/batch 34.62 | loss  5.62 | ppl   276.93
  25. | epoch   2 |  1600/ 2981 batches | lr 4.75 | ms/batch 34.62 | loss  5.66 | ppl   287.64
  26. | epoch   2 |  1800/ 2981 batches | lr 4.75 | ms/batch 34.63 | loss  5.59 | ppl   268.86
  27. | epoch   2 |  2000/ 2981 batches | lr 4.75 | ms/batch 34.62 | loss  5.63 | ppl   277.73
  28. | epoch   2 |  2200/ 2981 batches | lr 4.75 | ms/batch 34.63 | loss  5.52 | ppl   249.01
  29. | epoch   2 |  2400/ 2981 batches | lr 4.75 | ms/batch 34.61 | loss  5.58 | ppl   265.86
  30. | epoch   2 |  2600/ 2981 batches | lr 4.75 | ms/batch 34.62 | loss  5.60 | ppl   269.12
  31. | epoch   2 |  2800/ 2981 batches | lr 4.75 | ms/batch 34.63 | loss  5.51 | ppl   248.37
  32. -----------------------------------------------------------------------------------------
  33. | end of epoch   2 | time: 107.58s | valid loss  5.60 | valid ppl   270.75
  34. -----------------------------------------------------------------------------------------
  35. | epoch   3 |   200/ 2981 batches | lr 4.51 | ms/batch 34.80 | loss  5.55 | ppl   257.31
  36. | epoch   3 |   400/ 2981 batches | lr 4.51 | ms/batch 34.63 | loss  5.56 | ppl   259.12
  37. | epoch   3 |   600/ 2981 batches | lr 4.51 | ms/batch 34.62 | loss  5.36 | ppl   213.08
  38. | epoch   3 |   800/ 2981 batches | lr 4.51 | ms/batch 34.63 | loss  5.44 | ppl   229.59
  39. | epoch   3 |  1000/ 2981 batches | lr 4.51 | ms/batch 34.63 | loss  5.37 | ppl   215.90
  40. | epoch   3 |  1200/ 2981 batches | lr 4.51 | ms/batch 34.64 | loss  5.41 | ppl   223.49
  41. | epoch   3 |  1400/ 2981 batches | lr 4.51 | ms/batch 34.63 | loss  5.43 | ppl   228.08
  42. | epoch   3 |  1600/ 2981 batches | lr 4.51 | ms/batch 34.62 | loss  5.47 | ppl   238.36
  43. | epoch   3 |  1800/ 2981 batches | lr 4.51 | ms/batch 34.58 | loss  5.40 | ppl   222.43
  44. | epoch   3 |  2000/ 2981 batches | lr 4.51 | ms/batch 34.56 | loss  5.44 | ppl   229.30
  45. | epoch   3 |  2200/ 2981 batches | lr 4.51 | ms/batch 34.55 | loss  5.32 | ppl   204.63
  46. | epoch   3 |  2400/ 2981 batches | lr 4.51 | ms/batch 34.54 | loss  5.39 | ppl   220.17
  47. | epoch   3 |  2600/ 2981 batches | lr 4.51 | ms/batch 34.55 | loss  5.41 | ppl   223.92
  48. | epoch   3 |  2800/ 2981 batches | lr 4.51 | ms/batch 34.55 | loss  5.34 | ppl   209.22
  49. -----------------------------------------------------------------------------------------
  50. | end of epoch   3 | time: 107.47s | valid loss  5.54 | valid ppl   253.71
  51. -----------------------------------------------------------------------------------------

使用测试集评价模型

使用测试集来测试模型。

  1. test_loss = evaluate(best_model, test_data)
  2. print('=' * 89)
  3. print('| End of training | test loss {:5.2f} | test ppl {:8.2f}'.format(
  4.     test_loss, math.exp(test_loss)))
  5. print('=' * 89)

输出:

  1. =========================================================================================
  2. | End of training | test loss  5.43 | test ppl   229.27
  3. =========================================================================================

脚本的总运行时间： (5分钟38.763秒）

Download Python source code:transformer_tutorial.py

Download Jupyter notebook:transformer_tutorial.ipynb