15.10. 预训练BERT¶ 在 SageMaker Studio Lab 中打开 Notebook
我们在 第 15.8 节中实现了BERT模型，并在 第 15.9 节中从WikiText-2数据集中生成了预训练样本。本节将在此基础上，在WikiText-2数据集上对BERT进行预训练。
import torch
from torch import nn
from d2l import torch as d2l
from mxnet import autograd, gluon, init, np, npx
from d2l import mxnet as d2l
npx.set_np()
首先，我们加载WikiText-2数据集，将其作为用于遮蔽语言模型和下一句预测的小批量预训练样本。批量大小是512，BERT输入序列的最大长度是64。请注意，在最初的BERT模型中，最大长度是512。
batch_size, max_len = 512, 64
train_iter, vocab = d2l.load_data_wiki(batch_size, max_len)
batch_size, max_len = 512, 64
train_iter, vocab = d2l.load_data_wiki(batch_size, max_len)
[22:11:29] ../src/storage/storage.cc:196: Using Pooled (Naive) StorageManager for CPU
15.10.1. 预训练BERT¶
最初的BERT模型有两个不同模型大小的版本 (Devlin et al., 2018)。基本模型（\(\textrm{BERT}_{\textrm{BASE}}\)）使用了12个层（Transformer编码器块），768个隐藏单元（隐藏大小）和12个自注意力头。大模型（\(\textrm{BERT}_{\textrm{LARGE}}\)）使用了24个层，1024个隐藏单元和16个自注意力头。值得注意的是，前者有1.1亿个参数，而后者有3.4亿个参数。为了便于演示，我们定义了一个小的BERT，使用了2个层、128个隐藏单元和2个自注意力头。
net = d2l.BERTModel(len(vocab), num_hiddens=128,
ffn_num_hiddens=256, num_heads=2, num_blks=2, dropout=0.2)
devices = d2l.try_all_gpus()
loss = nn.CrossEntropyLoss()
net = d2l.BERTModel(len(vocab), num_hiddens=128, ffn_num_hiddens=256,
num_heads=2, num_blks=2, dropout=0.2)
devices = d2l.try_all_gpus()
net.initialize(init.Xavier(), ctx=devices)
loss = gluon.loss.SoftmaxCELoss()
[22:12:33] ../src/storage/storage.cc:196: Using Pooled (Naive) StorageManager for GPU
[22:12:34] ../src/storage/storage.cc:196: Using Pooled (Naive) StorageManager for GPU
在定义训练循环之前，我们先定义一个辅助函数 _get_batch_loss_bert。给定训练样本的一个切片，该函数计算遮蔽语言模型和下一句预测任务的损失。请注意，BERT预训练的最终损失只是遮蔽语言模型损失和下一句预测损失的和。
#@save
def _get_batch_loss_bert(net, loss, vocab_size, tokens_X,
segments_X, valid_lens_x,
pred_positions_X, mlm_weights_X,
mlm_Y, nsp_y):
# Forward pass
_, mlm_Y_hat, nsp_Y_hat = net(tokens_X, segments_X,
valid_lens_x.reshape(-1),
pred_positions_X)
# Compute masked language model loss
mlm_l = loss(mlm_Y_hat.reshape(-1, vocab_size), mlm_Y.reshape(-1)) *\
mlm_weights_X.reshape(-1, 1)
mlm_l = mlm_l.sum() / (mlm_weights_X.sum() + 1e-8)
# Compute next sentence prediction loss
nsp_l = loss(nsp_Y_hat, nsp_y)
l = mlm_l + nsp_l
return mlm_l, nsp_l, l
#@save
def _get_batch_loss_bert(net, loss, vocab_size, tokens_X_shards,
segments_X_shards, valid_lens_x_shards,
pred_positions_X_shards, mlm_weights_X_shards,
mlm_Y_shards, nsp_y_shards):
mlm_ls, nsp_ls, ls = [], [], []
for (tokens_X_shard, segments_X_shard, valid_lens_x_shard,
pred_positions_X_shard, mlm_weights_X_shard, mlm_Y_shard,
nsp_y_shard) in zip(
tokens_X_shards, segments_X_shards, valid_lens_x_shards,
pred_positions_X_shards, mlm_weights_X_shards, mlm_Y_shards,
nsp_y_shards):
# Forward pass
_, mlm_Y_hat, nsp_Y_hat = net(
tokens_X_shard, segments_X_shard, valid_lens_x_shard.reshape(-1),
pred_positions_X_shard)
# Compute masked language model loss
mlm_l = loss(
mlm_Y_hat.reshape((-1, vocab_size)), mlm_Y_shard.reshape(-1),
mlm_weights_X_shard.reshape((-1, 1)))
mlm_l = mlm_l.sum() / (mlm_weights_X_shard.sum() + 1e-8)
# Compute next sentence prediction loss
nsp_l = loss(nsp_Y_hat, nsp_y_shard)
nsp_l = nsp_l.mean()
mlm_ls.append(mlm_l)
nsp_ls.append(nsp_l)
ls.append(mlm_l + nsp_l)
npx.waitall()
return mlm_ls, nsp_ls, ls
通过调用上述两个辅助函数，下面的 train_bert 函数定义了在WikiText-2（train_iter）数据集上预训练BERT（net）的过程。训练BERT可能需要很长时间。与在 train_ch13 函数（见 第 14.1 节）中指定训练的迭代轮数不同，下面函数的输入 num_steps 指定了训练的迭代步数。
def train_bert(train_iter, net, loss, vocab_size, devices, num_steps):
net(*next(iter(train_iter))[:4])
net = nn.DataParallel(net, device_ids=devices).to(devices[0])
trainer = torch.optim.Adam(net.parameters(), lr=0.01)
step, timer = 0, d2l.Timer()
animator = d2l.Animator(xlabel='step', ylabel='loss',
xlim=[1, num_steps], legend=['mlm', 'nsp'])
# Sum of masked language modeling losses, sum of next sentence prediction
# losses, no. of sentence pairs, count
metric = d2l.Accumulator(4)
num_steps_reached = False
while step < num_steps and not num_steps_reached:
for tokens_X, segments_X, valid_lens_x, pred_positions_X,\
mlm_weights_X, mlm_Y, nsp_y in train_iter:
tokens_X = tokens_X.to(devices[0])
segments_X = segments_X.to(devices[0])
valid_lens_x = valid_lens_x.to(devices[0])
pred_positions_X = pred_positions_X.to(devices[0])
mlm_weights_X = mlm_weights_X.to(devices[0])
mlm_Y, nsp_y = mlm_Y.to(devices[0]), nsp_y.to(devices[0])
trainer.zero_grad()
timer.start()
mlm_l, nsp_l, l = _get_batch_loss_bert(
net, loss, vocab_size, tokens_X, segments_X, valid_lens_x,
pred_positions_X, mlm_weights_X, mlm_Y, nsp_y)
l.backward()
trainer.step()
metric.add(mlm_l, nsp_l, tokens_X.shape[0], 1)
timer.stop()
animator.add(step + 1,
(metric[0] / metric[3], metric[1] / metric[3]))
step += 1
if step == num_steps:
num_steps_reached = True
break
print(f'MLM loss {metric[0] / metric[3]:.3f}, '
f'NSP loss {metric[1] / metric[3]:.3f}')
print(f'{metric[2] / timer.sum():.1f} sentence pairs/sec on '
f'{str(devices)}')
def train_bert(train_iter, net, loss, vocab_size, devices, num_steps):
trainer = gluon.Trainer(net.collect_params(), 'adam',
{'learning_rate': 0.01})
step, timer = 0, d2l.Timer()
animator = d2l.Animator(xlabel='step', ylabel='loss',
xlim=[1, num_steps], legend=['mlm', 'nsp'])
# Sum of masked language modeling losses, sum of next sentence prediction
# losses, no. of sentence pairs, count
metric = d2l.Accumulator(4)
num_steps_reached = False
while step < num_steps and not num_steps_reached:
for batch in train_iter:
(tokens_X_shards, segments_X_shards, valid_lens_x_shards,
pred_positions_X_shards, mlm_weights_X_shards,
mlm_Y_shards, nsp_y_shards) = [gluon.utils.split_and_load(
elem, devices, even_split=False) for elem in batch]
timer.start()
with autograd.record():
mlm_ls, nsp_ls, ls = _get_batch_loss_bert(
net, loss, vocab_size, tokens_X_shards, segments_X_shards,
valid_lens_x_shards, pred_positions_X_shards,
mlm_weights_X_shards, mlm_Y_shards, nsp_y_shards)
for l in ls:
l.backward()
trainer.step(1)
mlm_l_mean = sum([float(l) for l in mlm_ls]) / len(mlm_ls)
nsp_l_mean = sum([float(l) for l in nsp_ls]) / len(nsp_ls)
metric.add(mlm_l_mean, nsp_l_mean, batch[0].shape[0], 1)
timer.stop()
animator.add(step + 1,
(metric[0] / metric[3], metric[1] / metric[3]))
step += 1
if step == num_steps:
num_steps_reached = True
break
print(f'MLM loss {metric[0] / metric[3]:.3f}, '
f'NSP loss {metric[1] / metric[3]:.3f}')
print(f'{metric[2] / timer.sum():.1f} sentence pairs/sec on '
f'{str(devices)}')
我们可以绘制出在BERT预训练过程中的遮蔽语言模型损失和下一句预测损失。
train_bert(train_iter, net, loss, len(vocab), devices, 50)
MLM loss 5.885, NSP loss 0.760
4413.2 sentence pairs/sec on [device(type='cuda', index=0), device(type='cuda', index=1)]
train_bert(train_iter, net, loss, len(vocab), devices, 50)
MLM loss 7.292, NSP loss 0.822
2417.3 sentence pairs/sec on [gpu(0), gpu(1)]
15.10.2. 用BERT表示文本¶
在预训练BERT之后，我们可以用它来表示单个文本、文本对或其中的任意词元。以下函数返回 tokens_a 和 tokens_b 中所有词元的BERT（net）表示。
def get_bert_encoding(net, tokens_a, tokens_b=None):
tokens, segments = d2l.get_tokens_and_segments(tokens_a, tokens_b)
token_ids = torch.tensor(vocab[tokens], device=devices[0]).unsqueeze(0)
segments = torch.tensor(segments, device=devices[0]).unsqueeze(0)
valid_len = torch.tensor(len(tokens), device=devices[0]).unsqueeze(0)
encoded_X, _, _ = net(token_ids, segments, valid_len)
return encoded_X
def get_bert_encoding(net, tokens_a, tokens_b=None):
tokens, segments = d2l.get_tokens_and_segments(tokens_a, tokens_b)
token_ids = np.expand_dims(np.array(vocab[tokens], ctx=devices[0]),
axis=0)
segments = np.expand_dims(np.array(segments, ctx=devices[0]), axis=0)
valid_len = np.expand_dims(np.array(len(tokens), ctx=devices[0]), axis=0)
encoded_X, _, _ = net(token_ids, segments, valid_len)
return encoded_X
考虑句子“a crane is flying”。回想一下在 第 15.8.4 节中讨论的BERT输入表示。在插入特殊词元“<cls>”（用于分类）和“<sep>”（用于分隔）后，BERT输入序列的长度为6。由于0是“<cls>”词元的索引，encoded_text[:, 0, :] 是整个输入句子的BERT表示。为了评估多义词“crane”，我们还打印出了该词元的BERT表示的前三个元素。
tokens_a = ['a', 'crane', 'is', 'flying']
encoded_text = get_bert_encoding(net, tokens_a)
# Tokens: '<cls>', 'a', 'crane', 'is', 'flying', '<sep>'
encoded_text_cls = encoded_text[:, 0, :]
encoded_text_crane = encoded_text[:, 2, :]
encoded_text.shape, encoded_text_cls.shape, encoded_text_crane[0][:3]
(torch.Size([1, 6, 128]),
torch.Size([1, 128]),
tensor([0.8414, 1.4830, 0.8226], device='cuda:0', grad_fn=<SliceBackward0>))
tokens_a = ['a', 'crane', 'is', 'flying']
encoded_text = get_bert_encoding(net, tokens_a)
# Tokens: '<cls>', 'a', 'crane', 'is', 'flying', '<sep>'
encoded_text_cls = encoded_text[:, 0, :]
encoded_text_crane = encoded_text[:, 2, :]
encoded_text.shape, encoded_text_cls.shape, encoded_text_crane[0][:3]
((1, 6, 128),
(1, 128),
array([-1.2760178, -0.79205 , -1.0534445], ctx=gpu(0)))
现在考虑一个句子对“a crane driver came”和“he just left”。同样，encoded_pair[:, 0, :] 是来自预训练BERT的整个句子对的编码结果。注意，多义词“crane”的前三个元素与上下文不同时的前三个元素是不同的。这表明BERT表示是上下文敏感的。
tokens_a, tokens_b = ['a', 'crane', 'driver', 'came'], ['he', 'just', 'left']
encoded_pair = get_bert_encoding(net, tokens_a, tokens_b)
# Tokens: '<cls>', 'a', 'crane', 'driver', 'came', '<sep>', 'he', 'just',
# 'left', '<sep>'
encoded_pair_cls = encoded_pair[:, 0, :]
encoded_pair_crane = encoded_pair[:, 2, :]
encoded_pair.shape, encoded_pair_cls.shape, encoded_pair_crane[0][:3]
(torch.Size([1, 10, 128]),
torch.Size([1, 128]),
tensor([0.0430, 1.6132, 0.0437], device='cuda:0', grad_fn=<SliceBackward0>))
tokens_a, tokens_b = ['a', 'crane', 'driver', 'came'], ['he', 'just', 'left']
encoded_pair = get_bert_encoding(net, tokens_a, tokens_b)
# Tokens: '<cls>', 'a', 'crane', 'driver', 'came', '<sep>', 'he', 'just',
# 'left', '<sep>'
encoded_pair_cls = encoded_pair[:, 0, :]
encoded_pair_crane = encoded_pair[:, 2, :]
encoded_pair.shape, encoded_pair_cls.shape, encoded_pair_crane[0][:3]
((1, 10, 128),
(1, 128),
array([-1.2759778 , -0.79211384, -1.0534613 ], ctx=gpu(0)))
在 第 16 章中，我们将微调预训练的BERT模型，用于下游的自然语言处理应用。
15.10.3. 小结¶
最初的BERT有两个版本，其中基本模型有1.1亿个参数，大模型有3.4亿个参数。
在预训练BERT之后，我们可以用它来表示单个文本、文本对或其中的任意词元。
在实验中，当上下文不同时，同一个词元有不同的BERT表示。这表明BERT表示是上下文敏感的。