参考：跟着论文《 Attention is All You Need》一步一步实现Attention和Transformer
对上面博客中提供的代码的一些细节进行注释。
由于是以机器翻译作为例子。对于没有接触过这方面的，特别是做视觉的会有很多细节不能理解，我花了一些时间，看了torchtext的使用以及机器翻译的过程，给代码做了写注释。
torchtext的使用：参考1，参考2，torchtext文档等等
代码分成两部分，一部分是NMT的部分，另一部分是模型
import numpy as np
import torch
import torch.nn as nn
import time
from torch.autograd import Variable
import matplotlib.pyplot as plt
import seaborn
seaborn.set_context(context="talk")
#%matplotlib inline
from torchtext import data, datasets
from model import *
#用于mask数据,产生source mask和target mask
class Batch:
""" 在训练期间使用mask处理数据 """
def __init__(self, src, trg=None, pad=0):
#src.size = batch_size, q_len
self.src = src
#src_mask.size = batch_size, 1, q_len
self.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()
@staticmethod
def make_std_mask(tgt, pad):
""" 创造一个mask来屏蔽补全词和字典外的词进行屏蔽"""
tgt_mask = (tgt != pad).unsqueeze(-2)
tgt_mask = tgt_mask & Variable(subsequent_mask(tgt.size(-1)).type_as(tgt_mask.data))
return tgt_mask
#将优化器再包一层，更方便
class NoamOpt:
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):
""" 更新参数和学习率 """
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):
""" lrate 实现"""
if step is None:
step = self._step
return self.factor * (self.model_size ** (-0.5) * min(step ** (-0.5), step * self.warmup ** (-1.5)))
#没用到，就是返回一个优化器，里面是一些设置
def get_std_up(model):
return NoamOpt(model.src_embed[0].d_model, 2, 4000,
torch.optim.Adam(model.param_groups(),
lr=0, betas=(0.9, 0.98), eps=1e-9))
'''size 是目标类别数目 smoothing这里使用,0.1'''
#平滑标签，将非真实目标的类别也给一个小的值
class LabelSmoothing(nn.Module):
""" 标签平滑实现 """
def __init__(self, size, padding_idx, smoothing=0.0):
super(LabelSmoothing, self).__init__()
#改成这样，不会有warning
self.criterion = nn.KLDivLoss(reduction='none')
self.padding_idx = padding_idx
self.confidence = 1.0 - smoothing
self.smoothing = smoothing
self.size = size
self.true_dist = None
#x是generator的输出[n, vocab_size]，也就是模型预测，target是真实目标,大小 n
def forward(self, x, target):
assert x.size(1) == self.size
true_dist = x.data.clone()
#为什么减去2???? 要减去padding_idx和正确的label本身
#size(x) = batch_size,
true_dist.fill_(self.smoothing / (self.size - 2))
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) #dim,index,val
self.true_dist = true_dist
return self.criterion(x, Variable(true_dist, requires_grad=False))
class MultiGPULossCompute:
"A multi-gpu loss compute and train function."
def __init__(self, generator, criterion, devices, opt=None, chunk_size=5):
# Send out to different gpus.
self.generator = generator
self.criterion = nn.parallel.replicate(criterion,
devices=devices)
self.opt = opt
self.devices = devices
self.chunk_size = chunk_size
#size(out) = batch_size, max_len, d_model
def __call__(self, out, targets, normalize):
total = 0.0
generator = nn.parallel.replicate(self.generator,
devices=self.devices)
out_scatter = nn.parallel.scatter(out,
target_gpus=self.devices)
out_grad = [[] for _ in out_scatter]
targets = nn.parallel.scatter(targets,
target_gpus=self.devices)
# Divide generating into chunks.
chunk_size = self.chunk_size
for i in range(0, out_scatter[0].size(1), chunk_size):
# Predict distributions
out_column = [[Variable(o[:, i:i + chunk_size].data,
requires_grad=self.opt is not None)]
for o in out_scatter]
gen = nn.parallel.parallel_apply(generator[:len(out_column)], out_column, )
# Compute loss.
y = [(g.contiguous().view(-1, g.size(-1)),
t[:, i:i + chunk_size].contiguous().view(-1))
for g, t in zip(gen, targets)]
loss = nn.parallel.parallel_apply(self.criterion[:len(y)], y)
# Sum and normalize loss
l = nn.parallel.gather(loss,
target_device=self.devices[0],dim=0)
l = l.sum() / normalize
total += l.data.item()
#因为上面对数据进行分割，定义了一个新的out_column，所以梯度传到这里就不会往前传了，需要手动计算梯度，再从out往前传播
# Backprop loss to output of transformer
if self.opt is not None:
l.backward() #累积
# sh = out_column[0].detach().cpu()
# input('here im !!!!!!!!!!!')
for j, l in enumerate(loss):#每个只有一项所以下标是0
out_grad[j].append(out_column[j][0].grad.data.clone())
# Backprop all loss through transformer.
if self.opt is not None:
#把位于同一gpu的部分cat起来（每个对应不同的chunk）
out_grad = [Variable(torch.cat(og, dim=1)) for og in out_grad]
o1 = out
o2 = nn.parallel.gather(out_grad,target_device=self.devices[0])
o1.backward(gradient=o2)
self.opt.step()
self.opt.optimizer.zero_grad()
return total * normalize
#训练时相类似BucketIterator 的作用，测试时则则按正常顺序
#pool 起到预先读取100个batch的作用，将他们排序
#每个batch都对长度进行排序
class MyIterator(data.Iterator):
def create_batches(self):
if self.train:
def pool(d, random_shuffler):
for p in data.batch(d, self.batch_size * 100):
p_batch = data.batch(
sorted(p, key=self.sort_key),
self.batch_size, self.batch_size_fn)
for b in random_shuffler(list(p_batch)): #about 100 times
#b为一个batch，list类型，它的元素是Example object，也就是一个训练样本
yield b
self.batches = pool(self.data(), self.random_shuffler)
else:
self.batches = []
for b in data.batch(self.data(), self.batch_size,
self.batch_size_fn):
self.batches.append(sorted(b, key=self.sort_key))
#src_mask mask掉那些padding
#贪心解码，
def greedy_decode(model, src, src_mask, max_len, start_symbol) :
memory = model.encode(src, src_mask)
ys = torch.ones(1, 1).fill_(start_symbol).type_as(src.data)
#根据当前的ys和src进行解码
for i in range(max_len - 1):
out = model.decode(memory, src_mask,
torch.Tensor(ys),
torch.Tensor(subsequent_mask(ys.size(1)).type_as(src.data)))
#根据输出,用generator转换成各个词的概率（一个线性层和softmax）
prob = model.generator(out[:, -1])
_, next_word = torch.max(prob, dim = 1)
next_word = next_word.data[0]
ys = torch.cat([ys, torch.ones(1, 1).type_as(src.data).fill_(next_word)], dim=1)
return ys
global max_src_in_batch, max_tgt_in_batch
#这个函数是为了使用动态batch设置的。batch大小根据迭代器设置的batch大小和当前已经加进来的样本长度得到
#在这里就是相当于设置的batch_size为每个batch 占用单位空间上限
#计算需要的总空间，new指的是当前batch新样本，count指的是，这个样本是当前batch的第几个
#(new example to add, current count of examples in the batch, and current effective batch size)
#returns the new effective batch size resulting from adding that example to a batch
def batch_size_fn(new, count, sofar):
""" 保持数据批量增加，并计算tokens+padding的总数 """
global max_src_in_batch, max_tgt_in_batch
if count == 1:
max_src_in_batch = 0
max_tgt_in_batch = 0
#new.src是当前训练样本的一个句子
max_src_in_batch = max(max_src_in_batch, len(new.src))
max_tgt_in_batch = max(max_tgt_in_batch, len(new.trg) + 2) # 2 表示的是前后共2个标志?
src_elements = count * max_src_in_batch
tgt_elements = count * max_tgt_in_batch
return max(src_elements, tgt_elements)
def run_epoch(data_iter, pad_idx, model, loss_compute):
""" 标准训练和日志函数 """
start = time.time()
total_tokens = 0
total_loss = 0
tokens = 0
for i, batch in enumerate(data_iter):
batch = Batch(batch.src,batch.trg,pad_idx)
out = model.forward(batch.src, batch.trg, batch.src_mask, batch.trg_mask)
#size(out) = [batch_size, q_len, d_model]
loss = loss_compute(out, batch.trg_y, batch.ntokens)
total_loss += loss
total_tokens += batch.ntokens
tokens += batch.ntokens
if i % 50 == 1:
elapsed = time.time() - start
print("Epoch Step: %d Loss : %f Tokens per Sec: %f " % (i, loss/ batch.ntokens, tokens / elapsed))
start = time.time()
tokens = 0
return total_loss / total_tokens
def main():
'''读取数据集'''
if True:
#spacy 用来做分词
import spacy
spacy_de = spacy.load('de')
spacy_en = spacy.load('en')
def tokenize_de(text):
return [tok.text for tok in spacy_de.tokenizer(text)]
def tokenize_en(text):
return [tok.text for tok in spacy_en.tokenizer(text)]
#起始标志，终止标志和填充词 begin of sentence / end of sentence
BOS_WORD = '<s>'
EOS_WORD = '</S>'
BLANK_WORD = "<blank>"
#加上batch first就不用后面转置一下了，原来的没加
SRC = data.Field(tokenize=tokenize_en, init_token=BOS_WORD, eos_token=EOS_WORD, pad_token=BLANK_WORD, batch_first=True)
TGT = data.Field(tokenize=tokenize_de, pad_token=BLANK_WORD, batch_first = True)
#长于MAX_LEN的丢掉
MAX_LEN = 220
#得到三个dataset
train, val, test = datasets.IWSLT.splits(
exts=('.en', '.de'), fields=(SRC, TGT),
filter_pred=lambda x: len(vars(x)['src']) <= MAX_LEN and len(vars(x)['trg']) <= MAX_LEN)
#词出现频率小于MIN_FREQ的丢掉
MIN_FREQ = 2
MIN_FREQ = 2
SRC.build_vocab(train.src, min_freq=MIN_FREQ)
TGT.build_vocab(train.trg, min_freq=MIN_FREQ)
# 需要使用的GPU
device_ids = [0,1] # 如果只有一个GPU，使用devices=[0]
device = torch.device(0)
'''构建模型'''
if True:
#pad_idx 一般好像都是0
pad_idx = TGT.vocab.stoi["<blank>"]
#模型在前面，这里先不管,N是encoder和decoder的层数
model = make_model(len(SRC.vocab), len(TGT.vocab), N=6)
model.cuda()
#标签平滑，这个也先不管
criterion = LabelSmoothing(size=len(TGT.vocab), padding_idx=pad_idx, smoothing=0.1)
criterion.cuda()
#这个batch_size不是句子数，输入占用空间大小的数目，见batch_size_fn这个函数
BATCH_SIZE = 1200
#自定义Iterator。 repeat应该是表示同一个迭代顺序要不要repeat多个epoch
train_iter = MyIterator(train, batch_size=BATCH_SIZE, device=device, repeat=False,
sort_key=lambda x : (len(x.src), len(x.trg)),
batch_size_fn=batch_size_fn, train=True)
valid_iter = MyIterator(val, batch_size=BATCH_SIZE, device= device, repeat=False,
sort_key=lambda x : (len(x.src), len(x.trg)),
batch_size_fn=batch_size_fn, train=False)
model_par = nn.DataParallel(model, device_ids=device_ids)
# 这里需要很大的内存，报内存错误很正常，可以直接用下面训练好的
# 或者调小BATCH_SIZE
'''开始训练'''
if True:
#优化器包装
model_opt = NoamOpt(model.src_embed[0].d_model, 1, 2000,
torch.optim.Adam(model.parameters(), lr=0, betas=(0.9, 0.98), eps=1e-9))
for epoch in range(10):
model_par.train()
run_epoch(train_iter, pad_idx, model_par,
MultiGPULossCompute(model.generator, criterion, devices=device_ids, opt=model_opt))
model_par.eval()
loss = run_epoch(valid_iter,pad_idx, model_par,
MultiGPULossCompute(model.generator, criterion, devices=device_ids, opt=None))
print("loss is: %f" % loss)
else: #load 已保存的模型
model = torch.load("iwslt.pt")
for i, batch in enumerate(valid_iter):
#取一条句子，大小为seq_len
src = batch.src[0]
#size(src_mask) = 1, 1, seq_len????
src_mask = (src != SRC.vocab.stoi["<blank>"]).unsqueeze(-2)
print('src_mask_size=',src_mask.size())
#out 是每个词在词典中的位置，还要转换成目标语单词
out = greedy_decode(model, src, src_mask, max_len=60, start_symbol=TGT.vocab.stoi["<s>"])
print("Translation: ", end="\t")
#输出模型的翻译
for i in range(1, out.size(1)):
sym = TGT.vocab.itos[out[0, i]]
if sym == "</s>":
print('meet end------------------')
break
print(sym, end=" ")
print()
#输出真实的目标 ground true
print("Target:", end="\t")
for i in range(1, batch.trg.size(0)):
sym = TGT.vocab.itos[batch.trg.data[i, 0]]
if sym == '</s>':
break
print(sym, end=" ")
print()
break
if __name__ == '__main__':
main()
import torch
from torch import nn
import torch.nn.functional as F
import numpy as np
import math, copy
#封装整个encoder和decoder
class EncoderDecoder(nn.Module):
"""
A stanard Encoder-Decoder architecture.Base fro this and many other models.
"""
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
self.tgt_embed = tgt_embed
self.generator = generator
def 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)
#一个分类层，把d_model转换成对应每个word的概率
#因为用的是KLDivLoss，所以这里输出log_softmax，把KLDivLoss改成CrossEntropyLoss，这里就直接输出logits即可
class Generator(nn.Module):
"""Define standard linear + softmax generation step."""
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)
#用于复制N个 module
def clones(module, N):
"Produce N identical layers."
return nn.ModuleList([copy.deepcopy(module) for _ in range(N)])
#将N个enconder layer封装起来
class Encoder(nn.Module):
"Core encoder is a stack of N layers"
def __init__(self, layer, N):
super(Encoder, self).__init__()
self.layers = clones(layer, N)
self.norm = LayerNorm(layer.size)
def forward(self, x, mask):
"Pass the input (and mask) through each layer in turn."
for layer in self.layers:
x = layer(x, mask)
return self.norm(x)
#layer norm, pytorch里面已经有了
class LayerNorm(nn.Module):
""" Construct a layernorm model (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 = eps
def 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
#layernorm + sublayer + residual
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 sanme size. """
return x + self.dropout(sublayer(self.norm(x)))
#encoder层 attention + poitwise_feedword层
class EncoderLayer(nn.Module):
"""Encoder is made up of self-attention and feed forward (defined below)"""
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):
"""Follow Figure 1 (left) for connection """
x = self.sublayer[0](x, lambda x: self.self_attn(x, x, x, mask))
return self.sublayer[1](x, self.feed_forward)
#decoder封装decoder层，memory是encoder的输出
#图的右边部分
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)
#两个attention(layernorm + resisual connection) + poitwise_feedward
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 = 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):
"""Follow Figure 1 (right) for connections"""
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)
'''
mask 例子 shape = (1,5,5)
[[[1 0 0 0 0 0],
[1 1 0 0 0 0],
[1 1 1 0 0 0],
[1 1 1 1 0 0],
[1 1 1 1 1 0],
]]
'''
def subsequent_mask(size):
"""Mask out subsequent positions. """
attn_shape = (1, size, size)
#k=1，对角线也是0
subsequent_mask = np.triu(np.ones(attn_shape), k=1).astype('uint8')
return torch.from_numpy(subsequent_mask) == 0
#用在multi-Head attention中
#Scaled Dot-Product Attention
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) # matmul矩阵相乘
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
#head数目h要整除d_model
class MultiHeadedAttention(nn.Module):
def __init__(self, h, d_model, dropout=0.1):
""" Take in model size and numbe of heads """
super(MultiHeadedAttention, self).__init__()7
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):
"""图片ModalNet-20的实现"""
if mask is not None:
# 同样的mask应用到所有heads
mask = mask.unsqueeze(1)
nbatches = query.size(0)
# 1. 批量做linear投影 => h x d_k
# query, key, value分别经过一h个线性变换（整合成一个）
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. 批量应用attention机制在所有的投影向量上
#attn 没有用到
x, self.attn = attention(query, key, value, mask=mask, dropout=self.dropout)
# 3. 使用view进行“Concat”并且进行最后一层的linear
x = x.transpose(1, 2).contiguous().view(nbatches, -1, self.h * self.d_k)
return self.linears[-1](x)
#除了Attention子层之外，Encoder和Decoder中的每个层都包含一个全连接前馈网络，
# 分别地应用于每个位置（每个word）。其中包括两个线性变换，然后使用ReLU作为激活函数。相当于两层1*1卷积，每个位置的特征就是对应一个channel,1
class PositionwiseFeedForward(nn.Module):
"""
FFN实现
d_model = 512
d_ff = 2048
"""
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))))
class Embeddings(nn.Module):
def __init__(self, d_model, vocab):
super(Embeddings, self).__init__()
self.lut = nn.Embedding(vocab, d_model) #look up matrix
self.d_model = d_model
def forward(self, x):
return self.lut(x) * math.sqrt(self.d_model)
class PositionalEncoding(nn.Module):
"""PE函数实现"""
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).float()
div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-(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)
pe.requires_grad = False
self.register_buffer('pe', pe)
def forward(self, x):
#print(x.type(),self.pe.type())
x = x + self.pe[:, :x.size(1)]
return self.dropout(x)
def make_model(src_vacab, tgt_vocab, N=6, d_model=512, d_ff=2048, h=8, dropout=0.1):
""" 构建模型"""
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_vacab), c(position)),
nn.Sequential(Embeddings(d_model, tgt_vocab), c(position)),
Generator(d_model, tgt_vocab)
)
# !!!import for the work
# 使用Glorot/ fan_avg初始化参数
for p in model.parameters():
if p.dim() > 1:
nn.init.xavier_uniform_(p)
return model
![Attention机制通俗介绍[图]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=)