基于Transformer语言模型：GPT-2

基于Transformer语言模型：GPT-2

  Transformer是Google在2017年提出的一种新型模型架构。它以自注意力机制取代传统的RNN和CNN对序列数据的建模，它在机器翻译、语言理解等任务上显示出强大的表示能力,目前已经成为自然语言处理领域的主流框架之一。主要特点有:

1. 完全基于注意力机制,没有循环神经网络或卷积神经网络。
2. 并行计算,不像RNN那样串行计算,更加高效。
3. 可处理远距离依赖,不受RNN的梯度消失问题影响。

  Transformer的基本结构包含N个相同的层(Layer)。每个层包含两个子层:

1. 多头自注意力机制(Multi-Head Attention):通过将给定向量投影到多个子空间,再在每个子空间中计算注意力,最后将结果拼接,实现对序列中每个位置的注意力计算。
2. 前馈神经网络(Feed Forward Neural Network): 通常包含两个线性变换与ReLU激活,实现位置编码的非线性映射。

模型结构

• $b$ : batch size
• $t$ : sequence length
• $c$ : embedding dims
• $N$ : vocabulary size

模型实现：Pytorch

• 简单实现，参考GPT-2

import os
import math
import time
import pandas as pd
from dataclasses import dataclass

import torch
import torch.nn as nn
from torch.nn import functional as F
from torch.utils.data import Dataset
from torch.utils.data.dataloader import DataLoader
from torch.utils.tensorboard import SummaryWriter

# 模型参数
@dataclass
class ModelConfig:
    vocab_size: int = None  
    n_embed : int = None
    n_hidden: int = None
    max_seq_length: int = None
    n_head: int = None
    n_layer: int = None

激活函数GLUEs

GELU(Gaussian Error Linear Units)是BERT作者在论文中提出的一种新型激活函数,广泛应用于GPT系列的模型中，函数的定义如下：
$$GELU(x)=xP(X<=x)=x\Phi (x)=x\frac{1}{2}[1+erf(x/\sqrt{2})]$$

• $\Phi (x)$ : $x$ 的累积分布函数

近似表示为：

$$GELU(x)\approx 0.5x(1+tanh[\sqrt{2/\pi }(x+0.044715x^3)])=x\sigma (1.702x)$$

class NewGELU(nn.Module):
    
    def forward(self, x):
        return 0.5 * x * (1.0 + torch.tanh(math.sqrt(2.0 / math.pi) * (x + 0.044715 * torch.pow(x, 3.0))))

因果自注意力

class CausalSelfAttention(nn.Module):
    """
    A multi-head masked self-attention layer with a projection at the end.
    It is possible to use torch.nn.MultiheadAttention here
    but I am including an explicit implementation here
    to show that there is nothing too scary here.
    """
    def __init__(self, config):
        super().__init__()
        assert config.n_embed % config.n_head == 0
        # key, query, value projections for all heads, but in a batch
        self.c_attn = nn.Linear(config.n_embed, 3 * config.n_embed)
        # output projection
        self.c_proj = nn.Linear(config.n_embed, config.n_embed)
        # causal mask to ensure that attention is only applied to the left in the input sequence
        self.register_buffer("bias", torch.tril(torch.ones(config.max_seq_length, config.max_seq_length))
                                     .view(1, 1, config.max_seq_length, config.max_seq_length))
        self.n_head = config.n_head
        self.n_embed = config.n_embed

    def forward(self, x):
        B, T, C = x.size() # batch size, sequence length, embedding dimensionality (n_embed)

        # calculate query, key, values for all heads in batch and move head forward to be the batch dim
        q, k ,v  = self.c_attn(x).split(self.n_embed, dim=2)
        k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
        q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)
        v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2) # (B, nh, T, hs)

        # causal self-attention; Self-attend: (B, nh, T, hs) x (B, nh, hs, T) -> (B, nh, T, T)
        att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
        att = att.masked_fill(self.bias[:,:,:T,:T] == 0, float('-inf'))
        att = F.softmax(att, dim=-1)
        y = att @ v # (B, nh, T, T) x (B, nh, T, hs) -> (B, nh, T, hs)
        y = y.transpose(1, 2).contiguous().view(B, T, C) # re-assemble all head outputs side by side

        # output projection
        y = self.c_proj(y)
        return y

Transformer Block

class Block(nn.Module):
    """ an unassuming Transformer block """

    def __init__(self, config):
        super().__init__()
        self.ln_1 = nn.LayerNorm(config.n_embed)
        self.attn = CausalSelfAttention(config)
        self.ln_2 = nn.LayerNorm(config.n_embed)
        self.mlp = nn.ModuleDict(dict(
            c_fc    = nn.Linear(config.n_embed, 4 * config.n_embed),
            c_proj  = nn.Linear(4 * config.n_embed, config.n_embed),
            act     = NewGELU(),
        ))
        m = self.mlp
        self.mlpf = lambda x: m.c_proj(m.act(m.c_fc(x))) # MLP forward

    def forward(self, x):
        x = x + self.attn(self.ln_1(x))
        x = x + self.mlpf(self.ln_2(x))
        return x

GPT Model

class Transformer(nn.Module):
    """ Transformer Language Model, exactly as seen in GPT-2 """

    def __init__(self, config):
        super().__init__()
        self.max_seq_length = config.max_seq_length

        self.transformer = nn.ModuleDict(dict(
            wte = nn.Embedding(config.vocab_size, config.n_embed),
            wpe = nn.Embedding(config.max_seq_length, config.n_embed),
            h = nn.ModuleList([Block(config) for _ in range(config.n_layer)]),
            ln_f = nn.LayerNorm(config.n_embed),
        ))
        self.lm_head = nn.Linear(config.n_embed, config.vocab_size, bias=False)

        # report number of parameters (note we don't count the decoder parameters in lm_head)
        n_params = sum(p.numel() for p in self.transformer.parameters())
        print("number of parameters: %.2fM" % (n_params/1e6,))

    def get_max_seq_length(self):
        return self.max_seq_length

    def forward(self, idx, targets=None):
        device = idx.device
        b, t = idx.size()
        assert t <= self.max_seq_length, f"Cannot forward sequence of length {t}, block size is only {self.max_seq_length}"
        pos = torch.arange(0, t, dtype=torch.long, device=device).unsqueeze(0) # shape (1, t)

        # forward the GPT model itself
        tok_emb = self.transformer.wte(idx) # token embeddings of shape (b, t, n_embed)
        pos_emb = self.transformer.wpe(pos) # position embeddings of shape (1, t, n_embed)
        x = tok_emb + pos_emb
        for block in self.transformer.h:
            x = block(x)
        x = self.transformer.ln_f(x)
        logits = self.lm_head(x)

        # if we are given some desired targets also calculate the loss
        loss = None
        if targets is not None:
            loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.view(-1), ignore_index=-1)

        return logits, loss

测试数据

数据集来10k+中文外卖评价数据集：

data = pd.read_csv('./dataset/waimai_10k.csv')
data.dropna(subset='review',inplace=True)
data['review_length'] = data.review.apply(lambda x:len(x))
data.sample(5)

语料统计信息：

data = data[data.review_length <=50] # 滤掉长度超过50的评论
words = data.review.tolist()
chars = sorted(list(set(''.join(words))))    
max_word_length = max(len(w) for w in words)

print(f"number of examples: {len(words)}")
print(f"max word length: {max_word_length}")
print(f"size of vocabulary: {len(chars)}")

number of examples: 10796
max word length: 50
size of vocabulary: 2272

划分训练/测试数据

test_set_size = min(1000, int(len(words) * 0.1)) 
rp = torch.randperm(len(words)).tolist()
train_words = [words[i] for i in rp[:-test_set_size]]
test_words = [words[i] for i in rp[-test_set_size:]]
print(f"split up the dataset into {len(train_words)} training examples and {len(test_words)} test examples")

split up the dataset into 9796 training examples and 1000 test examples

构造字符数据集[tensor]

• < BLANK> : 0
• token seqs : [1, 2, 3, 4, 5, 6]
• x : [0, 1, 2, 3, 4, 5, 6]
• y : [1, 2, 3, 4, 5, 6, 0]

class CharDataset(Dataset):

    def __init__(self, words, chars, max_word_length):
        self.words = words
        self.chars = chars
        self.max_word_length = max_word_length
        # char-->index-->char
        self.char2i = {ch:i+1 for i,ch in enumerate(chars)}
        self.i2char = {i:s for s,i in self.char2i.items()}    

    def __len__(self):
        return len(self.words)

    def contains(self, word):
        return word in self.words

    def get_vocab_size(self):
        return len(self.chars) + 1      

    def get_output_length(self):
        return self.max_word_length + 1

    def encode(self, word):
        # char sequece ---> index sequence
        ix = torch.tensor([self.char2i[w] for w in word], dtype=torch.long)
        return ix

    def decode(self, ix):
        # index sequence ---> char sequence
        word = ''.join(self.i2char[i] for i in ix)
        return word

    def __getitem__(self, idx):
        word = self.words[idx]
        ix = self.encode(word)
        x = torch.zeros(self.max_word_length + 1, dtype=torch.long)
        y = torch.zeros(self.max_word_length + 1, dtype=torch.long)
        x[1:1+len(ix)] = ix
        y[:len(ix)] = ix
        y[len(ix)+1:] = -1 # index -1 will mask the loss
        return x, y

数据加载器[DataLoader]

class InfiniteDataLoader:
    
    def __init__(self, dataset, **kwargs):
        train_sampler = torch.utils.data.RandomSampler(dataset, replacement=True, num_samples=int(1e10))
        self.train_loader = DataLoader(dataset, sampler=train_sampler, **kwargs)
        self.data_iter = iter(self.train_loader)

    def next(self):
        try:
            batch = next(self.data_iter)
        except StopIteration: # this will technically only happen after 1e10 samples... (i.e. basically never)
            self.data_iter = iter(self.train_loader)
            batch = next(self.data_iter)
        return batch

训练模型

# 模型评估
@torch.inference_mode()
def evaluate(model, dataset, batch_size=10, max_batches=None):
    model.eval()
    loader = DataLoader(dataset, shuffle=True, batch_size=batch_size, num_workers=0)
    losses = []
    for i, batch in enumerate(loader):
        batch = [t.to('cuda') for t in batch]
        X, Y = batch
        logits, loss = model(X, Y)
        losses.append(loss.item())
        if max_batches is not None and i >= max_batches:
            break
    mean_loss = torch.tensor(losses).mean().item()
    model.train() # reset model back to training mode
    return mean_loss

环境初始化:

torch.manual_seed(seed=12345)
torch.cuda.manual_seed_all(seed=12345)

work_dir = "./GPT2_log"
os.makedirs(work_dir, exist_ok=True)
writer = SummaryWriter(log_dir=work_dir)

模型初始化：

config = ModelConfig(vocab_size=len(chars)+1,
                     n_embed=128,
                     n_hidden=64,
                     max_seq_length=max_word_length+1,
                     n_head=4,
                     n_layer=4)

model = Transformer(config)

model.to('cuda')

number of parameters: 1.09M





Transformer(
  (transformer): ModuleDict(
    (wte): Embedding(2273, 128)
    (wpe): Embedding(51, 128)
    (h): ModuleList(
      (0-3): 4 x Block(
        (ln_1): LayerNorm((128,), eps=1e-05, elementwise_affine=True)
        (attn): CausalSelfAttention(
          (c_attn): Linear(in_features=128, out_features=384, bias=True)
          (c_proj): Linear(in_features=128, out_features=128, bias=True)
        )
        (ln_2): LayerNorm((128,), eps=1e-05, elementwise_affine=True)
        (mlp): ModuleDict(
          (c_fc): Linear(in_features=128, out_features=512, bias=True)
          (c_proj): Linear(in_features=512, out_features=128, bias=True)
          (act): NewGELU()
        )
      )
    )
    (ln_f): LayerNorm((128,), eps=1e-05, elementwise_affine=True)
  )
  (lm_head): Linear(in_features=128, out_features=2273, bias=False)
)

初始化数据：

train_dataset = CharDataset(train_words, chars, max_word_length)
test_dataset = CharDataset(test_words, chars, max_word_length)

train_dataset[0][0].shape, train_dataset[0][1].shape

(torch.Size([51]), torch.Size([51]))

Training:文章来源地址https://www.toymoban.com/news/detail-469140.html

# init optimizer
optimizer = torch.optim.AdamW(model.parameters(), lr=5e-4, weight_decay=0.01, betas=(0.9, 0.99), eps=1e-8)
# init dataloader
batch_loader = InfiniteDataLoader(train_dataset, batch_size=256, pin_memory=True, num_workers=4)

# training loop
best_loss = None
step = 0
train_losses, test_losses = [],[]
while True:

    t0 = time.time()

    # get the next batch, ship to device, and unpack it to input and target
    batch = batch_loader.next()
    batch = [t.to('cuda') for t in batch]
    X, Y = batch
    # feed into the model
    logits, loss = model(X, Y)

    # calculate the gradient, update the weights
    model.zero_grad(set_to_none=True)
    loss.backward()
    optimizer.step()
    # wait for all CUDA work on the GPU to finish then calculate iteration time taken
    torch.cuda.synchronize()
    t1 = time.time()

    # logging
    if step % 1000 == 0:
        print(f"step {step} | loss {loss.item():.4f} | step time {(t1-t0)*1000:.2f}ms")

    # evaluate the model
    if step > 0 and step % 100 == 0:
        train_loss = evaluate(model, train_dataset, batch_size=100, max_batches=10)
        test_loss  = evaluate(model, test_dataset,  batch_size=100, max_batches=10)
        train_losses.append(train_loss)
        test_losses.append(test_loss)
        # save the model to disk if it has improved
        if best_loss is None or test_loss < best_loss:
            out_path = os.path.join(work_dir, "model.pt")
            print(f"test loss {test_loss} is the best so far, saving model to {out_path}")
            torch.save(model.state_dict(), out_path)
            best_loss = test_loss

    step += 1
    # termination conditions
    if step > 10100:
        break

step 0 | loss 7.8996 | step time 424.90ms
test loss 4.594789028167725 is the best so far, saving model to ./GPT2_log/model.pt
test loss 3.9983901977539062 is the best so far, saving model to ./GPT2_log/model.pt
test loss 3.762165069580078 is the best so far, saving model to ./GPT2_log/model.pt
test loss 3.6443073749542236 is the best so far, saving model to ./GPT2_log/model.pt
test loss 3.5818755626678467 is the best so far, saving model to ./GPT2_log/model.pt
test loss 3.565037250518799 is the best so far, saving model to ./GPT2_log/model.pt
step 1000 | loss 2.4028 | step time 30.07ms
step 2000 | loss 1.1732 | step time 30.60ms
step 3000 | loss 0.7114 | step time 29.94ms
step 4000 | loss 0.5963 | step time 29.27ms
step 5000 | loss 0.5811 | step time 30.56ms
step 6000 | loss 0.5321 | step time 30.99ms
step 7000 | loss 0.5324 | step time 29.52ms
step 8000 | loss 0.5611 | step time 30.47ms
step 9000 | loss 0.5524 | step time 30.72ms
step 10000 | loss 0.5481 | step time 31.09ms

测试：外卖评价生成器

# laod save best model
model.load_state_dict(torch.load('./GPT2_log/model.pt'))

<All keys matched successfully>

@torch.no_grad()
def generate(model, idx, max_new_tokens, temperature=1.0, do_sample=False, top_k=None):
    for _ in range(max_new_tokens):
        # forward the model to get the logits for the index in the sequence
        logits, _ = model(idx)
        # pluck the logits at the final step and scale by desired temperature
        logits = logits[:,-1,:] / temperature
        # optionally crop the logits to only the top k options
        if top_k is not None:
            v, _ = torch.topk(logits, top_k)
            logits[logits < v[:, [-1]]] = -float('Inf')
        # apply softmax to convert logits to (normalized) probabilities
        probs = F.softmax(logits, dim=-1)
        # either sample from the distribution or take the most likely element
        if do_sample:
            idx_next = torch.multinomial(probs, num_samples=1)
        else:
            _, idx_next = torch.topk(probs, k=1, dim=-1)
         
        # append sampled index to the running sequence and continue
        idx = torch.cat((idx, idx_next), dim=-1)
    return idx

def print_samples(num=13):
    # inital 0 tokens
    X_init = torch.zeros((num, 1), dtype=torch.long).to('cuda')
    steps = train_dataset.get_output_length() - 1 # -1 because we already start with <START> token (index 0)
    X_samp = generate(model, X_init, steps, top_k=None, do_sample=True).to('cuda')
    new_samples = []
    for i in range(X_samp.size(0)):
        # get the i'th row of sampled integers, as python list
        row = X_samp[i, 1:].tolist() # note: we need to crop out the first <START> token
        # token 0 is the <END> token, so we crop the output sequence at that point
        crop_index = row.index(0) if 0 in row else len(row)
        row = row[:crop_index]
        word_samp = train_dataset.decode(row)
        new_samples.append(word_samp)
    return new_samples

print_samples(num=10)

['送餐很快，师傅辛苦了，味道非常好，服务挺好！',
 '还不错,有点辣',
 '花。不好吃，送餐单人是不长，肘子皮白粥包装好。很细心酱服务态度，很值！！！',
 '一如既往的神子还不错，真心不了',
 '师傅洒了快',
 '估汁买的太大了，我实在哪里的一道是80分，袖蹄放只有股卷饼腻。没夏怪卷饼，怎么好吃完成。',
 '一个半小时！现在外卖来了！太慢了已送到50多啊！好差辣椒腐柳，也不是虑就不怎么怀疑址，。',
 '忘餐厅到这次不放挺热的',
 '骑士态度很好，门给送餐员但饮料。这种纯）目少',
 '好吃好吃，味道也没有！']

到了这里，关于基于Transformer语言模型：GPT-2的文章就介绍完了。如果您还想了解更多内容，请在右上角搜索TOY模板网以前的文章或继续浏览下面的相关文章，希望大家以后多多支持TOY模板网！