PyTorch_Transformer/train.py at main · N-Thander/PyTorch_Transformer · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
import torch
import torch.nn as nn
from torch.utils.data import Dataset, DataLoader, random_split

from dataset import BilingualDataset, casual_mask
from config import get_weights_file_path, get_config

from model import build_transformer

from datasets import load_dataset

from tokenizers import Tokenizer
from tokenizers.models import WordLevel
from tokenizers.trainers import WordLevelTrainer
from tokenizers.pre_tokenizers import Whitespace

from torch.utils.tensorboard import SummaryWriter

from tqdm import tqdm

import warnings
from pathlib import Path

def get_all_sentences(ds, lang):
    for item in ds:
        yield item['translation'][lang]

def get_or_build_tokenizer(config, ds, lang):
    tokenizer_path = Path(config['tokenizer_file']).format('lang')
    if not Path.exists(tokenizer_path):
        tokenizer = Tokenizer(WordLevel(unk_token="[UNK]"))
        tokenizer.pre_tokenizer = Whitespace()
        # UNK -> unknown token
        # PAD -> padding token
        # SOS -> start of sentence token
        # EOS -> end of sentence token
        trainer = WordLevelTrainer(special_tokens=["[UNK]", "[PAD]", "[SOS]", "[EOS]"], min_frequency=2)
        tokenizer.train_from_iterator(get_all_sentences(ds, lang), trainer=trainer)
        tokenizer.save(str(tokenizer_path))
    else:
        tokenizer = Tokenizer.from_file(str(tokenizer_path))

    return tokenizer

def get_ds(config):
    ds_raw = load_dataset("opus_books", f"{config['lang_src']}-{config['lang_tgt']}", split="train")

    # build tokenizer
    tokenizer_src = get_or_build_tokenizer(config, ds_raw, config['lang_src'])
    tokenizer_tgt = get_or_build_tokenizer(config, ds_raw, config['lang_tgt'])

    # keeping 90% for training and 10% for validation
    train_ds_size = int(len(ds_raw) * 0.9)
    val_ds_size = len(ds_raw) - train_ds_size
    train_ds_raw, val_ds_raw = random_split(ds_raw, [train_ds_size, val_ds_size])

    train_ds = BilingualDataset(train_ds_raw, tokenizer_src, tokenizer_tgt, config['lang_src'], config['lang_tgt'], config['seq_len'])
    val_ds = BilingualDataset(val_ds_raw, tokenizer_src, tokenizer_tgt, config['lang_src'], config['lang_tgt'], config['seq_len'])

    max_len_src = 0
    max_len_tgt = 0

    for item in ds_raw:
        src_ids = tokenizer_src.encode(item['translation'][config['lang_src']]).ids
        tgt_ids = tokenizer_src.encode(item['translation'][config['lang_tgt']]).ids

        max_len_src = max(max_len_src, len(src_ids))
        max_len_tgt = max(max_len_tgt, len(tgt_ids))

    print(f'Max length of source text: {max_len_src}')
    print(f'Max length of target text: {max_len_tgt}')

    train_dataloader = DataLoader(train_ds, batch_size=config('batch_size'), shuffle=True)
    val_dataloader = DataLoader(val_ds, batch_size=1, shuffle=False)

    return train_dataloader, val_dataloader, tokenizer_src, tokenizer_tgt


def get_model(config, vocab_src_len, vocab_tgt_len):
    model = build_transformer(vocab_src_len, vocab_tgt_len, config['seq_len'], config['seq_len'], config['d_model'])
    return model


def train_model(config):
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    print(f'Using device: {device}')

    Path(config['model_folder']).mkdir(parents=True, exist_ok=True)

    train_dataloader, val_dataloader, tokenizer_src, tokenizer_tgt = get_ds(config)

    model = get_model(config, tokenizer_src.get_vocab_size(), tokenizer_tgt.get_vocab_size()).to(device)

    writer = SummaryWriter(config['experiment_name'])

    optimizer = torch.optim.Adam(model.parameters(), lr=config['lr'], eps=1e-9)

    inital_epoch = 0
    global_step = 0

    if config['preload']:
        model_filename = get_weights_file_path(config, config['preload'])
        print(f'Preloading model {model_filename}')
        state = torch.load(model_filename)

        inital_epoch = state['epoch'] + 1
        optimizer.load_state_dict(state['optimizer_state_dict'])
        global_step = state['global_step']

    loss_fn = nn.CrossEntropyLoss(ignore_index=tokenizer_src.token_to_id('[PAD]'), label_smoothing=0.1).to(device)

    for epoch in range(inital_epoch, config['num_epochs']):
        model.train()
        batch_iterator = tqdm(train_dataloader, desc=f'Processing epoch {epoch:0.2d}')

        for batch in batch_iterator:
            encoder_input = batch['encoder_input'].to(device)
            decoder_input = batch['decoder_input'].to(device)
            encoder_mask = batch['encoder_mask'].to(device)
            decoder_mask = batch['decoder_mask'].to(device)


            encoder_output = model.encode(encoder_input, encoder_mask)
            decoder_output = model.decode(encoder_output, encoder_mask, decoder_input, decoder_mask)
            proj_output = model.project(decoder_output)

            lable = batch['label'].to(device)

            loss = loss_fn(proj_output.view(-1, tokenizer_tgt.get_vocab_size()), lable.view(-1))
            batch_iterator.set_postfix({f"loss": f"{loss.item():6.3f}"})

            writer.add_scalar('train_loss', loss.item(), global_step)
            writer.flush()

            loss.backward()

            optimizer.step()
            optimizer.zero_grad()

            global_step += 1

        model_filename = get_weights_file_path(config, epoch)
        torch.save({
            'epoch': epoch,
            'model_state_dict': model.state_dict(),
            'optimizer_state_dict': optimizer.state_dict(),
            'global_step': global_step
        }, model_filename)

if __name__ == '__main__':
    warnings.filterwarnings("ignore")
    config = get_config()
    train_model(config)