File size: 9,767 Bytes
50f0fbb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
from dataclasses import dataclass
from transformers import (
    MegatronBertConfig,
    MegatronBertForPreTraining,
    AutoTokenizer,
)
from pytorch_lightning import (
    LightningModule,
    Trainer,
)
from pytorch_lightning.callbacks import (
    LearningRateMonitor,
)
import argparse
import torch
import os
import numpy as np
import time
from fengshen.data.universal_datamodule import UniversalDataModule
from fengshen.data.data_utils.sop_utils import get_a_and_b_segments
from fengshen.data.data_utils.truncate_utils import truncate_segments
from fengshen.data.data_utils.token_type_utils import create_tokens_and_tokentypes
from fengshen.data.data_utils.mask_utils import create_masked_lm_predictions
from fengshen.models.model_utils import (
    add_module_args,
    configure_optimizers,
    get_total_steps,
)
from fengshen.utils.universal_checkpoint import UniversalCheckpoint
from torch.utils.data._utils.collate import default_collate

SHOW_DATA = False


@dataclass
class ErLangShenCollator:
    '''
    由input处理成samples,也就是最终模型的输入
    其中主要处理逻辑在__call__里
    包含Mask和Sop任务
    '''
    tokenizer: None  # 分词
    max_seq_length: 512
    masked_lm_prob: 0.15
    content_key: str = 'text'
    # 一些预处理操作

    def setup(self):
        from fengshen.data.data_utils.sentence_split import ChineseSentenceSplitter
        self.sentence_split = ChineseSentenceSplitter()
        self.np_rng = np.random.RandomState(seed=((int(time.time()) % 2**32)))
        inv_vocab = {v: k for k, v in self.tokenizer.vocab.items()}
        self.vocab_id_list = list(inv_vocab.keys())
        self.vocab_id_to_token_dict = inv_vocab

    def __call__(self, samples):
        '''
        samples: 一个sample长这样{"text": "hello world"}
        '''
        model_inputs = []
        for s in samples:
            sentences = self.sentence_split.tokenize(s[self.content_key])
            # Divide sample into two segments (A and B).
            tokenized_sentences = [self.tokenizer.convert_tokens_to_ids(
                self.tokenizer.tokenize(sent)) for sent in sentences]
            if len(tokenized_sentences) == 0:
                print('find empty sentence')
                continue
            if len(tokenized_sentences) > 1:
                tokens_a, tokens_b, is_next_random = get_a_and_b_segments(tokenized_sentences,
                                                                          self.np_rng)
            else:
                tokens_a = tokenized_sentences[0]
                tokens_b = []
                is_next_random = False
            # max_seq_length - 3因为还需要拼上[CLS] [SEP] [SEP]
            if len(tokens_a) == 0:
                continue
            _ = truncate_segments(tokens_a, tokens_b, len(tokens_a),
                                  len(tokens_b), self.max_seq_length-3, self.np_rng)
            # Build tokens and toketypes.
            tokens, tokentypes = create_tokens_and_tokentypes(tokens_a, tokens_b,
                                                              self.tokenizer.cls_token_id, self.tokenizer.sep_token_id)
            # Masking.
            max_predictions_per_seq = self.masked_lm_prob * len(tokens)
            (tokens, masked_positions, masked_labels, _, _) = create_masked_lm_predictions(
                tokens, self.vocab_id_list, self.vocab_id_to_token_dict, self.masked_lm_prob,
                self.tokenizer.cls_token_id, self.tokenizer.sep_token_id, self.tokenizer.mask_token_id,
                max_predictions_per_seq, self.np_rng,
                masking_style='bert')

            # Some checks.
            num_tokens = len(tokens)
            padding_length = self.max_seq_length - num_tokens
            assert padding_length >= 0
            assert len(tokentypes) == num_tokens
            assert len(masked_positions) == len(masked_labels)

            # Tokens and token types.
            filler = [self.tokenizer.pad_token_id] * padding_length
            tokens_np = np.array(tokens + filler, dtype=np.int64)
            tokentypes_np = np.array(tokentypes + filler, dtype=np.int64)

            # Padding mask.
            padding_mask_np = np.array([1] * num_tokens + [0] * padding_length,
                                       dtype=np.int64)

            # Lables and loss mask.
            labels = [-100] * self.max_seq_length
            for i in range(len(masked_positions)):
                assert masked_positions[i] < num_tokens
                labels[masked_positions[i]] = masked_labels[i]
            labels_np = np.array(labels, dtype=np.int64)
            model_inputs.append(
                {
                    'input_ids': tokens_np,
                    'attention_mask': padding_mask_np,
                    'token_type_ids': tokentypes_np,
                    'labels': labels_np,
                    'next_sentence_label': int(is_next_random)
                }
            )
        return default_collate(model_inputs)


class ErLangShenBert(LightningModule):
    @staticmethod
    def add_module_specific_args(parent_parser):
        parser = parent_parser.add_argument_group('Erlangshen Bert')
        parser.add_argument('--masked_lm_prob', type=float, default=0.15)
        parser.add_argument('--max_seq_length', type=int, default=512)
        parser.add_argument('--sample_content_key', type=str, default='text')
        return parent_parser

    def __init__(self, args, tokenizer, **kwargs) -> None:
        super().__init__()
        self.save_hyperparameters(args)
        config = MegatronBertConfig.from_pretrained(args.model_path)
        self.config = config
        self.tokenizer = tokenizer
        self.model = MegatronBertForPreTraining(config)

    def setup(self, stage) -> None:
        if stage == 'fit':
            self.total_steps = get_total_steps(self.trainer, self.hparams)
            print('Total steps: {}' .format(self.total_steps))

    def configure_optimizers(self):
        return configure_optimizers(self)

    def forward(self, **batch):
        return self.model(**batch)

    def detokenize(self, token_ids):
        toks = self.tokenizer.convert_ids_to_tokens(token_ids)
        return self.tokenizer.convert_tokens_to_string(toks)

    def comput_metrix(self, logits, labels):
        y_pred = torch.argmax(logits, dim=-1)
        y_pred = y_pred.view(size=(-1,))
        y_true = labels.view(size=(-1,)).float()
        corr = torch.eq(y_pred, y_true)
        acc = torch.sum(corr.float())/labels.shape[0]
        return acc

    def training_step(self, batch, batch_idx):
        if self.trainer.global_rank == 0:
            global SHOW_DATA
            if not SHOW_DATA:
                print(self.config)
                print(self.model)
                SHOW_DATA = True
                print('source: {}'.format(batch['input_ids'][0]))
                print('target: {}'.format(batch['labels'][0]))
                print('source: {}'.format(self.detokenize(batch['input_ids'][0])))
                label_idx = batch['labels'][0] != -100
                print('target: {}'.format(self.detokenize(
                    batch['labels'][0][label_idx])))
        output = self(**batch)
        self.log('train_loss', output.loss, sync_dist=True)
        label_idx = batch['labels'] != -100
        acc = self.comput_metrix(
            output.prediction_logits[label_idx].view(-1, output.prediction_logits.size(-1)), batch['labels'][label_idx])
        self.log('train_acc', acc, sync_dist=True)
        return output.loss

    def validation_step(self, batch, batch_idx):
        output = self(**batch)
        self.log('val_loss', output.loss, sync_dist=True)
        return output.loss

    def on_load_checkpoint(self, checkpoint) -> None:
        # 兼容低版本lightning,低版本lightning从ckpt起来时steps数会被重置为0
        global_step_offset = checkpoint["global_step"]
        if 'global_samples' in checkpoint:
            self.consumed_samples = checkpoint['global_samples']
        self.trainer.fit_loop.epoch_loop._batches_that_stepped = global_step_offset


if __name__ == '__main__':
    args_parser = argparse.ArgumentParser()
    args_parser = add_module_args(args_parser)
    args_parser = UniversalDataModule.add_data_specific_args(args_parser)
    args_parser = Trainer.add_argparse_args(args_parser)
    args_parser = ErLangShenBert.add_module_specific_args(args_parser)
    args_parser = UniversalCheckpoint.add_argparse_args(args_parser)
    args = args_parser.parse_args()

    tokenizer = AutoTokenizer.from_pretrained(args.model_path)
    collate_fn = ErLangShenCollator(
        tokenizer=tokenizer,
        max_seq_length=args.max_seq_length,
        masked_lm_prob=args.masked_lm_prob,
        content_key=args.sample_content_key,
    )
    collate_fn.setup()
    data_module = UniversalDataModule(tokenizer=tokenizer, args=args, collate_fn=collate_fn)
    print('data load complete')

    model = ErLangShenBert(args, tokenizer=tokenizer)
    print('model load complete')

    lr_monitor = LearningRateMonitor(logging_interval='step')
    checkpoint_callback = UniversalCheckpoint(args)

    # 做兼容,如果目录不存在的话把这个参数去掉,不然会报错
    if args.load_ckpt_path is not None and \
            not os.path.exists(args.load_ckpt_path):
        print('--------warning no checkpoint found--------, remove args')
        args.load_ckpt_path = None

    trainer = Trainer.from_argparse_args(args,
                                         callbacks=[
                                             lr_monitor,
                                             checkpoint_callback])

    trainer.fit(model, data_module, ckpt_path=args.load_ckpt_path)