run_plm.py 20.9 KB
Newer Older
1
#!/usr/bin/env python
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
# coding=utf-8
# Copyright 2020 The HuggingFace Team All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Fine-tuning the library models for permutation language modeling.
"""
# You can also adapt this script on your own permutation language modeling task. Pointers for this are left as comments.

import logging
import math
import os
import sys
from dataclasses import dataclass, field
from typing import Optional

from datasets import load_dataset

import transformers
from transformers import (
    AutoConfig,
    AutoTokenizer,
    DataCollatorForPermutationLanguageModeling,
    HfArgumentParser,
    Trainer,
    TrainingArguments,
    XLNetConfig,
    XLNetLMHeadModel,
    set_seed,
)
42
from transformers.trainer_utils import get_last_checkpoint
43
from transformers.utils import check_min_version
44
from transformers.utils.versions import require_version
45
46


47
# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
Sylvain Gugger's avatar
Sylvain Gugger committed
48
check_min_version("4.9.0.dev0")
Sylvain Gugger's avatar
Sylvain Gugger committed
49

50
require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")
51

52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
logger = logging.getLogger(__name__)


@dataclass
class ModelArguments:
    """
    Arguments pertaining to which model/config/tokenizer we are going to fine-tune, or train from scratch.
    """

    model_name_or_path: Optional[str] = field(
        default=None,
        metadata={
            "help": "The model checkpoint for weights initialization."
            "Don't set if you want to train a model from scratch."
        },
    )
    config_name: Optional[str] = field(
        default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
    )
71
72
73
74
75
76
77
    config_overrides: Optional[str] = field(
        default=None,
        metadata={
            "help": "Override some existing default config settings when a model is trained from scratch. Example: "
            "n_embd=10,resid_pdrop=0.2,scale_attn_weights=false,summary_type=cls_index"
        },
    )
78
79
80
81
    tokenizer_name: Optional[str] = field(
        default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
    )
    cache_dir: Optional[str] = field(
82
83
        default=None,
        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
84
85
86
87
88
    )
    use_fast_tokenizer: bool = field(
        default=True,
        metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
    )
89
90
91
92
93
94
95
96
97
98
99
    model_revision: str = field(
        default="main",
        metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."},
    )
    use_auth_token: bool = field(
        default=False,
        metadata={
            "help": "Will use the token generated when running `transformers-cli login` (necessary to use this script "
            "with private models)."
        },
    )
100

101
102
103
104
105
106
    def __post_init__(self):
        if self.config_overrides is not None and (self.config_name is not None or self.model_name_or_path is not None):
            raise ValueError(
                "--config_overrides can't be used in combination with --config_name or --model_name_or_path"
            )

107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127

@dataclass
class DataTrainingArguments:
    """
    Arguments pertaining to what data we are going to input our model for training and eval.
    """

    dataset_name: Optional[str] = field(
        default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
    )
    dataset_config_name: Optional[str] = field(
        default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
    )
    train_file: Optional[str] = field(default=None, metadata={"help": "The input training data file (a text file)."})
    validation_file: Optional[str] = field(
        default=None,
        metadata={"help": "An optional input evaluation data file to evaluate the perplexity on (a text file)."},
    )
    overwrite_cache: bool = field(
        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
    )
128
129
130
131
132
133
    validation_split_percentage: Optional[int] = field(
        default=5,
        metadata={
            "help": "The percentage of the train set used as validation set in case there's no validation split"
        },
    )
134
135
    max_seq_length: int = field(
        default=512,
136
137
        metadata={
            "help": "The maximum total input sequence length after tokenization. Sequences longer "
138
            "than this will be truncated."
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
        },
    )
    preprocessing_num_workers: Optional[int] = field(
        default=None,
        metadata={"help": "The number of processes to use for the preprocessing."},
    )
    plm_probability: float = field(
        default=1 / 6,
        metadata={
            "help": "Ratio of length of a span of masked tokens to surrounding context length for "
            "permutation language modeling."
        },
    )
    max_span_length: int = field(
        default=5, metadata={"help": "Maximum length of a span of masked tokens for permutation language modeling."}
    )
155
156
157
158
159
160
161
162
163
164
165
    line_by_line: bool = field(
        default=False,
        metadata={"help": "Whether distinct lines of text in the dataset are to be handled as distinct sequences."},
    )
    pad_to_max_length: bool = field(
        default=False,
        metadata={
            "help": "Whether to pad all samples to `max_seq_length`. "
            "If False, will pad the samples dynamically when batching to the maximum length in the batch."
        },
    )
166
167
168
169
170
171
172
    max_train_samples: Optional[int] = field(
        default=None,
        metadata={
            "help": "For debugging purposes or quicker training, truncate the number of training examples to this "
            "value if set."
        },
    )
173
    max_eval_samples: Optional[int] = field(
174
175
        default=None,
        metadata={
176
            "help": "For debugging purposes or quicker training, truncate the number of evaluation examples to this "
177
178
179
            "value if set."
        },
    )
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

    def __post_init__(self):
        if self.dataset_name is None and self.train_file is None and self.validation_file is None:
            raise ValueError("Need either a dataset name or a training/validation file.")
        else:
            if self.train_file is not None:
                extension = self.train_file.split(".")[-1]
                assert extension in ["csv", "json", "txt"], "`train_file` should be a csv, a json or a txt file."
            if self.validation_file is not None:
                extension = self.validation_file.split(".")[-1]
                assert extension in ["csv", "json", "txt"], "`validation_file` should be a csv, a json or a txt file."


def main():
    # See all possible arguments in src/transformers/training_args.py
    # or by passing the --help flag to this script.
    # We now keep distinct sets of args, for a cleaner separation of concerns.

    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
        # If we pass only one argument to the script and it's the path to a json file,
        # let's parse it to get our arguments.
        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
    else:
        model_args, data_args, training_args = parser.parse_args_into_dataclasses()

    # Setup logging
    logging.basicConfig(
208
        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
209
        datefmt="%m/%d/%Y %H:%M:%S",
210
        handlers=[logging.StreamHandler(sys.stdout)],
211
    )
212
    logger.setLevel(logging.INFO if training_args.should_log else logging.WARN)
213
214
215
216
217
218
219

    # Log on each process the small summary:
    logger.warning(
        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
        + f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
    )
    # Set the verbosity to info of the Transformers logger (on main process only):
220
    if training_args.should_log:
221
        transformers.utils.logging.set_verbosity_info()
222
223
        transformers.utils.logging.enable_default_handler()
        transformers.utils.logging.enable_explicit_format()
224
    logger.info(f"Training/evaluation parameters {training_args}")
225

226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
    # Detecting last checkpoint.
    last_checkpoint = None
    if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
        last_checkpoint = get_last_checkpoint(training_args.output_dir)
        if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
            raise ValueError(
                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
                "Use --overwrite_output_dir to overcome."
            )
        elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
            logger.info(
                f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
                "the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
            )

241
242
243
244
245
    # Set seed before initializing model.
    set_seed(training_args.seed)

    # Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
    # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
Sylvain Gugger's avatar
Sylvain Gugger committed
246
    # (the dataset will be downloaded automatically from the datasets Hub).
247
    #
Sylvain Gugger's avatar
Sylvain Gugger committed
248
249
    # For CSV/JSON files, this script will use the column called 'text' or the first column if no column called
    # 'text' is found. You can easily tweak this behavior (see below).
250
251
252
253
254
    #
    # In distributed training, the load_dataset function guarantee that only one local process can concurrently
    # download the dataset.
    if data_args.dataset_name is not None:
        # Downloading and loading a dataset from the hub.
255
        datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name, cache_dir=model_args.cache_dir)
256
257
258
259
260
        if "validation" not in datasets.keys():
            datasets["validation"] = load_dataset(
                data_args.dataset_name,
                data_args.dataset_config_name,
                split=f"train[:{data_args.validation_split_percentage}%]",
261
                cache_dir=model_args.cache_dir,
262
263
264
265
266
            )
            datasets["train"] = load_dataset(
                data_args.dataset_name,
                data_args.dataset_config_name,
                split=f"train[{data_args.validation_split_percentage}%:]",
267
                cache_dir=model_args.cache_dir,
268
            )
269
270
271
272
273
    else:
        data_files = {}
        if data_args.train_file is not None:
            data_files["train"] = data_args.train_file
        if data_args.validation_file is not None:
274
            data_files["validation"] = data_args.validation_file
275
276
277
        extension = data_args.train_file.split(".")[-1]
        if extension == "txt":
            extension = "text"
278
        datasets = load_dataset(extension, data_files=data_files, cache_dir=model_args.cache_dir)
279
280
281
282
283
284
285
286
    # See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at
    # https://huggingface.co/docs/datasets/loading_datasets.html.

    # Load pretrained model and tokenizer
    #
    # Distributed training:
    # The .from_pretrained methods guarantee that only one local process can concurrently
    # download model & vocab.
287
288
289
290
291
    config_kwargs = {
        "cache_dir": model_args.cache_dir,
        "revision": model_args.model_revision,
        "use_auth_token": True if model_args.use_auth_token else None,
    }
292
    if model_args.config_name:
293
        config = AutoConfig.from_pretrained(model_args.config_name, **config_kwargs)
294
    elif model_args.model_name_or_path:
295
        config = AutoConfig.from_pretrained(model_args.model_name_or_path, **config_kwargs)
296
297
298
    else:
        config = XLNetConfig()
        logger.warning("You are instantiating a new config instance from scratch.")
299
300
301
        if model_args.config_overrides is not None:
            logger.info(f"Overriding config: {model_args.config_overrides}")
            config.update_from_string(model_args.config_overrides)
302

303
304
305
306
307
308
    tokenizer_kwargs = {
        "cache_dir": model_args.cache_dir,
        "use_fast": model_args.use_fast_tokenizer,
        "revision": model_args.model_revision,
        "use_auth_token": True if model_args.use_auth_token else None,
    }
309
    if model_args.tokenizer_name:
310
        tokenizer = AutoTokenizer.from_pretrained(model_args.tokenizer_name, **tokenizer_kwargs)
311
    elif model_args.model_name_or_path:
312
        tokenizer = AutoTokenizer.from_pretrained(model_args.model_name_or_path, **tokenizer_kwargs)
313
314
315
316
317
318
319
320
321
322
323
324
    else:
        raise ValueError(
            "You are instantiating a new tokenizer from scratch. This is not supported by this script."
            "You can do it from another script, save it, and load it from here, using --tokenizer_name."
        )

    if model_args.model_name_or_path:
        model = XLNetLMHeadModel.from_pretrained(
            model_args.model_name_or_path,
            from_tf=bool(".ckpt" in model_args.model_name_or_path),
            config=config,
            cache_dir=model_args.cache_dir,
325
326
            revision=model_args.model_revision,
            use_auth_token=True if model_args.use_auth_token else None,
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
        )
    else:
        logger.info("Training new model from scratch")
        model = XLNetLMHeadModel.from_config(config)

    model.resize_token_embeddings(len(tokenizer))

    # Preprocessing the datasets.
    # First we tokenize all the texts.
    if training_args.do_train:
        column_names = datasets["train"].column_names
    else:
        column_names = datasets["validation"].column_names
    text_column_name = "text" if "text" in column_names else column_names[0]

342
    if data_args.max_seq_length > tokenizer.model_max_length:
343
        logger.warning(
344
345
346
347
348
            f"The max_seq_length passed ({data_args.max_seq_length}) is larger than the maximum length for the"
            f"model ({tokenizer.model_max_length}). Using max_seq_length={tokenizer.model_max_length}."
        )
    max_seq_length = min(data_args.max_seq_length, tokenizer.model_max_length)

349
350
351
352
353
354
355
    if data_args.line_by_line:
        # When using line_by_line, we just tokenize each nonempty line.
        padding = "max_length" if data_args.pad_to_max_length else False

        def tokenize_function(examples):
            # Remove empty lines
            examples["text"] = [line for line in examples["text"] if len(line) > 0 and not line.isspace()]
356
            return tokenizer(examples["text"], padding=padding, truncation=True, max_length=max_seq_length)
357
358
359
360
361
362
363

        tokenized_datasets = datasets.map(
            tokenize_function,
            batched=True,
            num_proc=data_args.preprocessing_num_workers,
            remove_columns=[text_column_name],
            load_from_cache_file=not data_args.overwrite_cache,
364
            desc="Running tokenizer on dataset line_by_line",
365
366
367
368
369
370
371
372
373
374
        )
    else:
        # Otherwise, we tokenize every text, then concatenate them together before splitting them in smaller parts.
        def tokenize_function(examples):
            return tokenizer(examples[text_column_name])

        tokenized_datasets = datasets.map(
            tokenize_function,
            batched=True,
            num_proc=data_args.preprocessing_num_workers,
375
            remove_columns=column_names,
376
            load_from_cache_file=not data_args.overwrite_cache,
377
            desc="Running tokenizer on every text in dataset",
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
        )

        # Main data processing function that will concatenate all texts from our dataset and generate chunks of
        # max_seq_length.
        def group_texts(examples):
            # Concatenate all texts.
            concatenated_examples = {k: sum(examples[k], []) for k in examples.keys()}
            total_length = len(concatenated_examples[list(examples.keys())[0]])
            # We drop the small remainder, we could add padding if the model supported it instead of this drop, you can
            # customize this part to your needs.
            total_length = (total_length // max_seq_length) * max_seq_length
            # Split by chunks of max_len.
            result = {
                k: [t[i : i + max_seq_length] for i in range(0, total_length, max_seq_length)]
                for k, t in concatenated_examples.items()
            }
            return result

        # Note that with `batched=True`, this map processes 1,000 texts together, so group_texts throws away a
        # remainder for each of those groups of 1,000 texts. You can adjust that batch_size here but a higher value
        # might be slower to preprocess.
        #
        # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
        # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
402

403
404
405
406
407
        tokenized_datasets = tokenized_datasets.map(
            group_texts,
            batched=True,
            num_proc=data_args.preprocessing_num_workers,
            load_from_cache_file=not data_args.overwrite_cache,
408
            desc=f"Grouping texts in chunks of {max_seq_length}",
409
        )
410

411
412
413
414
415
416
417
418
419
420
421
    if training_args.do_train:
        if "train" not in tokenized_datasets:
            raise ValueError("--do_train requires a train dataset")
        train_dataset = tokenized_datasets["train"]
        if data_args.max_train_samples is not None:
            train_dataset = train_dataset.select(range(data_args.max_train_samples))

    if training_args.do_eval:
        if "validation" not in tokenized_datasets:
            raise ValueError("--do_eval requires a validation dataset")
        eval_dataset = tokenized_datasets["validation"]
422
423
        if data_args.max_eval_samples is not None:
            eval_dataset = eval_dataset.select(range(data_args.max_eval_samples))
424

425
426
427
428
429
430
431
432
433
434
435
    # Data collator
    data_collator = DataCollatorForPermutationLanguageModeling(
        tokenizer=tokenizer,
        plm_probability=data_args.plm_probability,
        max_span_length=data_args.max_span_length,
    )

    # Initialize our Trainer
    trainer = Trainer(
        model=model,
        args=training_args,
436
437
        train_dataset=train_dataset if training_args.do_train else None,
        eval_dataset=eval_dataset if training_args.do_eval else None,
438
439
440
441
442
443
        tokenizer=tokenizer,
        data_collator=data_collator,
    )

    # Training
    if training_args.do_train:
444
445
446
447
        checkpoint = None
        if training_args.resume_from_checkpoint is not None:
            checkpoint = training_args.resume_from_checkpoint
        elif last_checkpoint is not None:
448
449
            checkpoint = last_checkpoint
        train_result = trainer.train(resume_from_checkpoint=checkpoint)
450
        trainer.save_model()  # Saves the tokenizer too for easy upload
451
        metrics = train_result.metrics
452

453
454
455
456
457
        max_train_samples = (
            data_args.max_train_samples if data_args.max_train_samples is not None else len(train_dataset)
        )
        metrics["train_samples"] = min(max_train_samples, len(train_dataset))

458
459
460
        trainer.log_metrics("train", metrics)
        trainer.save_metrics("train", metrics)
        trainer.save_state()
461

462
463
464
465
    # Evaluation
    if training_args.do_eval:
        logger.info("*** Evaluate ***")

466
        metrics = trainer.evaluate()
467

468
469
        max_eval_samples = data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)
        metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset))
470
471
472
473
        try:
            perplexity = math.exp(metrics["eval_loss"])
        except OverflowError:
            perplexity = float("inf")
474
        metrics["perplexity"] = perplexity
475

476
477
        trainer.log_metrics("eval", metrics)
        trainer.save_metrics("eval", metrics)
478

Sylvain Gugger's avatar
Sylvain Gugger committed
479
    if training_args.push_to_hub:
Sylvain Gugger's avatar
Sylvain Gugger committed
480
        kwargs = {"finetuned_from": model_args.model_name_or_path, "tasks": "language-modeling"}
Sylvain Gugger's avatar
Sylvain Gugger committed
481
482
483
484
485
486
487
488
489
        if data_args.dataset_name is not None:
            kwargs["dataset_tags"] = data_args.dataset_name
            if data_args.dataset_config_name is not None:
                kwargs["dataset_args"] = data_args.dataset_config_name
                kwargs["dataset"] = f"{data_args.dataset_name} {data_args.dataset_config_name}"
            else:
                kwargs["dataset"] = data_args.dataset_name

        trainer.push_to_hub(**kwargs)
Sylvain Gugger's avatar
Sylvain Gugger committed
490

491
492
493
494
495
496
497
498

def _mp_fn(index):
    # For xla_spawn (TPUs)
    main()


if __name__ == "__main__":
    main()