"docs/source/en/model_doc/swin.md" did not exist on "bb300ac686f2102b52091935e22879254822baf6"
run_translation.py 25.7 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
# coding=utf-8
# Copyright The HuggingFace Team and The HuggingFace Inc. 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 sequence to sequence.
"""
# You can also adapt this script on your own sequence to sequence task. Pointers for this are left as comments.

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

27
import datasets
28
29
30
31
32
33
34
35
36
37
import numpy as np
from datasets import load_dataset, load_metric

import transformers
from transformers import (
    AutoConfig,
    AutoModelForSeq2SeqLM,
    AutoTokenizer,
    DataCollatorForSeq2Seq,
    HfArgumentParser,
38
39
40
    M2M100Tokenizer,
    MBart50Tokenizer,
    MBart50TokenizerFast,
41
    MBartTokenizer,
42
    MBartTokenizerFast,
43
44
45
46
47
    Seq2SeqTrainer,
    Seq2SeqTrainingArguments,
    default_data_collator,
    set_seed,
)
48
from transformers.trainer_utils import get_last_checkpoint
49
from transformers.utils import check_min_version
50
from transformers.utils.versions import require_version
51
52


53
# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
Lysandre's avatar
Lysandre committed
54
check_min_version("4.8.0.dev0")
55
require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/translation/requirements.txt")
56

57
58
logger = logging.getLogger(__name__)

59
60
61
# A list of all multilingual tokenizer which require src_lang and tgt_lang attributes.
MULTILINGUAL_TOKENIZERS = [MBartTokenizer, MBartTokenizerFast, MBart50Tokenizer, MBart50TokenizerFast, M2M100Tokenizer]

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

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

    model_name_or_path: str = field(
        metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
    )
    config_name: Optional[str] = field(
        default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
    )
    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(
        default=None,
        metadata={"help": "Where to store the pretrained models downloaded from huggingface.co"},
    )
    use_fast_tokenizer: bool = field(
        default=True,
        metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
    )
    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)."
        },
    )


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

105
106
107
    source_lang: str = field(default=None, metadata={"help": "Source language id for translation."})
    target_lang: str = field(default=None, metadata={"help": "Target language id for translation."})

108
109
110
111
112
113
    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)."}
    )
114
    train_file: Optional[str] = field(default=None, metadata={"help": "The input training data file (a jsonlines)."})
115
116
    validation_file: Optional[str] = field(
        default=None,
117
        metadata={
118
119
            "help": "An optional input evaluation data file to evaluate the metrics (sacreblue) on "
            "a jsonlines file."
120
121
122
123
124
        },
    )
    test_file: Optional[str] = field(
        default=None,
        metadata={
125
            "help": "An optional input test data file to evaluate the metrics (sacreblue) on " "a jsonlines file."
126
        },
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
    )
    overwrite_cache: bool = field(
        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
    )
    preprocessing_num_workers: Optional[int] = field(
        default=None,
        metadata={"help": "The number of processes to use for the preprocessing."},
    )
    max_source_length: Optional[int] = field(
        default=1024,
        metadata={
            "help": "The maximum total input sequence length after tokenization. Sequences longer "
            "than this will be truncated, sequences shorter will be padded."
        },
    )
    max_target_length: Optional[int] = field(
        default=128,
        metadata={
            "help": "The maximum total sequence length for target text after tokenization. Sequences longer "
            "than this will be truncated, sequences shorter will be padded."
        },
    )
    val_max_target_length: Optional[int] = field(
150
        default=None,
151
152
        metadata={
            "help": "The maximum total sequence length for validation target text after tokenization. Sequences longer "
153
            "than this will be truncated, sequences shorter will be padded. Will default to `max_target_length`."
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
            "This argument is also used to override the ``max_length`` param of ``model.generate``, which is used "
            "during ``evaluate`` and ``predict``."
        },
    )
    pad_to_max_length: bool = field(
        default=False,
        metadata={
            "help": "Whether to pad all samples to model maximum sentence length. "
            "If False, will pad the samples dynamically when batching to the maximum length in the batch. More "
            "efficient on GPU but very bad for TPU."
        },
    )
    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
    max_predict_samples: Optional[int] = field(
181
182
        default=None,
        metadata={
183
            "help": "For debugging purposes or quicker training, truncate the number of prediction examples to this "
184
185
186
187
188
189
190
191
192
193
            "value if set."
        },
    )
    num_beams: Optional[int] = field(
        default=None,
        metadata={
            "help": "Number of beams to use for evaluation. This argument will be passed to ``model.generate``, "
            "which is used during ``evaluate`` and ``predict``."
        },
    )
194
195
196
197
198
199
    ignore_pad_token_for_loss: bool = field(
        default=True,
        metadata={
            "help": "Whether to ignore the tokens corresponding to padded labels in the loss computation or not."
        },
    )
200
201
202
    source_prefix: Optional[str] = field(
        default=None, metadata={"help": "A prefix to add before every source text (useful for T5 models)."}
    )
203
204
205
206
207
208
209
210
    forced_bos_token: Optional[str] = field(
        default=None,
        metadata={
            "help": "The token to force as the first generated token after the :obj:`decoder_start_token_id`."
            "Useful for multilingual models like :doc:`mBART <../model_doc/mbart>` where the first generated token "
            "needs to be the target language token.(Usually it is the target language token)"
        },
    )
211
212
213
214

    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.")
215
216
217
218
219
220
221
222
223
        elif self.source_lang is None or self.target_lang is None:
            raise ValueError("Need to specify the source language and the target language.")

        if self.train_file is not None:
            extension = self.train_file.split(".")[-1]
            assert extension == "json", "`train_file` should be a json file."
        if self.validation_file is not None:
            extension = self.validation_file.split(".")[-1]
            assert extension == "json", "`validation_file` should be a json file."
224
225
        if self.val_max_target_length is None:
            self.val_max_target_length = self.max_target_length
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240


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, Seq2SeqTrainingArguments))
    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()

241
242
243
244
245
246
    # Setup logging
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s -   %(message)s",
        datefmt="%m/%d/%Y %H:%M:%S",
        handlers=[logging.StreamHandler(sys.stdout)],
    )
247

248
    log_level = training_args.get_process_log_level()
249
250
251
    logger.setLevel(log_level)
    datasets.utils.logging.set_verbosity(log_level)
    transformers.utils.logging.set_verbosity(log_level)
252
253
254
255
256
257
258
259

    # 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}"
    )
    logger.info(f"Training/evaluation parameters {training_args}")

260
261
262
263
264
265
266
267
268
269
270
271
    if data_args.source_prefix is None and model_args.model_name_or_path in [
        "t5-small",
        "t5-base",
        "t5-large",
        "t5-3b",
        "t5-11b",
    ]:
        logger.warning(
            "You're running a t5 model but didn't provide a source prefix, which is expected, e.g. with "
            "`--source_prefix 'translate English to German: ' `"
        )

272
273
274
275
276
277
278
279
280
    # 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."
            )
281
        elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
282
283
284
285
            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."
            )
286
287
288
289

    # Set seed before initializing model.
    set_seed(training_args.seed)

290
    # Get the datasets: you can either provide your own JSON training and evaluation files (see below)
291
292
293
294
295
296
297
298
299
300
    # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
    # (the dataset will be downloaded automatically from the datasets Hub).
    #
    # For translation, only JSON files are supported, with one field named "translation" containing two keys for the
    # source and target languages (unless you adapt what follows).
    #
    # 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.
301
302
303
        raw_datasets = load_dataset(
            data_args.dataset_name, data_args.dataset_config_name, cache_dir=model_args.cache_dir
        )
304
305
306
307
308
309
310
311
    else:
        data_files = {}
        if data_args.train_file is not None:
            data_files["train"] = data_args.train_file
            extension = data_args.train_file.split(".")[-1]
        if data_args.validation_file is not None:
            data_files["validation"] = data_args.validation_file
            extension = data_args.validation_file.split(".")[-1]
312
313
314
        if data_args.test_file is not None:
            data_files["test"] = data_args.test_file
            extension = data_args.test_file.split(".")[-1]
315
        raw_datasets = load_dataset(extension, data_files=data_files, cache_dir=model_args.cache_dir)
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
    # 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.
    config = AutoConfig.from_pretrained(
        model_args.config_name if model_args.config_name else model_args.model_name_or_path,
        cache_dir=model_args.cache_dir,
        revision=model_args.model_revision,
        use_auth_token=True if model_args.use_auth_token else None,
    )
    tokenizer = AutoTokenizer.from_pretrained(
        model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path,
        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,
    )
    model = AutoModelForSeq2SeqLM.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,
        revision=model_args.model_revision,
        use_auth_token=True if model_args.use_auth_token else None,
    )

Suraj Patil's avatar
Suraj Patil committed
346
347
    model.resize_token_embeddings(len(tokenizer))

348
    # Set decoder_start_token_id
349
350
351
352
353
354
    if model.config.decoder_start_token_id is None and isinstance(tokenizer, (MBartTokenizer, MBartTokenizerFast)):
        if isinstance(tokenizer, MBartTokenizer):
            model.config.decoder_start_token_id = tokenizer.lang_code_to_id[data_args.target_lang]
        else:
            model.config.decoder_start_token_id = tokenizer.convert_tokens_to_ids(data_args.target_lang)

355
356
357
    if model.config.decoder_start_token_id is None:
        raise ValueError("Make sure that `config.decoder_start_token_id` is correctly defined")

358
    prefix = data_args.source_prefix if data_args.source_prefix is not None else ""
359

360
361
362
    # Preprocessing the datasets.
    # We need to tokenize inputs and targets.
    if training_args.do_train:
363
        column_names = raw_datasets["train"].column_names
364
    elif training_args.do_eval:
365
        column_names = raw_datasets["validation"].column_names
366
    elif training_args.do_predict:
367
        column_names = raw_datasets["test"].column_names
368
369
370
    else:
        logger.info("There is nothing to do. Please pass `do_train`, `do_eval` and/or `do_predict`.")
        return
371
372
373

    # For translation we set the codes of our source and target languages (only useful for mBART, the others will
    # ignore those attributes).
374
375
376
377
378
379
380
381
382
383
384
385
386
387
    if isinstance(tokenizer, tuple(MULTILINGUAL_TOKENIZERS)):
        assert data_args.target_lang is not None and data_args.source_lang is not None, (
            f"{tokenizer.__class__.__name__} is a multilingual tokenizer which requires --source_lang and "
            "--target_lang arguments."
        )

        tokenizer.src_lang = data_args.source_lang
        tokenizer.tgt_lang = data_args.target_lang

        # For multilingual translation models like mBART-50 and M2M100 we need to force the target language token
        # as the first generated token. We ask the user to explicitly provide this as --forced_bos_token argument.
        forced_bos_token_id = (
            tokenizer.lang_code_to_id[data_args.forced_bos_token] if data_args.forced_bos_token is not None else None
        )
388
        model.config.forced_bos_token_id = forced_bos_token_id
389

390
391
392
    # Get the language codes for input/target.
    source_lang = data_args.source_lang.split("_")[0]
    target_lang = data_args.target_lang.split("_")[0]
393
394
395
396
397

    # Temporarily set max_target_length for training.
    max_target_length = data_args.max_target_length
    padding = "max_length" if data_args.pad_to_max_length else False

398
    if training_args.label_smoothing_factor > 0 and not hasattr(model, "prepare_decoder_input_ids_from_labels"):
399
        logger.warning(
400
401
402
403
            "label_smoothing is enabled but the `prepare_decoder_input_ids_from_labels` method is not defined for"
            f"`{model.__class__.__name__}`. This will lead to loss being calculated twice and will take up more memory"
        )

404
    def preprocess_function(examples):
405
406
        inputs = [ex[source_lang] for ex in examples["translation"]]
        targets = [ex[target_lang] for ex in examples["translation"]]
407
        inputs = [prefix + inp for inp in inputs]
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
        model_inputs = tokenizer(inputs, max_length=data_args.max_source_length, padding=padding, truncation=True)

        # Setup the tokenizer for targets
        with tokenizer.as_target_tokenizer():
            labels = tokenizer(targets, max_length=max_target_length, padding=padding, truncation=True)

        # If we are padding here, replace all tokenizer.pad_token_id in the labels by -100 when we want to ignore
        # padding in the loss.
        if padding == "max_length" and data_args.ignore_pad_token_for_loss:
            labels["input_ids"] = [
                [(l if l != tokenizer.pad_token_id else -100) for l in label] for label in labels["input_ids"]
            ]

        model_inputs["labels"] = labels["input_ids"]
        return model_inputs

    if training_args.do_train:
425
        if "train" not in raw_datasets:
426
            raise ValueError("--do_train requires a train dataset")
427
        train_dataset = raw_datasets["train"]
428
429
430
431
432
433
434
435
        if data_args.max_train_samples is not None:
            train_dataset = train_dataset.select(range(data_args.max_train_samples))
        train_dataset = train_dataset.map(
            preprocess_function,
            batched=True,
            num_proc=data_args.preprocessing_num_workers,
            remove_columns=column_names,
            load_from_cache_file=not data_args.overwrite_cache,
436
            desc="Running tokenizer on train dataset",
437
438
439
440
        )

    if training_args.do_eval:
        max_target_length = data_args.val_max_target_length
441
        if "validation" not in raw_datasets:
442
            raise ValueError("--do_eval requires a validation dataset")
443
        eval_dataset = raw_datasets["validation"]
444
445
        if data_args.max_eval_samples is not None:
            eval_dataset = eval_dataset.select(range(data_args.max_eval_samples))
446
447
448
449
450
451
        eval_dataset = eval_dataset.map(
            preprocess_function,
            batched=True,
            num_proc=data_args.preprocessing_num_workers,
            remove_columns=column_names,
            load_from_cache_file=not data_args.overwrite_cache,
452
            desc="Running tokenizer on validation dataset",
453
454
        )

455
456
    if training_args.do_predict:
        max_target_length = data_args.val_max_target_length
457
        if "test" not in raw_datasets:
458
            raise ValueError("--do_predict requires a test dataset")
459
        predict_dataset = raw_datasets["test"]
460
461
462
        if data_args.max_predict_samples is not None:
            predict_dataset = predict_dataset.select(range(data_args.max_predict_samples))
        predict_dataset = predict_dataset.map(
463
464
465
466
467
            preprocess_function,
            batched=True,
            num_proc=data_args.preprocessing_num_workers,
            remove_columns=column_names,
            load_from_cache_file=not data_args.overwrite_cache,
468
            desc="Running tokenizer on prediction dataset",
469
470
        )

471
472
473
474
475
    # Data collator
    label_pad_token_id = -100 if data_args.ignore_pad_token_for_loss else tokenizer.pad_token_id
    if data_args.pad_to_max_length:
        data_collator = default_data_collator
    else:
476
477
        data_collator = DataCollatorForSeq2Seq(
            tokenizer,
478
            model=model,
479
480
481
            label_pad_token_id=label_pad_token_id,
            pad_to_multiple_of=8 if training_args.fp16 else None,
        )
482
483

    # Metric
484
    metric = load_metric("sacrebleu")
485

486
487
    def postprocess_text(preds, labels):
        preds = [pred.strip() for pred in preds]
488
        labels = [[label.strip()] for label in labels]
489
490
491

        return preds, labels

492
493
494
495
496
497
498
499
500
501
502
    def compute_metrics(eval_preds):
        preds, labels = eval_preds
        if isinstance(preds, tuple):
            preds = preds[0]
        decoded_preds = tokenizer.batch_decode(preds, skip_special_tokens=True)
        if data_args.ignore_pad_token_for_loss:
            # Replace -100 in the labels as we can't decode them.
            labels = np.where(labels != -100, labels, tokenizer.pad_token_id)
        decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)

        # Some simple post-processing
503
        decoded_preds, decoded_labels = postprocess_text(decoded_preds, decoded_labels)
504

505
506
        result = metric.compute(predictions=decoded_preds, references=decoded_labels)
        result = {"bleu": result["score"]}
507
508
509

        prediction_lens = [np.count_nonzero(pred != tokenizer.pad_token_id) for pred in preds]
        result["gen_len"] = np.mean(prediction_lens)
510
        result = {k: round(v, 4) for k, v in result.items()}
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
        return result

    # Initialize our Trainer
    trainer = Seq2SeqTrainer(
        model=model,
        args=training_args,
        train_dataset=train_dataset if training_args.do_train else None,
        eval_dataset=eval_dataset if training_args.do_eval else None,
        tokenizer=tokenizer,
        data_collator=data_collator,
        compute_metrics=compute_metrics if training_args.predict_with_generate else None,
    )

    # Training
    if training_args.do_train:
526
527
528
529
        checkpoint = None
        if training_args.resume_from_checkpoint is not None:
            checkpoint = training_args.resume_from_checkpoint
        elif last_checkpoint is not None:
530
531
            checkpoint = last_checkpoint
        train_result = trainer.train(resume_from_checkpoint=checkpoint)
532
533
        trainer.save_model()  # Saves the tokenizer too for easy upload

534
535
536
537
538
        metrics = train_result.metrics
        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))
539

540
541
542
        trainer.log_metrics("train", metrics)
        trainer.save_metrics("train", metrics)
        trainer.save_state()
543
544

    # Evaluation
545
    results = {}
546
547
548
    if training_args.do_eval:
        logger.info("*** Evaluate ***")

549
        metrics = trainer.evaluate(
550
            max_length=data_args.val_max_target_length, num_beams=data_args.num_beams, metric_key_prefix="eval"
551
        )
552
553
        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))
554

555
556
        trainer.log_metrics("eval", metrics)
        trainer.save_metrics("eval", metrics)
557

558
    if training_args.do_predict:
559
        logger.info("*** Predict ***")
560

561
562
563
        predict_results = trainer.predict(
            predict_dataset,
            metric_key_prefix="predict",
564
565
566
            max_length=data_args.val_max_target_length,
            num_beams=data_args.num_beams,
        )
567
568
569
570
571
        metrics = predict_results.metrics
        max_predict_samples = (
            data_args.max_predict_samples if data_args.max_predict_samples is not None else len(predict_dataset)
        )
        metrics["predict_samples"] = min(max_predict_samples, len(predict_dataset))
572

573
574
        trainer.log_metrics("predict", metrics)
        trainer.save_metrics("predict", metrics)
575

576
        if trainer.is_world_process_zero():
577
            if training_args.predict_with_generate:
578
579
                predictions = tokenizer.batch_decode(
                    predict_results.predictions, skip_special_tokens=True, clean_up_tokenization_spaces=True
580
                )
581
582
                predictions = [pred.strip() for pred in predictions]
                output_prediction_file = os.path.join(training_args.output_dir, "generated_predictions.txt")
583
                with open(output_prediction_file, "w", encoding="utf-8") as writer:
584
                    writer.write("\n".join(predictions))
585

Sylvain Gugger's avatar
Sylvain Gugger committed
586
    if training_args.push_to_hub:
Sylvain Gugger's avatar
Sylvain Gugger committed
587
        kwargs = {"finetuned_from": model_args.model_name_or_path, "tasks": "translation"}
Sylvain Gugger's avatar
Sylvain Gugger committed
588
589
590
591
592
593
594
595
596
597
598
599
600
        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

        languages = [l for l in [data_args.source_lang, data_args.target_lang] if l is not None]
        if len(languages) > 0:
            kwargs["language"] = languages

        trainer.push_to_hub(**kwargs)
Sylvain Gugger's avatar
Sylvain Gugger committed
601

602
603
    return results

604
605
606
607
608
609
610
611

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


if __name__ == "__main__":
    main()