File size: 31,957 Bytes
52869cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
497a54e
 
 
 
 
 
 
 
 
 
52869cd
 
 
 
 
 
 
 
 
497a54e
52869cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
497a54e
52869cd
baaee71
5c23c43
60b3f40
52869cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
497a54e
52869cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
497a54e
 
 
 
 
 
 
 
 
52869cd
 
 
 
 
 
 
 
497a54e
 
52869cd
 
497a54e
 
52869cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
497a54e
 
 
 
 
 
52869cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
497a54e
52869cd
 
 
 
 
 
 
 
497a54e
52869cd
 
 
 
 
 
 
497a54e
52869cd
 
 
 
 
 
 
497a54e
52869cd
 
497a54e
52869cd
497a54e
52869cd
 
 
 
 
 
 
 
 
497a54e
52869cd
 
 
 
 
 
 
 
497a54e
 
 
 
 
 
 
 
 
52869cd
497a54e
 
52869cd
 
497a54e
 
 
 
 
 
52869cd
497a54e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
52869cd
497a54e
 
 
 
 
 
 
52869cd
497a54e
 
 
 
 
 
 
52869cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
497a54e
 
52869cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
497a54e
 
 
 
 
52869cd
 
 
497a54e
 
 
52869cd
 
 
497a54e
52869cd
 
 
 
 
497a54e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
52869cd
 
497a54e
 
 
 
 
 
 
 
52869cd
 
 
 
497a54e
 
52869cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
497a54e
52869cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
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
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
#!/usr/bin/env python
# coding=utf-8
# Copyright 2022 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 sequence to sequence speech recognition
with 🤗 Datasets' streaming mode.
"""
# This progam was modified by Michael Kamfonas (mkamfonas@infokarta.com) on Dec 11 2022
# - added options for drpout, gradient_checkpointing, use_cache, stopping_strategy and streaming
# - restructured it to enable both streaming and non-streaming modes 
# - allows concatenation of mutiple datasets (single-string comma-separated) for interleaving
#     The following params must have the same number of comma-separated (,) elements:
#       dataset_name, 
#       dataset_config_name, 
#       train_split_name and eval_split_name (each element plus-separated (+) for multiple splits), 
#       text_column_name and audio_column_name 
 

import logging
import os
import sys
from dataclasses import dataclass, field
from typing import Any, Dict, List, Optional, Union

import datasets
import torch
from datasets import Audio, DatasetDict, IterableDatasetDict, interleave_datasets, load_dataset
from torch.utils.data import IterableDataset

import evaluate
import transformers
from transformers import (
    AutoConfig,
    AutoFeatureExtractor,
    AutoModelForSpeechSeq2Seq,
    AutoProcessor,
    AutoTokenizer,
    HfArgumentParser,
    Seq2SeqTrainer,
    Seq2SeqTrainingArguments,
    TrainerCallback,
    set_seed,
)
from transformers.trainer_pt_utils import IterableDatasetShard
from transformers.trainer_utils import get_last_checkpoint, is_main_process
from transformers.utils import check_min_version, send_example_telemetry
from transformers.utils.versions import require_version
from transformers.models.whisper.english_normalizer import BasicTextNormalizer

#TEXT_COL_NAME="text"
TEXT_COL_NAME="sentence,transcription"
AUDIO_COL_NAME="audio"

# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
check_min_version("4.25.0.dev0")

require_version("datasets>=1.18.2", "To fix: pip install -r examples/pytorch/speech-recognition/requirements.txt")

logger = logging.getLogger(__name__)


@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"}
    )
    feature_extractor_name: Optional[str] = field(
        default=None, metadata={"help": "feature extractor 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 `huggingface-cli login` (necessary to use this script "
                "with private models)."
            )
        },
    )
    freeze_feature_encoder: bool = field(
        default=True, metadata={"help": "Deprecated - Whether to freeze the feature encoder layers of the model."}
    )
    freeze_encoder: bool = field(
        default=False, metadata={"help": "Whether to freeze the entire encoder of the seq2seq model."}
    )
    forced_decoder_ids: List[List[int]] = field(
        default=None,
        metadata={
            "help": (
                "A list of pairs of integers which indicates a mapping from generation indices to token indices "
                "that will be forced before sampling. For example, [[0, 123]] means the first generated token "
                "will always be a token of index 123."
            )
        },
    )
    suppress_tokens: List[int] = field(
        default=None, metadata={"help": "A list of tokens that will be suppressed at generation."}
    )
    model_index_name: str = field(default=None, metadata={"help": "Pretty name for the model card."})

    ## added by Michael Kamfonas
    use_cache: bool = field(
        default=False, metadata={"help": "Whether to use cache."}
    )

    dropout: float = field(
        default = 0.0, metadata = {"help": "dropout probability."}
    )


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

    dataset_name: 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)."}
    )
    text_column: Optional[str] = field(
        default=None,
        metadata={"help": "The name of the column in the datasets containing the full texts (for summarization)."},
    )
    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."
            )
        },
    )
    max_eval_samples: Optional[int] = field(
        default=None,
        metadata={
            "help": (
                "For debugging purposes or quicker training, truncate the number of evaluation examples to this "
                "value if set."
            )
        },
    )
    audio_column_name: str = field(
        default="audio",
        metadata={"help": "The name of the dataset column containing the audio data. Defaults to 'audio'"},
    )
    text_column_name: str = field(
        default="text",
        metadata={"help": "The name of the dataset column containing the text data. Defaults to 'text'"},
    )
    max_duration_in_seconds: float = field(
        default=20.0,
        metadata={
            "help": (
                "Truncate audio files that are longer than `max_duration_in_seconds` seconds to"
                " 'max_duration_in_seconds`"
            )
        },
    )
    min_duration_in_seconds: float = field(
        default=0.0, metadata={"help": "Filter audio files that are shorter than `min_duration_in_seconds` seconds"}
    )
    train_split_name: str = field(
        default="train",
        metadata={
            "help": "The name of the training data set split to use (via the datasets library). Defaults to 'train'"
        },
    )
    eval_split_name: str = field(
        default="test",
        metadata={
            "help": "The name of the training data set split to use (via the datasets library). Defaults to 'train'"
        },
    )
    do_lower_case: bool = field(
        default=False,
        metadata={"help": "Whether the target text should be lower cased."},
    )
    do_remove_punctuation: bool = field(
        default=False,
        metadata={"help": "Whether the target text should be striped of punctuation."},
    )
    do_normalize_eval: bool = field(
        default=True,
        metadata={"help": "Whether to normalise the references and predictions in the eval WER calculation."},
    )
    language: str = field(
        default=None,
        metadata={
            "help": (
                "Language for multilingual fine-tuning. This argument should be set for multilingual fine-tuning "
                "only. For English speech recognition, it should be set to `None`."
            )
        },
    )
    task: str = field(
        default="transcribe",
        metadata={"help": "Task, either `transcribe` for speech recognition or `translate` for speech translation."},
    )
    shuffle_buffer_size: Optional[int] = field(
        default=500,
        metadata={
            "help": (
                "The number of streamed examples to download before shuffling them. The large the buffer, "
                "the closer it is to real offline shuffling."
            )
        },
    )
    stopping_strategy: Optional[str] = field(
        default="all_exhausted",
        metadata={
             "help": "Strategy used to consume interleaved data. Default = 'all_exhausted'"
        }
    )
    streaming: bool = field(
        default=True,
        metadata={"help": "Whether to use streaming mode to load and pre-process the data."},
    )

@dataclass
class DataCollatorSpeechSeq2SeqWithPadding:
    """
    Data collator that will dynamically pad the inputs received.
    Args:
        processor ([`WhisperProcessor`])
            The processor used for processing the data.
        decoder_start_token_id (`int`)
            The begin-of-sentence of the decoder.
    """

    processor: Any
    decoder_start_token_id: int

    def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]:
        # split inputs and labels since they have to be of different lengths and need
        # different padding methods
        model_input_name = self.processor.model_input_names[0]
        input_features = [{model_input_name: feature[model_input_name]} for feature in features]
        label_features = [{"input_ids": feature["labels"]} for feature in features]

        batch = self.processor.feature_extractor.pad(input_features, return_tensors="pt")

        labels_batch = self.processor.tokenizer.pad(label_features, return_tensors="pt")

        # replace padding with -100 to ignore loss correctly
        labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100)

        # if bos token is appended in previous tokenization step,
        # cut bos token here as it's append later anyways
        if (labels[:, 0] == self.decoder_start_token_id).all().cpu().item():
            labels = labels[:, 1:]

        batch["labels"] = labels

        return batch


def load_streaming_dataset(dataset_name, dataset_config_name, split="train", **kwargs):
    """
    Utility function to load a dataset in streaming mode. For datasets with multiple splits,
    each split is loaded individually and then splits combined by taking alternating examples from
    each (interleaving).
    """
    if "+" in split:
        # load multiple splits separated by the `+` symbol with streaming mode
        dataset_splits = [
            load_dataset(dataset_name, dataset_config_name, split=split_name, streaming=True, **kwargs)
            for split_name in split.split("+")
        ]
        # interleave multiple splits to form one dataset
        interleaved_dataset = interleave_datasets(dataset_splits)
        return interleaved_dataset
    else:
        # load a single split *with* streaming mode
        dataset = load_dataset(dataset_name, dataset_config_name, split=split, streaming=True, **kwargs)
        return dataset

def load_multiple_streaming_datasets(
    dataset_names: List,
    dataset_config_names: List,
    splits: Optional[List] = None,
    text_column_names: Optional[List] = None,
    audio_column_names: Optional[List] = None,
    sampling_rate: Optional[int] = 16000,
    stopping_strategy: Optional[str] = "all_exhausted",
    streaming = True,
    **kwargs
    ):

    if len(dataset_names) != len(dataset_config_names):
        raise ValueError(
            f"Ensure one config is passed for each dataset, got {len(dataset_names)} datasets and"
            f" {len(dataset_config_names)} configs."
        )

    if splits is not None and len(splits) != len(dataset_names):
        raise ValueError(
            f"Ensure one train_split is passed for each dataset, got {len(dataset_names)} datasets and {len(splits)} splits."
        )

    if text_column_names is not None and len(text_column_names) != len(dataset_names):
        raise ValueError(
            f"Ensure one text column name is passed for each dataset, got {len(dataset_names)} datasets and"
            f" {len(text_column_names)} text column names."
        )

    if audio_column_names is not None and len(audio_column_names) != len(dataset_names):
        raise ValueError(
            f"Ensure one text column name is passed for each dataset, got {len(dataset_names)} datasets and"
            f" {len(audio_column_names)} text column names."
        )

    splits = splits if splits is not None \
        else ["train" for i in range(len(dataset_names))]

    text_column_names = (
        text_column_names if text_column_names is not None \
            else [TEXT_COL_NAME for i in range(len(dataset_names))]
    )

    audio_column_names = (
        audio_column_names if audio_column_names is not None \
            else [AUDIO_COL_NAME for i in range(len(dataset_names))]
    )

    all_data_splits = []
    # iterate over the datasets we want to interleave
    for dset, cfgNm, splt,  txtColNm, audColNm in zip(dataset_names,dataset_config_names,\
                                                splits,text_column_names, audio_column_names):

        dset_splits = [load_dataset(dset, cfgNm, split=c, streaming=streaming, **kwargs) \
            for c in splt.split('+') if c != '-']

        if streaming:
            dset_splits = [ds if TEXT_COL_NAME  in ds.features else ds.rename_column(txtColNm, TEXT_COL_NAME) \
                for ds in dset_splits ]
            dset_splits = [ds if AUDIO_COL_NAME in ds.features else ds.rename_column(audColNm, AUDIO_COL_NAME) \
                for ds in dset_splits]

            if len(dset_splits)>0 and sampling_rate != next(iter(dset_splits[0]))[AUDIO_COL_NAME]['sampling_rate']:
                dset_splits = [ds.cast_column(AUDIO_COL_NAME, Audio(sampling_rate)) for ds in dset_splits]
        else:

            dset_splits = [ds if TEXT_COL_NAME  in ds.column_names else ds.rename_column(txtColNm, TEXT_COL_NAME) \
                for ds in dset_splits ]
            dset_splits = [ds if AUDIO_COL_NAME in ds.column_names else ds.rename_column(audColNm, AUDIO_COL_NAME) \
                for ds in dset_splits]
        
        if len(dset_splits)>0 and sampling_rate != next(iter(dset_splits[0]))[AUDIO_COL_NAME]['sampling_rate']:
            dset_splits = [ds.cast_column(AUDIO_COL_NAME, Audio(sampling_rate)) for ds in dset_splits]

        cols2keep = set([AUDIO_COL_NAME, TEXT_COL_NAME])

        dset_splits = [ds.remove_columns(set(ds.features.keys()) - cols2keep) for ds in dset_splits]

        all_data_splits +=   dset_splits

    return interleave_datasets(all_data_splits, stopping_strategy=stopping_strategy)

def main():
    # 1. Parse input arguments
    # 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()

    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
    # information sent is the one passed as arguments along with your Python/PyTorch versions.
    send_example_telemetry("run_speech_recognition_seq2seq_streaming", model_args, data_args)

    # 2. 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)],
    )
    log_level = training_args.get_process_log_level()
    logger.setLevel(log_level)
    datasets.utils.logging.set_verbosity(log_level)
    transformers.utils.logging.set_verbosity(log_level)
    transformers.utils.logging.enable_default_handler()
    transformers.utils.logging.enable_explicit_format()

    logger.setLevel(logging.INFO if is_main_process(training_args.local_rank) else logging.WARN)

    # 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}")

    # Set the verbosity to info of the Transformers logger (on main process only):
    if is_main_process(training_args.local_rank):
        transformers.utils.logging.set_verbosity_info()
    logger.info("Training/evaluation parameters %s", training_args)

    # 3. Detecting last checkpoint and eventually continue from 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."
            )

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

    # 5. Load pretrained model, tokenizer, and feature extractor
    #
    # Distributed training:
    # The .from_pretrained methods guarantee that only one local process can concurrently
    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,
    )

    config.update({ "forced_decoder_ids": model_args.forced_decoder_ids, 
                    "suppress_tokens": model_args.suppress_tokens})

    feature_extractor = AutoFeatureExtractor.from_pretrained(
        model_args.feature_extractor_name if model_args.feature_extractor_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 = AutoModelForSpeechSeq2Seq.from_pretrained(
        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,
    )

    model.config.use_cache = model_args.use_cache
    model.config.dropout = model_args.dropout
    if training_args.gradient_checkpointing:
        model.gradient_checkpointing_enable()

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

    # deprecated
    #if model_args.freeze_feature_encoder:
    #    model.freeze_feature_encoder()

    if model_args.freeze_encoder:
        model.freeze_encoder()
        model.model.encoder.gradient_checkpointing = False

    if data_args.language is not None:
        # We only need to set the task id when the language is specified (i.e. in a multilingual setting)
        tokenizer.set_prefix_tokens(language=data_args.language, task=data_args.task)


    # 4. Load dataset
    raw_datasets = IterableDatasetDict() if data_args.streaming else DatasetDict()

    # if training_args.do_train:
    #     raw_datasets["train"] = load_streaming_dataset(
    #         data_args.dataset_name,
    #         data_args.dataset_config_name,
    #         split=data_args.train_split_name,
    #         use_auth_token=True if model_args.use_auth_token else None,
    #     )

    # if training_args.do_eval:
    #     raw_datasets["eval"] = load_streaming_dataset(
    #         data_args.dataset_name,
    #         data_args.dataset_config_name,
    #         split=data_args.eval_split_name,
    #         use_auth_token=True if model_args.use_auth_token else None,
    #     )

    if training_args.do_train:
        raw_datasets["train"] = load_multiple_streaming_datasets(
            dataset_names=data_args.dataset_name.split(","), 
            dataset_config_names=data_args.dataset_config_name.split(","),
            splits = data_args.train_split_name.split(","),
            text_column_names = data_args.text_column_name.split(","),
            sampling_rate  = feature_extractor.sampling_rate, 
            streaming=data_args.streaming,
            use_auth_token=True if model_args.use_auth_token else None,
        )

    if training_args.do_eval:
        raw_datasets["eval"] = load_multiple_streaming_datasets(
            dataset_names=data_args.dataset_name.split(","), 
            dataset_config_names=data_args.dataset_config_name.split(","),
            splits  = data_args.eval_split_name.split(","),
            text_column_names = data_args.text_column_name.split(","),
            sampling_rate  = feature_extractor.sampling_rate,
            streaming=False,
            use_auth_token=True if model_args.use_auth_token else None,
        )

    raw_datasets_features = list(next(iter(raw_datasets.values())).features.keys())

    if AUDIO_COL_NAME not in raw_datasets_features:
        raise ValueError(
            f"--audio_column_name '{data_args.audio_column_name}' not found in dataset '{data_args.dataset_name}'. "
            "Make sure to set `--audio_column_name` to the correct audio column - one of "
            f"{', '.join(raw_datasets_features)}."
        )

    if TEXT_COL_NAME not in raw_datasets_features:
        raise ValueError(
            f"--text_column_name {TEXT_COL_NAME} not found in dataset. "
            "Make sure to set `--text_column_name` to the the respective correct text columns."
        )


    # 6. Resample speech dataset if necessary
    #dataset_sampling_rate = next(iter(raw_datasets.values())).features[data_args.audio_column_name].sampling_rate
    #if dataset_sampling_rate != feature_extractor.sampling_rate:
    #    raw_datasets = raw_datasets.cast_column(
    #        data_args.audio_column_name, datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate)
    #    )

    # 7. Preprocessing the datasets.
    # We need to read the audio files as arrays and tokenize the targets.
    max_input_length = data_args.max_duration_in_seconds * feature_extractor.sampling_rate
    min_input_length = data_args.min_duration_in_seconds * feature_extractor.sampling_rate
    audio_column_name = AUDIO_COL_NAME
    text_column_name = TEXT_COL_NAME
    model_input_name = feature_extractor.model_input_names[0]
    do_lower_case = data_args.do_lower_case
    do_remove_punctuation = data_args.do_remove_punctuation
    normalizer = BasicTextNormalizer()  # 'official' text normalizer from OpenAI

    if data_args.max_train_samples is not None:
        raw_datasets["train"] = (
            raw_datasets["train"].take(data_args.max_train_samples)
            if data_args.streaming
            else raw_datasets["train"].select(range(data_args.max_train_samples))
        )

    if data_args.max_eval_samples is not None:
        raw_datasets["eval"] = (
            raw_datasets["eval"].take(data_args.max_eval_samples)
            if data_args.streaming
            else raw_datasets["eval"].select(range(data_args.max_eval_samples))
        )

    def prepare_dataset(batch):
        # process audio
        sample = batch[audio_column_name]
        inputs = feature_extractor(sample["array"], sampling_rate=sample["sampling_rate"])
        # process audio length
        batch[model_input_name] = inputs.get(model_input_name)[0]
        batch["input_length"] = len(sample["array"])

        # process targets
        input_str = batch[text_column_name].lower() if do_lower_case else batch[text_column_name]
        if do_remove_punctuation:
            input_str = normalizer(input_str).strip()
        batch["labels"] = tokenizer(input_str).input_ids
        return batch

    with training_args.main_process_first(desc="dataset map pre-processing"):
        vectorized_datasets = raw_datasets.map(
            prepare_dataset,
            remove_columns=raw_datasets_features
        ).with_format("torch")

        if training_args.do_train and data_args.streaming:
            # manually shuffle if streaming (done by the trainer for non-streaming)
            vectorized_datasets["train"] = vectorized_datasets["train"].shuffle(
                buffer_size=data_args.shuffle_buffer_size,
                seed=training_args.seed,
            )

    # filter training data that is shorter than min_input_length or longer than
    # max_input_length
    def is_audio_in_length_range(length):
        return min_input_length < length < max_input_length

    if training_args.do_train:
        vectorized_datasets["train"] = vectorized_datasets["train"].filter(
            is_audio_in_length_range,
            input_columns=["input_length"],
        )

    # 8. Load Metric
    metric = evaluate.load("wer")
    do_normalize_eval = data_args.do_normalize_eval

    def compute_metrics(pred):
        pred_ids = pred.predictions

        pred.label_ids[pred.label_ids == -100] = tokenizer.pad_token_id

        pred_str = tokenizer.batch_decode(pred_ids, skip_special_tokens=True)
        # we do not want to group tokens when computing the metrics
        label_str = tokenizer.batch_decode(pred.label_ids, skip_special_tokens=True)

        if do_normalize_eval:
            pred_str = [normalizer(pred) for pred in pred_str]
            label_str = [normalizer(label) for label in label_str]
            # filtering step to only evaluate the samples that correspond to non-zero references:
            pred_str = [pred_str[i] for i in range(len(pred_str)) if len(label_str[i]) > 0]
            label_str = [label_str[i] for i in range(len(label_str)) if len(label_str[i]) > 0]

        wer = 100 * metric.compute(predictions=pred_str, references=label_str)

        return {"wer": wer}

    # 9. Create a single speech processor
    if is_main_process(training_args.local_rank):
        # save feature extractor, tokenizer and config
        feature_extractor.save_pretrained(training_args.output_dir)
        tokenizer.save_pretrained(training_args.output_dir)
        config.save_pretrained(training_args.output_dir)

    processor = AutoProcessor.from_pretrained(training_args.output_dir)

    # 10. Define data collator
    data_collator = DataCollatorSpeechSeq2SeqWithPadding(
        processor=processor,
        decoder_start_token_id=model.config.decoder_start_token_id,
    )

    # 11. Configure Trainer
    # Trainer callback to reinitialise and reshuffle the streamable datasets at the beginning of each epoch
    # Only required for streaming: Trainer automatically shuffles non-streaming datasets
    class ShuffleCallback(TrainerCallback):
        def on_epoch_begin(self, args, state, control, train_dataloader, **kwargs):
            if isinstance(train_dataloader.dataset, IterableDatasetShard):
                pass  # set_epoch() is handled by the Trainer
            elif isinstance(train_dataloader.dataset, IterableDataset):
                train_dataloader.dataset.set_epoch(train_dataloader.dataset._epoch + 1)

    # Initialize Trainer
    trainer = Seq2SeqTrainer(
        model=model,
        args=training_args,
        train_dataset=vectorized_datasets["train"] if training_args.do_train else None,
        eval_dataset=vectorized_datasets["eval"] if training_args.do_eval else None,
        tokenizer=feature_extractor,
        data_collator=data_collator,
        compute_metrics=compute_metrics if training_args.predict_with_generate else None,
        callbacks=[ShuffleCallback()] if data_args.streaming else None,
    )

    # 12. Training
    if training_args.do_train:
        checkpoint = None
        if training_args.resume_from_checkpoint is not None:
            checkpoint = training_args.resume_from_checkpoint
        elif last_checkpoint is not None:
            checkpoint = last_checkpoint
        train_result = trainer.train(resume_from_checkpoint=checkpoint)
        trainer.save_model()  # Saves the feature extractor too for easy upload

        metrics = train_result.metrics
        if data_args.max_train_samples:
            metrics["train_samples"] = data_args.max_train_samples
        trainer.log_metrics("train", metrics)
        trainer.save_metrics("train", metrics)
        trainer.save_state()

    # 13. Evaluation
    results = {}
    if training_args.do_eval:
        logger.info("*** Evaluate ***")
        metrics = trainer.evaluate(
            metric_key_prefix="eval",
            max_length=training_args.generation_max_length,
            num_beams=training_args.generation_num_beams,
        )
        if data_args.max_eval_samples:
            metrics["eval_samples"] = data_args.max_eval_samples

        trainer.log_metrics("eval", metrics)
        trainer.save_metrics("eval", metrics)

    # 14. Write Training Stats
    kwargs = {
        "finetuned_from": model_args.model_name_or_path,
        "tasks": "automatic-speech-recognition",
        "tags": "whisper-event",
    }
    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"] = f"{data_args.dataset_name} {data_args.dataset_config_name}"
        else:
            kwargs["dataset"] = data_args.dataset_name
        if "common_voice" in data_args.dataset_name:
            kwargs["language"] = data_args.dataset_config_name[:2]
        if model_args.model_index_name is not None:
            kwargs["model_name"] = model_args.model_index_name

    if training_args.push_to_hub:
        trainer.push_to_hub(**kwargs)
    else:
        trainer.create_model_card(**kwargs)

    return results


if __name__ == "__main__":
    main()