run_speech_recognition_seq2seq_streaming.py

#!/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.
"""
# You can also adapt this script for your own sequence to sequence speech
# recognition task. Pointers for this are left as comments.

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

import datasets
import torch
import wandb
from datasets import 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.models.whisper.english_normalizer import BasicTextNormalizer
from transformers.models.whisper.tokenization_whisper import TO_LANGUAGE_CODE, LANGUAGES
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

# 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__)

SENDING_NOTIFICATION = "*** Sending notification to email ***"
RECIPIENT_ADDRESS = "[email protected]"

wandb_token = os.environ.get("WANDB_TOKEN", "None")
hf_token = os.environ.get("HF_TOKEN", None)
if (hf_token is None or wandb_token == "None") and os.path.exists("./creds.txt"):
    with open("./creds.txt", "r") as f:
        lines = f.readlines()
    for line in lines:
        key, value = line.split("=")
        if key == "HF_TOKEN":
            hf_token = value.strip()
        if key == "WANDB_TOKEN":
            wandb_token = value.strip()
        if key == "EMAIL_ADDRESS":
            os.environ["EMAIL_ADDRESS"] = value.strip()
        if key == "EMAIL_PASSWORD":
            os.environ["EMAIL_PASSWORD"] = value.strip()

if hf_token is not None:
    try:
        os.makedirs("/root/.huggingface", exist_ok=True)
        with open("/root/.huggingface/token", "w") as f:
            f.write(hf_token)
        logger.info("Huggingface API key set")
    except (PermissionError, OSError):
        logger.warning("Huggingface API key not set, relying on ~/.huggingface/token")
else:
    logger.warning("Huggingface API key not set, relying on ~/.huggingface/token")

wandb.login(key=wandb_token, relogin=True, timeout=5)
wandb.init(project="whisper", entity="pn-aa")

logger.info("Wandb API key set, logging to wandb")


@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": "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."})


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

    dataset_train_name: str = field(
        default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
    )
    dataset_train_config_name: Optional[str] = field(
        default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
    )
    dataset_eval_name: str = field(
        default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
    )
    dataset_eval_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_train: 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`."
            )
        },
    )
    language_eval: 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."
            )
        },
    )
    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
    task_id: int
    # TODO: remove - infer language from dataset
    language_id: int = -100

    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]
        # lang_features = [f"<|{TO_LANGUAGE_CODE[feature['language']]}|>" 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
        
        # lang_token_ids = self.processor.tokenizer(lang_features).input_ids
        # # Replace language and task if they are in the beginning, otherwise add them
        # if (labels[:, 1] == self.task_id).all().cpu().item():
        #     labels[:, 0] = lang_token_ids
        #     labels[:, 1] = torch.full_like(labels[:, 1], self.task_id)
        # else:
        #     # convert task id to tensor of labels dim to concatenate
        #     task_id = torch.full_like(labels[:, 0], self.task_id)
        #     labels = torch.cat((lang_token_ids, task_id, labels), dim=1)

        # Set language to pad token
        if (labels[:, 0] == self.decoder_start_token_id).all().cpu().item():
            labels[:, 1] = torch.full_like(labels[:, 1], -100)
        # labels[:, 0] = torch.full_like(labels[:, 0], -100)
        # labels[:, 1] = torch.full_like(labels[:, 1], -100)

        # remove start of sentence token from labels
        # if (labels[:, 0] == self.decoder_start_token_id).all().cpu().item():
        #     labels = labels[:, 1:]

        # # add start of sentence token to labels + language + task
        # labels = torch.cat((torch.full_like(labels[:, 0], self.task_id).unsqueeze(0).T, labels), dim=-1)
        # labels = torch.cat((torch.full_like(labels[:, 0], self.language_id).unsqueeze(0).T, labels), dim=-1)
        # labels = torch.cat((torch.full_like(labels[:, 0], self.decoder_start_token_id).unsqueeze(0).T, labels), dim=-1)

        batch["labels"] = labels

        return batch


def notify_me(recipient, message=None):
    """
    Send an email to the specified address with the specified message
    """
    sender = os.environ.get("EMAIL_ADDRESS", None)
    password = os.environ.get("EMAIL_PASSWORD", None)
    if sender is None:
        logging.warning("No email address specified, not sending notification")
    if password is None:
        logging.warning("No email password specified, not sending notification")
    if message is None:
        message = "Training is finished!"

    if sender is not None:
        import smtplib
        from email.mime.text import MIMEText

        msg = MIMEText(message)
        msg["Subject"] = "Training updates..."
        msg["From"] = "[email protected]"
        msg["To"] = recipient

        # send the email
        smtp_obj = smtplib.SMTP("smtp.gmail.com", 587)
        smtp_obj.starttls()
        smtp_obj.login(sender, password)
        smtp_obj.sendmail(sender, recipient, msg.as_string())
        smtp_obj.quit()


def rename_col_and_resample(dataset, dataset_name, text_column_names, text_col_name_ref, audio_column_name, sampling_rate):
    raw_datasets_features = list(dataset.features.keys())
    logger.info(f"Dataset {dataset_name} - Features: {raw_datasets_features}")
                
    if text_col_name_ref not in raw_datasets_features:
        if len(text_column_names) == 1:
            raise ValueError("None of the text column names provided found in dataset."
                                f"Text columns: {text_column_names}"
                                f"Dataset columns: {raw_datasets_features}")
        flag = False
        for text_column_name in text_column_names:
            if text_column_name in raw_datasets_features:
                logger.info(f"Renaming text column {text_column_name} to {text_col_name_ref}")
                dataset = dataset.rename_column(text_column_name, text_col_name_ref)
                flag = True
                break
        if flag is False:
            raise ValueError("None of the text column names provided found in dataset."
                                f"Text columns: {text_column_names}"
                                f"Dataset columns: {raw_datasets_features}")
    if audio_column_name is not None and sampling_rate is not None:
        ds_sr = int(dataset.features[audio_column_name].sampling_rate)
        if ds_sr != sampling_rate:
            dataset = dataset.cast_column(
                audio_column_name, datasets.features.Audio(sampling_rate=sampling_rate)
            )
    
    raw_datasets_features = list(dataset.features.keys())
    raw_datasets_features.remove(audio_column_name)
    raw_datasets_features.remove(text_col_name_ref)
    # Keep only audio and sentence
    dataset = dataset.remove_columns(column_names=raw_datasets_features)
    return dataset


def load_maybe_streaming_dataset(
    dataset_names,
    dataset_config_names,
    split="train",
    streaming=True,
    audio_column_name=None,
    sampling_rate=None,
    **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).
    """
    text_column_names = None
    if "text_column_name" in kwargs:
        text_column_names = kwargs.pop("text_column_name").split(",")
        text_col_name_ref = text_column_names[0]

    if "," in dataset_names or "+" in split:
        # load multiple splits separated by the `+` symbol with streaming mode
        dataset_splits = []
        for dataset_name, dataset_config_name, split_names in zip(
            dataset_names.split(","), dataset_config_names.split(","), split.split(",")
        ):
            for split_name in split_names.split("+"):
                if dataset_config_name:
                    dataset = load_dataset(dataset_name, dataset_config_name, split=split_name, streaming=streaming, **kwargs)
                else:
                    dataset = load_dataset(dataset_name, split=split_name, streaming=streaming, **kwargs)              
                
                dataset = rename_col_and_resample(
                    dataset,
                    dataset_name,
                    text_column_names,
                    text_col_name_ref,
                    audio_column_name,
                    sampling_rate
                )

                dataset_splits.append(dataset)

        # interleave multiple splits to form one dataset
        interleaved_dataset = interleave_datasets(dataset_splits, stopping_strategy="all_exhausted")
        return interleaved_dataset
    else:
        # load a single split *with* streaming mode
        
        dataset = load_dataset(dataset_names, dataset_config_names, split=split, streaming=streaming, **kwargs)
        dataset = rename_col_and_resample(
            dataset,
            dataset_names,
            text_column_names,
            text_col_name_ref,
            audio_column_name,
            sampling_rate
        )
        return dataset


def print_data_samples(dataset, tokenizer, max_samples=5):
    shown_samples = 0
    for batch in dataset:
        print("Target: ", tokenizer.decode(batch["labels"]))
        shown_samples += len(batch)
        if shown_samples >= max_samples:
            break


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.
    logger.info("*** Parse args ***")
    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
    logger.info("*** 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)

    # Load feature extractor
    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=hf_token if model_args.use_auth_token else None,
    )

    # 4. Load dataset
    logger.info("*** Load dataset ***")
    raw_datasets = IterableDatasetDict() if data_args.streaming else DatasetDict()

    if len(data_args.language_eval.split(",")) > 1:
        raise ValueError("Implementation does not support multiple language evaluation.")

    if training_args.do_train:
        raw_datasets["train"] = load_maybe_streaming_dataset(
            data_args.dataset_train_name,
            data_args.dataset_train_config_name,
            split=data_args.train_split_name,
            use_auth_token=hf_token if model_args.use_auth_token else None,
            streaming=data_args.streaming,
            text_column_name=data_args.text_column_name,
            audio_column_name=data_args.audio_column_name,
            sampling_rate=int(feature_extractor.sampling_rate),
            # language=data_args.language_train
        )

    if training_args.do_eval:
        raw_datasets["eval"] = load_maybe_streaming_dataset(
            data_args.dataset_eval_name,
            data_args.dataset_eval_config_name,
            split=data_args.eval_split_name,
            use_auth_token=hf_token if model_args.use_auth_token else None,
            streaming=data_args.streaming,
            text_column_name=data_args.text_column_name,
            audio_column_name=data_args.audio_column_name,
            sampling_rate=int(feature_extractor.sampling_rate),
            # language=data_args.language_eval
        )

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

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

    data_args.text_column_name = data_args.text_column_name.split(",")[0]
    if data_args.text_column_name not in raw_datasets_features:
        raise ValueError(
            f"--text_column_name {data_args.text_column_name} not found in dataset. "
            "Make sure to set `--text_column_name` to the correct text column - one of "
            f"{', '.join(raw_datasets_features)}."
        )

    # 5. Load pretrained model, tokenizer, and feature extractor
    logger.info("*** 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=hf_token if model_args.use_auth_token else None
    )

    # Forced decoder ids will be overwritten before evaluation 
    config.update({"forced_decoder_ids": model_args.forced_decoder_ids, "suppress_tokens": model_args.suppress_tokens})

    if training_args.gradient_checkpointing:
        config.update({"use_cache": False})

    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=hf_token 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=hf_token if model_args.use_auth_token else None,
    )

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

    if model_args.freeze_feature_encoder:
        model.freeze_feature_encoder()

    if model_args.freeze_encoder:
        model.freeze_encoder()
    
    tokenizer.set_prefix_tokens(language="swedish", task=data_args.task)

    # if data_args.language_train is not None and len(data_args.language_train.split(",")) == 1:
    #     # We only need to set the task id when the language is specified (i.e. in a multilingual setting)
    #     # If more than a langugae is specified, it will be specified in the data collator
    #     tokenizer.set_prefix_tokens(language=data_args.language_train, task=data_args.task)
    # elif data_args.language_train is not None and len(data_args.language_train.split(",")) > 1:
    #     # make sure language and task are not stored in the model config
    #     model.config.forced_decoder_ids = None

    # 6. Resample speech dataset if necessary
    # logger.info("*** Resample dataset ***")
    # dataset_sampling_rate = next(iter(raw_datasets.values())).features[data_args.audio_column_name].sampling_rate
    # if dataset_sampling_rate != feature_extractor.sampling_rate:
    

    # 7. Preprocessing the datasets.
    # We need to read the audio files as arrays and tokenize the targets.
    logger.info("*** Preprocess dataset ***")
    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 = data_args.audio_column_name
    text_column_name = data_args.text_column_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"):
        # raw_datasets_features.remove("language")
        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
    logger.info("*** 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
    logger.info("*** Init 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
    task_token = data_args.task
    if not task_token.startswith('<|'):
        task_token = f'<{task_token}>'
    task_id = tokenizer(task_token).input_ids[0]
    data_collator = DataCollatorSpeechSeq2SeqWithPadding(
        processor=processor,
        decoder_start_token_id=model.config.decoder_start_token_id,
        task_id=task_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
    logger.info("*** Set shuffle callback ***")
    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
    logger.info("*** Init 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,
    )
    logger.info("*** Trainer initialized ***")

    orig_push_to_hub = trainer.args.push_to_hub 
    trainer.args.push_to_hub = False

    # 12. Training
    if training_args.do_train:
        logger.info("*** Train ***")
        print_data_samples(vectorized_datasets["train"], tokenizer)
        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)
        logger.info("*** Training completed ***")
        logger.info("*** Saving model ***")
        # We don't want to push the model to the hub now
        # so we temporarily set to false the push_to_hub attribute
        # and then reset it to the original value
        trainer.save_model()  # Saves the feature extractor too for easy upload
        logger.info("*** Model saved ***")
        metrics = train_result.metrics
        if data_args.max_train_samples:
            metrics["train_samples"] = data_args.max_train_samples
        logger.info("*** Logging metrics ***")
        trainer.log_metrics("train", metrics)
        logger.info("*** Metrics logged ***")
        logger.info("*** Saving metrics ***")
        trainer.save_metrics("train", metrics)
        logger.info("*** Metrics saved ***")
        logger.info("*** Saving state ***")
        trainer.save_state()
        logger.info("*** State saved ***")

    # Run a test prediction to check outputs
    predictions = trainer.predict(
        test_dataset=vectorized_datasets["eval"].shuffle(seed=training_args.seed).take(5),
        metric_key_prefix="test",
        max_length=training_args.generation_max_length,
        num_beams=training_args.generation_num_beams,
    )
    logger.info("*** Test prediction done ***")
    preds = tokenizer.batch_decode(predictions.predictions)
    labels = tokenizer.batch_decode(predictions.label_ids)
    pred_labels = [f"Prediction: {pred}\nLabel: {label}\n" for pred, label in zip(preds, labels)]
    logger.info("Before setting language and task")
    logger.info(f"{pred_labels}")
    language_name = LANGUAGES[data_args.language_eval]
    trainer.model.config.forced_decoder_ids = \
        tokenizer.get_decoder_prompt_ids(language=language_name, task=data_args.task, no_timestamps=True)
    preds = tokenizer.batch_decode(predictions.predictions)
    labels = tokenizer.batch_decode(predictions.label_ids)
    pred_labels = [f"Prediction: {pred}\nLabel: {label}\n" for pred, label in zip(preds, labels)]
    logger.info("After setting language and task")
    logger.info(f"{pred_labels}")

    # 13. Evaluation
    results = {}
    if training_args.do_eval:
        logger.info("*** Evaluate ***")
        print_data_samples(vectorized_datasets["eval"], tokenizer)
        metrics = trainer.evaluate(
            metric_key_prefix="eval",
            max_length=training_args.generation_max_length,
            num_beams=training_args.generation_num_beams,
        )
        logger.info("*** Evaluation done ***")
        if data_args.max_eval_samples:
            metrics["eval_samples"] = data_args.max_eval_samples
        logger.info("*** Logging metrics ***")
        trainer.log_metrics("eval", metrics)
        logger.info("*** Metrics logged ***")
        logger.info("*** Saving metrics ***")
        trainer.save_metrics("eval", metrics)
        logger.info("*** Metrics saved ***")

    # 14. Write Training Stats
    logger.info("*** Writing training stats ***")
    kwargs = {
        "finetuned_from": model_args.model_name_or_path,
        "tasks": "automatic-speech-recognition",
        "tags": "whisper-event",
    }
    if data_args.dataset_train_name is not None:
        dataset_names = list(data_args.dataset_train_name.split(","))
        kwargs["dataset_tags"] = dataset_names
        # if data_args.dataset_train_config_name is not None:
        #     dataset_config_names = list(data_args.dataset_train_config_name.split(","))
        #     dataset_config_names_list = [f"{ds_name} {ds_cfg_name}" for ds_name, ds_cfg_name in zip(dataset_names, dataset_config_names)]
        # else:
        #     dataset_config_names_list = dataset_names
        # kwargs["dataset"] = "\n".join(dataset_config_names_list)
        # if "common_voice" in data_args.dataset_name:
        #     kwargs["language"] = data_args.dataset_config_name[:2]
        if data_args.language_train is not None:
            languages = list(set(data_args.language_train.split(",")))
            kwargs["language"] = languages
        if model_args.model_index_name is not None:
            kwargs["model_name"] = model_args.model_index_name

    logger.info("*** Training stats written ***")
    logger.info(json.dumps(kwargs, indent=4))

    # Training complete notification
    logger.info("*** Training and eval complete ***")
    logger.info(SENDING_NOTIFICATION)
    with open(os.path.join(training_args.output_dir, "train_results.json"), "r") as f:
            train_results = json.load(f)
    with open(os.path.join(training_args.output_dir, "eval_results.json"), "r") as f:
            eval_results = json.load(f)
    notify_me(recipient=RECIPIENT_ADDRESS,
              message=f"Training complete! {train_results = } {eval_results = }")
    
    trainer.args.push_to_hub = orig_push_to_hub
    if training_args.push_to_hub:
        logger.info("*** Pushing to hub ***")
        trainer.push_to_hub(**kwargs)
        logger.info("*** Pushed to hub ***")
        logger.info(SENDING_NOTIFICATION)
    else:
        logger.info("*** Creating model card ***")
        trainer.create_model_card(**kwargs)
        logger.info("*** Model card created ***")
        logger.info(SENDING_NOTIFICATION)

    with open(os.path.join(training_args.output_dir, "README.md"), "r") as f:
        readme = f.read()
    notify_me(recipient=RECIPIENT_ADDRESS,
              message=f"Model pushed to hub! {readme = }")

    return results


if __name__ == "__main__":
    main()