run_flax_speech_recognition_seq2seq_streaming.py

#!/usr/bin/env python
# coding=utf-8
# Original code Copyright 2023 The HuggingFace Inc. team. All rights reserved.
# Additions and modifications Copyright 2023 National Library of Norway. All rights reserved.
#
# This code is based on the original script developed by HuggingFace Inc.
# Substantial additions and modifications have been made by the AiLab at the
# National Library of Norway, with contributions from Per Egil Kummervold
# and Javier de la Rosa, including TPU Pod support, Dataset Streaming, 
# performance enhancements, and support for new features.
#
# 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 Flax library models for sequence to sequence speech recognition.
"""
# You can also adapt this script on your own sequence to sequence task. Pointers for this are left as comments.

import os
import itertools
import json
import logging
import shutil
import socket
import sys
import tempfile
import time
from dataclasses import field
from datetime import datetime
from functools import partial
from importlib import import_module
from pathlib import Path
from typing import Any, Callable, Dict, Generator, List, Optional, Union

import flax
import jax
import jax.numpy as jnp 
import numpy as np
import optax
import pandas as pd
import torch
from flax import jax_utils, traverse_util
from flax.jax_utils import pad_shard_unpad, unreplicate
from flax.training import train_state
from flax.training.common_utils import get_metrics, onehot, shard, shard_prng_key
from torch.utils.data import IterableDataset
from tqdm import tqdm

import datasets
import evaluate
import transformers
from datasets import Dataset, DatasetDict, IterableDatasetDict, interleave_datasets, load_dataset
from datasets.distributed import split_dataset_by_node
from huggingface_hub import Repository, create_repo
from transformers import (
    AutoConfig,
    AutoFeatureExtractor,
    AutoProcessor,
    AutoTokenizer,
    FlaxAutoModelForSpeechSeq2Seq,
    HfArgumentParser,
    Seq2SeqTrainingArguments,
    is_tensorboard_available,
)
from transformers.modelcard import TrainingSummary
from transformers.models.whisper.english_normalizer import BasicTextNormalizer
from transformers.models.whisper.tokenization_whisper import TO_LANGUAGE_CODE
from transformers.file_utils import get_full_repo_name
from transformers.utils import check_min_version, send_example_telemetry
from transformers.utils.versions import require_version

from flax.training import checkpoints

# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
check_min_version("4.27.0.dev0")

require_version("datasets>=1.18.2",
                "To fix: pip install datasets>=1.18.2")


logger = logging.getLogger(__name__)


@flax.struct.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 `transformers-cli login` (necessary to use this script "
            "with private models)."
        },
    )
    dtype: Optional[str] = field(
        default="float32",
        metadata={
            "help": (
                "Floating-point format in which the model weights should be initialized and trained. Choose one of"
                " `[float32, float16, bfloat16]`."
            )
        },
    )
    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 evaluation."
            )
        },
    )
    dropout: Optional[float] = field(
        default=None, metadata={"help": "The dropout ratio for the dropout layer probabilities."}
    )
    attention_dropout: Optional[float] = field(
        default=None, metadata={"help": "The dropout ratio for the attention probabilities."}
    )
    activation_dropout: Optional[float] = field(
        default=None, metadata={"help": "The dropout ratio for activations inside the fully connected layer."}
    )
    encoder_dropout: Optional[float] = field(
        default=None, metadata={"help": "The dropout ratio for the encoder layer dropout probabilities."}
    )
    decoder_dropout: Optional[float] = field(
        default=None, metadata={"help": "The dropout ratio for the decoder layer dropout probabilities."}
    )

@flax.struct.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)."},
    )
    dataset_cache_dir: Optional[str] = field(
        default=None, metadata={"help": "Path to cache directory for saving and loading datasets"}
    )
    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_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."
        },
    )
    max_predict_samples: Optional[int] = field(
        default=None,
        metadata={
            "help": "Truncate the number of prediction examples (test set) 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=30.0,
        metadata={
            "help": "Filter audio files that are longer than `max_duration_in_seconds` seconds"},
    )
    min_duration_in_seconds: float = field(
        default=0.0,
        metadata={
            "help": "Filter audio files that are shorter than `min_duration_in_seconds` seconds"},
    )
    max_label_length: Optional[int] = field(
        default=256,
        metadata={
            "help": "Truncate transcriptions that are longer `max_label_length` tokens."},
    )
    pad_input_to_multiple_of: Optional[int] = field(
        default=None,
        metadata={
            "help": "If set will pad the input sequence to a multiple of the provided value. "
            "This is important to avoid triggering recompilations on TPU. If unspecified, will default to padding the inputs to max length."
        },
    )
    pad_target_to_multiple_of: Optional[int] = field(
        default=None,
        metadata={
            "help": "If set will pad the target sequence to a multiple of the provided value. "
            "This is important to avoid triggering recompilations on TPU. If unspecified, will default to padding the targets to max length."
        },
    )
    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="validation",
        metadata={
            "help": "The name of the evaluation data set split to use (via the datasets library). Defaults to 'validation'"
        },
    )
    test_split_name: str = field(
        default="test",
        metadata={
            "help": "The name of the prediction data set split to use (via the datasets library). Defaults to 'test'"
        },
    )
    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."},
    )
    num_train_steps: int = field(
        default=50000,
        metadata={
            "help": "The number of training steps."
        },
    )
    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."},
    )


def shift_tokens_right(label_ids: np.array, decoder_start_token_id: int) -> np.ndarray:
    """
    Shift label ids one token to the right.
    """
    shifted_label_ids = np.zeros_like(label_ids)
    shifted_label_ids[:, 1:] = label_ids[:, :-1]
    shifted_label_ids[:, 0] = decoder_start_token_id

    return shifted_label_ids


@flax.struct.dataclass
class FlaxDataCollatorSpeechSeq2SeqWithPadding:
    """
    Data collator that will dynamically pad the inputs received.
    Args:
        processor ([`Wav2Vec2Processor`])
            The processor used for proccessing the data.
        decoder_start_token_id (:obj: `int`)
            The begin-of-sentence of the decoder.
        input_padding (:obj:`bool`, :obj:`str` or :class:`~transformers.tokenization_utils_base.PaddingStrategy`, `optional`, defaults to :obj:`True`):
            Select a strategy to pad the returned input sequences (according to the model's padding side and padding index)
            among:
            * :obj:`True` or :obj:`'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
              sequence if provided).
            * :obj:`'max_length'`: Pad to a maximum length specified with the argument :obj:`max_length` or to the
              maximum acceptable input length for the model if that argument is not provided.
            * :obj:`False` or :obj:`'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of
              different lengths).
        target_padding (:obj:`bool`, :obj:`str` or :class:`~transformers.tokenization_utils_base.PaddingStrategy`, `optional`, defaults to :obj:`True`):
            Select a strategy to pad the returned target sequences (according to the model's padding side and padding index).
            See above for details.
        max_input_length (:obj:`float`, `optional`):
            Maximum length of the ``input_values`` of the returned list and optionally padding length (see above).
        max_target_length (:obj:`int`, `optional`):
            Maximum length of the ``labels`` of the returned list and optionally padding length (see above).
        pad_input_to_multiple_of (:obj:`int`, `optional`):
            If set will pad the input sequence to a multiple of the provided value.
            This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability >=
            7.5 (Volta).
        pad_target_to_multiple_of (:obj:`int`, `optional`):
            If set will pad the target sequence to a multiple of the provided value.
            This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability >=
            7.5 (Volta).
    """

    processor: Any
    decoder_start_token_id: int
    input_padding: Union[bool, str] = "longest"
    target_padding: Union[bool, str] = "max_length"
    max_input_length: Optional[float] = None
    max_target_length: Optional[int] = None
    pad_input_to_multiple_of: Optional[int] = None
    pad_target_to_multiple_of: Optional[int] = None
    forward_attention_mask: Optional[bool] = False

    def __call__(self, features: List[Dict[str, Union[List[int], np.ndarray]]]) -> Dict[str, np.ndarray]:
        model_input_name = self.processor.model_input_names[0]
        input_features = {model_input_name: features[model_input_name]}
        label_features = {"input_ids": features["labels"]}

        # reformat list to dict and set to pytorch format
        batch = self.processor.feature_extractor.pad(
            input_features,
            max_length=self.max_input_length,
            padding=self.input_padding,
            pad_to_multiple_of=self.pad_input_to_multiple_of,
            return_tensors="np",
        )

        labels_batch = self.processor.tokenizer.pad(
            label_features,
            max_length=self.max_target_length,
            padding=self.target_padding,
            pad_to_multiple_of=self.pad_target_to_multiple_of,
            return_tensors="np",
        )

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

        # replace padding with -100 to ignore correctly when computing the loss
        labels = np.ma.array(labels, mask=np.not_equal(
            labels_batch.attention_mask, 1))
        labels = labels.filled(fill_value=-100)

        batch["labels"] = labels
        batch["decoder_input_ids"] = decoder_input_ids
        if self.forward_attention_mask:
            batch["attention_mask"] = np.array([feature["attention_mask"] for feature in features])

        return batch


def collate_batch(samples: List) -> Dict[str, List]:
    """
    Collate function to transform a dataset batch into a format suitable for Torch
    """
    return {key: [feature[key] for feature in samples] for key in samples[0]}


def data_loader(
    dataset: Dataset,
    batch_size: int,
    drop_last: bool=True,
    num_workers: int=0,
) -> Generator:
    """
    Returns batches of size `batch_size` from `dataset`. If `drop_last` is set to `False`, the final batch may be incomplete,
    and range in size from 1 to `batch_size`.
    """
    data_loader_iterator = iter(torch.utils.data.DataLoader(
        batch_size=batch_size,
        dataset=dataset.with_format("numpy"),
        num_workers=num_workers,
        collate_fn=collate_batch,
        drop_last=drop_last,
    ))
    return data_loader_iterator


class TrainState(train_state.TrainState):
    dropout_rng: jnp.ndarray

    def replicate(self):
        return jax_utils.replicate(self).replace(dropout_rng=shard_prng_key(self.dropout_rng))


def create_learning_rate_fn(
    num_train_steps: int, num_warmup_steps: int, learning_rate: float, start_step: int=0, warmup_init_value: float=0.0, decay_end_value: float=0.0,
) -> Callable[[int], jnp.array]:
    """
    Returns a linear warmup, linear_decay learning rate function. `start_step` is an optional parameter (defaults to 0) to
    shift the learning rate schedule by a specified number of steps."""
    warmup_fn = optax.linear_schedule(
        init_value=warmup_init_value, end_value=learning_rate, transition_steps=num_warmup_steps)
    decay_fn = optax.linear_schedule(
        init_value=learning_rate, end_value=decay_end_value, transition_steps=num_train_steps - num_warmup_steps
    )
    schedule_fn = optax.join_schedules(
        schedules=[warmup_fn, decay_fn], boundaries=[num_warmup_steps])
    
    def learning_rate_fn(step: int) -> Callable[[int], jnp.array]:
        return schedule_fn(step + start_step)
    
    return learning_rate_fn


def main():
    # 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 JAX/Flax versions.
    send_example_telemetry("run_speech_recognition_seq2seq",
                           model_args, data_args, framework="flax")

    # Setup logging
    # Make one log on every process with the configuration for debugging.
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
        datefmt="%m/%d/%Y %H:%M:%S",
        handlers=[logging.StreamHandler(sys.stdout)],
    )

    # Number of hosts
    num_of_hosts = jax.process_count()
    current_host_idx = jax.process_index()

    if current_host_idx == 0:
        datasets.utils.logging.set_verbosity_warning()
        transformers.utils.logging.set_verbosity_info()
    else:
        datasets.utils.logging.set_verbosity_error()
        transformers.utils.logging.set_verbosity_error()
    
    logger.setLevel(logging.INFO)
    logger.info("Training/evaluation parameters %s", training_args)

    if num_of_hosts and not training_args.push_to_hub:
        logger.warning(
            f"If you are on a TPU Pod or a multinode setup, it is impossible to recover from local checkpoints. You will therefore need to set `--push_to_hub` to save checkpoints and be able to restore a training."
        )
    if num_of_hosts and not training_args.overwrite_output_dir and training_args.resume_from_checkpoint:
        logger.error(
            f"If you are on a TPU Pod or a multinode setup, you need to set --overwrite_output_dir to be able to resume from a pushed checkpoint."
        )
        sys.exit(1)

    # Check the output dir is valid
    if os.path.exists(training_args.output_dir):
        if (
            os.listdir(training_args.output_dir)
            and training_args.do_train
            and not training_args.overwrite_output_dir
        ):
            raise ValueError(
                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
                "Use `--overwrite_output_dir` to overcome."
            )
        elif training_args.overwrite_output_dir:
            logger.warning(f"Removing path {training_args.output_dir}")
            shutil.rmtree(training_args.output_dir)
      
    # Handle the repository creation
    output_dir = Path(training_args.output_dir)
    repo_name = ""
    if training_args.push_to_hub:
        if training_args.hub_model_id is None:
            repo_name = get_full_repo_name(
                output_dir.absolute().name,
                token=training_args.hub_token,
                organization=training_args.push_to_hub_organization,
            )
        else:
            repo_name = training_args.hub_model_id

        repo_url = None  
        while not repo_url:
            # Workaround for an internal HuggingFace error if the repo is being created by another worker
            try:
                repo_url = create_repo(
                    repo_name, exist_ok=True, token=training_args.hub_token, private=training_args.hub_private_repo
                )
            except:
                logger.info(f"Waiting for the creation of repository '{repo_name}'. Make sure you are logged in.")
                time.sleep(1)

        repo = Repository(training_args.output_dir,
                          clone_from=repo_name, token=training_args.hub_token)

    # Set the model_name_or_path
    model_name_or_path = model_args.model_name_or_path

    # Try to detect last checkpoint and continue if possible
    training_state = {"step": 0, "eval_lines": []}
    if training_args.resume_from_checkpoint:
        if (output_dir / "flax_model.msgpack").exists() and (output_dir / "training_state.bin").exists():
            training_state = json.loads((output_dir / "training_state.bin").read_text())
            model_name_or_path = os.path.join(training_args.output_dir)
            logger.info(
                f"Checkpoint detected, resuming training from {training_args.output_dir} at step {training_state['step']}."
            )
        else:
            logger.info(
                f"No valid checkpoint found in {training_args.output_dir}. Starting from {model_name_or_path}."
            )
    
    
    # Load dataset
    raw_datasets = IterableDatasetDict() if data_args.streaming else DatasetDict()

    if training_args.do_train:
        raw_datasets["train"] = load_dataset(
            data_args.dataset_name,
            data_args.dataset_config_name,
            split=data_args.train_split_name,
            cache_dir=data_args.dataset_cache_dir,
            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_dataset(
            data_args.dataset_name,
            data_args.dataset_config_name,
            split=data_args.eval_split_name,
            cache_dir=data_args.dataset_cache_dir,
            streaming=data_args.streaming,
            use_auth_token=True if model_args.use_auth_token else None,
        )

    if training_args.do_predict:
        raw_datasets["test"] = load_dataset(
            data_args.dataset_name,
            data_args.dataset_config_name,
            split=data_args.test_split_name,
            cache_dir=data_args.dataset_cache_dir,
            streaming=data_args.streaming,
            use_auth_token=True if model_args.use_auth_token else None,
        )

    if not training_args.do_train and not training_args.do_eval and not training_args.do_predict:
        raise ValueError(
            "There is nothing to do. Please pass `do_train`, `do_eval` and/or `do_predict`."
        )

    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 '{data_args.dataset_name}'. "
            "Make sure to set `--audio_column_name` to the correct audio column - one of "
            f"{', '.join(raw_datasets_features)}."
        )

    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 '{data_args.dataset_name}'. "
            "Make sure to set `--text_column_name` to the correct text column - one of "
            f"{', '.join(raw_datasets_features)}."
        )

    # Load pretrained model, tokenizer, and feature extractor
    config = AutoConfig.from_pretrained(
        model_args.config_name if model_args.config_name else 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,
    )
       
    # Update config with arguments. Use values set by model_args if they are not None, otherwise use values from config
    config.update({
        "dropout": model_args.dropout or getattr(config, "dropout", 0.0),
        "attention_dropout": model_args.attention_dropout or getattr(config, "attention_dropout", 0.0),
        "activation_dropout": model_args.activation_dropout or getattr(config, "activation_dropout", 0.0),
        "decoder_layerdrop": model_args.decoder_dropout or getattr(config, "decoder_dropout", 0.0),
        "encoder_layerdrop": model_args.encoder_dropout or getattr(config, "encoder_dropout", 0.0),
    })
    
    feature_extractor = AutoFeatureExtractor.from_pretrained(
        model_args.feature_extractor_name if model_args.feature_extractor_name else 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_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 = FlaxAutoModelForSpeechSeq2Seq.from_pretrained(
        model_name_or_path,
        config=config,
        dtype=getattr(jnp, model_args.dtype),
        cache_dir=model_args.cache_dir,
        revision=model_args.model_revision,
        use_auth_token=True if model_args.use_auth_token else None,
    )

    logger.info(
        f"Successfully loaded the model '{model_name_or_path}'."
    )
    
    if model.config.decoder_start_token_id is None:
        raise ValueError(
            "Make sure that `config.decoder_start_token_id` is correctly defined")

    # Resample speech dataset: `datasets` takes care of automatically loading and resampling the audio,
    # so we just need to set the correct target sampling rate.
    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)
        )

    # Preprocessing the datasets.
    # We need to read the audio files as arrays and tokenize the targets.
    max_input_length = int(
        data_args.max_duration_in_seconds * feature_extractor.sampling_rate)
    min_input_length = int(
        data_args.min_duration_in_seconds * feature_extractor.sampling_rate)
    max_label_length = (
        data_args.max_label_length if data_args.max_label_length is not None else model.config.max_length
    )
    pad_input_to_multiple_of = data_args.pad_input_to_multiple_of
    pad_target_to_multiple_of = data_args.pad_target_to_multiple_of
    audio_column_name = data_args.audio_column_name
    num_workers = training_args.dataloader_num_workers
    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 SpecAugment is used for whisper models, return attention_mask to guide the mask along time axis
    forward_attention_mask = (
        getattr(config, "model_type", None) == "whisper"
        and getattr(config, "apply_spec_augment", False)
        and getattr(config, "mask_time_prob", 0) > 0
    )

    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)
    
    if training_args.do_train and 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 training_args.do_eval and 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))
        )

    if training_args.do_predict and data_args.max_predict_samples is not None:
        raw_datasets["test"] = (
            raw_datasets["test"].take(data_args.max_predict_samples)
            if data_args.streaming
            else raw_datasets["test"].select(range(data_args.max_predict_samples))
        )


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

        # 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, truncation=True, max_length=max_label_length).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,
        )

    # Filter training data with inputs 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"],
        )

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

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

    # Load metrics and write stats
    metric_wer = evaluate.load("wer")
    metric_cer = evaluate.load("cer")
    do_normalize_eval = data_args.do_normalize_eval

    def compute_metrics(pred_ids, label_ids):
        # Replace padded labels by the padding token
        for idx in range(len(label_ids)):
            label_ids[idx][label_ids[idx] == -100] = tokenizer.pad_token_id

        predictions = tokenizer.batch_decode(pred_ids, skip_special_tokens=True)
        # We do not want to group tokens when computing the metrics
        labels = tokenizer.batch_decode(label_ids, skip_special_tokens=True)

        if do_normalize_eval:
            pred_str = [normalizer(pred) for pred in predictions]
            label_str = [normalizer(label) for label in labels]
            # 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]
        else:
            pred_str = predictions
            label_str = labels

        wer = 100 * metric_wer.compute(predictions=pred_str, references=label_str)
        cer = 100 * metric_cer.compute(predictions=pred_str, references=label_str)
        return {"wer": wer, "cer": cer}

    def update_training_state(training_state, train_metrics, eval_metrics, step):
        safe_value = lambda x: float(x.tolist() if isinstance(x, jnp.ndarray) else x)
        state = {"step": step}
        eval_lines = training_state["eval_lines"]
       
        train_metrics = get_metrics(train_metrics)
        train_metrics_dict = {}
        for metric_name, values in train_metrics.items():
            tag = f"train_{metric_name}"
            for i, value in enumerate(values):
                train_metrics_dict[step - len(values) + i + 1] = {tag: safe_value(value)}

        eval_metrics_dict = {}
        for metric_name, value in eval_metrics.items():
            tag = f"eval_{metric_name}"
            eval_metrics_dict.update({
                "step": step,
                tag: safe_value(value),
            })
            if step in train_metrics_dict:
                eval_metrics_dict.update(train_metrics_dict[step])
        eval_lines.append(eval_metrics_dict)
        return {**state, "eval_lines": eval_lines}

    def write_metric(summary_writer, train_metrics, eval_metrics, train_time, step, do_predict=False):
        if not do_predict:
            summary_writer.scalar("train_time", train_time, step)

            train_metrics = get_metrics(train_metrics)
            for key, vals in train_metrics.items():
                tag = f"train_{key}"
                for i, val in enumerate(vals):
                    summary_writer.scalar(tag, val, step - len(vals) + i + 1)
            summary_prefix = "eval"
        else:
            summary_prefix = "test"

        for metric_name, value in eval_metrics.items():
            summary_writer.scalar(f"{summary_prefix}_{metric_name}", value, step)

    # 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)

    data_collator = FlaxDataCollatorSpeechSeq2SeqWithPadding(
        processor=processor,
        decoder_start_token_id=model.config.decoder_start_token_id,
        input_padding="longest",
        target_padding="longest",
        max_target_length=max_label_length,
        pad_input_to_multiple_of=pad_input_to_multiple_of,
        pad_target_to_multiple_of=pad_target_to_multiple_of if pad_target_to_multiple_of else max_label_length,
    )

    # Enable tensorboard only on the master node
    has_tensorboard = is_tensorboard_available()
    if has_tensorboard and current_host_idx == 0:
        try:
            from flax.metrics.tensorboard import SummaryWriter

            summary_writer = SummaryWriter(
                log_dir=output_dir / "runs" / f"{datetime.now():%b%d_%H-%M-%S}_{socket.gethostname()}")
        except ImportError as ie:
            has_tensorboard = False
            logger.warning(
                f"Unable to display metrics through TensorBoard because some packages are not installed: {ie}"
            )
    else:
        if current_host_idx == 0:
            logger.warning(
                "Unable to display metrics through TensorBoard because the package is not installed: "
                "Please run pip install tensorboard to enable."
            )

    # Initialize our training
    rng = jax.random.PRNGKey(training_args.seed)
    rng, dropout_rng = jax.random.split(rng)

    # Store some constant
    train_batch_size = int(
        training_args.per_device_train_batch_size) * jax.device_count()
    eval_batch_size = int(
        training_args.per_device_eval_batch_size) * jax.device_count()

    # Create learning rate schedule
    lr_scheduler_types = {"linear", "constant", "constant_with_warmup"}
    if training_args.lr_scheduler_type not in lr_scheduler_types:
        raise ValueError(
            f"lr_scheduler_type of type {training_args.lr_scheduler_type} not supported, choose from {lr_scheduler_types}."
        )
    elif training_args.lr_scheduler_type == "constant":
        warmup_init_value = training_args.learning_rate
        decay_end_value = training_args.learning_rate
    elif training_args.lr_scheduler_type == "constant_with_warmup":
        warmup_init_value = 0.0
        decay_end_value = training_args.learning_rate
    else:
        warmup_init_value = 0.0
        decay_end_value = 0.0
        
    linear_decay_lr_schedule_fn = create_learning_rate_fn(
        data_args.num_train_steps,
        training_args.warmup_steps,
        training_args.learning_rate,
        start_step=training_state["step"],
        warmup_init_value=warmup_init_value,
        decay_end_value=decay_end_value
    )
    
    # We use Optax's "masking" functionality to not apply weight decay
    # to bias and LayerNorm scale parameters. decay_mask_fn returns a
    # mask boolean with the same structure as the parameters.
    # The mask is True for parameters that should be decayed.
    def decay_mask_fn(params):
        flat_params = traverse_util.flatten_dict(params)
        # find out all LayerNorm parameters
        layer_norm_candidates = ["layer_norm", "self_attn_layer_norm", "final_layer_norm", "encoder_attn_layer_norm"]
        layer_norm_named_params = set(
            [
                layer[-2:]
                for layer_norm_name in layer_norm_candidates
                for layer in flat_params.keys()
                if layer_norm_name in "".join(layer).lower()
            ]
        )
        flat_mask = {path: (path[-1] != "bias" and path[-2:] not in layer_norm_named_params) for path in flat_params}
        return traverse_util.unflatten_dict(flat_mask)
        
    # Create adam optimizer
    optimizer = optax.adamw(
        learning_rate=linear_decay_lr_schedule_fn,
        b1=training_args.adam_beta1,
        b2=training_args.adam_beta2,
        eps=training_args.adam_epsilon,
        weight_decay=training_args.weight_decay,
        mask=decay_mask_fn,
    )
    if training_args.gradient_accumulation_steps > 1:
        optimizer = optax.MultiSteps(
            optimizer, training_args.gradient_accumulation_steps
        )

    # Setup train state
    state = TrainState.create(
        apply_fn=model.__call__, params=model.params, tx=optimizer, dropout_rng=dropout_rng)

    # Label smoothed cross entropy
    def loss_fn(logits, labels, label_smoothing_factor=0.0):
        """
        The label smoothing implementation is adapted from Flax's official example:
        https://github.com/google/flax/blob/87a211135c6a377c8f29048a1cac3840e38b9da4/examples/wmt/train.py#L104
        """
        vocab_size = logits.shape[-1]
        confidence = 1.0 - label_smoothing_factor
        low_confidence = (1.0 - confidence) / (vocab_size - 1)
        normalizing_constant = -(
            confidence * jnp.log(confidence) + (vocab_size - 1) *
            low_confidence * jnp.log(low_confidence + 1e-20)
        )
        soft_labels = onehot(labels, vocab_size,
                             on_value=confidence, off_value=low_confidence)

        loss = optax.softmax_cross_entropy(logits, soft_labels)
        loss = loss - normalizing_constant

        # Ignore padded tokens from loss, i.e. where labels are not set to -100
        padding_mask = labels >= 0
        loss = loss * padding_mask
        loss = loss.sum()
        num_labels = padding_mask.sum()
        return loss, num_labels

    # Define gradient update step fn
    def train_step(state, batch, label_smoothing_factor=0.0):
        
        dropout_rng, new_dropout_rng = jax.random.split(state.dropout_rng)

        def compute_loss(params):
            labels = batch.pop("labels")
            logits = state.apply_fn(
                **batch, params=params, dropout_rng=dropout_rng, train=True)[0]
            loss, num_labels = loss_fn(logits, labels, label_smoothing_factor)
            return loss, num_labels

        grad_fn = jax.value_and_grad(compute_loss, has_aux=True)
        (loss, num_labels), grad = grad_fn(state.params)
        num_labels = jax.lax.psum(num_labels, "batch")

        # True loss = total loss / total samples
        loss = jax.lax.psum(loss, "batch")
        loss = jax.tree_util.tree_map(lambda x: x / num_labels, loss)

        # True grad = total grad / total samples
        grad = jax.lax.psum(grad, "batch")
        grad = jax.tree_util.tree_map(lambda x: x / num_labels, grad)
        new_state = state.apply_gradients(
            grads=grad, dropout_rng=new_dropout_rng)

        metrics = {"loss": loss,
                   "learning_rate": linear_decay_lr_schedule_fn(state.step)}

        return new_state, metrics

    # Define eval fn
    def eval_step(params, batch, label_smoothing_factor=0.0):
        labels = batch.pop("labels")
        logits = model(**batch, params=params, train=False)[0]

        loss, num_labels = loss_fn(logits, labels, label_smoothing_factor)
        num_labels = jax.lax.psum(num_labels, "batch")

        # True loss = total loss / total samples
        loss = jax.lax.psum(loss, "batch")
        loss = jax.tree_util.tree_map(lambda x: x / num_labels, loss)

        metrics = {"loss": loss}
        return metrics

    # Define generation function
    num_beams = model_args.num_beams if model_args.num_beams is not None else model.config.num_beams
    gen_kwargs = {"max_length": max_label_length, "num_beams": num_beams}

     
    def generate_step(params, batch):
        model.params = params
        attention_mask = batch.get("attention_mask")
        output_ids = model.generate(batch[model_input_name], attention_mask=attention_mask, **gen_kwargs)
        return output_ids.sequences

    # Create parallel version of the train and eval step
    p_train_step = jax.pmap(
        partial(train_step, label_smoothing_factor=training_args.label_smoothing_factor), "batch", donate_argnums=(0, )
    )
    p_eval_step = jax.pmap(partial(
        eval_step, label_smoothing_factor=training_args.label_smoothing_factor), "batch")
    p_generate_step = jax.pmap(generate_step, "batch")

    # Replicate the train state on each device
    state = state.replicate()
    
    # Logging
    logger.info("***** Running training *****")
    logger.info(
        f"  Original model = {model_args.model_name_or_path}")
    if training_args.push_to_hub:
        logger.info(
        f"  Hub model id = {training_args.hub_model_id}")
    logger.info(
        f"  Dataset name = {data_args.dataset_name}")
    logger.info(
        f"  Dataset config name = {data_args.dataset_config_name}")
    logger.info(
        f"  Learning rate = {training_args.learning_rate}")
    logger.info(
        f"  Scheduler = {training_args.lr_scheduler_type}")
    logger.info(
        f"  Num examples = {data_args.num_train_steps * train_batch_size:,}")
    if model_args.num_beams:
        logger.info(
        f"  Num beams evaluation = {model_args.num_beams}")
    logger.info(
        f"  Number of hosts = {num_of_hosts}")
    if num_of_hosts > 1:
        logger.info(
            f"  Current host idx = {current_host_idx}")
    logger.info(
        f"  Instantaneous batch size per device = {training_args.per_device_train_batch_size}")
    logger.info(
        f"  Total train batch size per node (w. parallel & distributed) = {train_batch_size // num_of_hosts:,}")
    logger.info(
        f"  Total train batch size (w. parallel & distributed) = {train_batch_size}")
    if training_args.gradient_accumulation_steps > 1:
        logger.info(
            f"  Gradient accumulation steps = {training_args.gradient_accumulation_steps}")
        logger.info(f"  ↪ Effective total batch size = {train_batch_size * training_args.gradient_accumulation_steps:,}")
    logger.info(f"  Total optimization steps = {data_args.num_train_steps - training_state['step']:,}")
    if training_state['step'] > 0:
        logger.info(f"  ↪ Starting at {training_state['step']:,} and finishing at {data_args.num_train_steps:,}")

    train_time = 0

    # Training summary
    language_code = None  # Maybe 'multilingual'?
    if data_args.language is not None:
        language = data_args.language.lower()
        if language in TO_LANGUAGE_CODE:
            language_code = TO_LANGUAGE_CODE[language]
        elif len(language) == 2:
            language_code = language
    training_summary = {
        "model_name": repo_name.split("/")[-1] if repo_name else model_name_or_path,
        "language": language_code,
        "tags": ["audio", "asr", "automatic-speech-recognition", "hf-asr-leaderboard"],
        "license": "apache-2.0",
        "finetuned_from": model_args.model_name_or_path,
        "tasks": ["asr"],
        "dataset": data_args.dataset_name,
        "dataset_args": {"name": data_args.dataset_config_name},
        "source": "flax",
        "eval_lines": [],
        "eval_results": None,
        "hyperparameters": {
            "learning_rate": training_args.learning_rate,
            "lr_scheduler_type": training_args.lr_scheduler_type,
            "per_device_train_batch_size": training_args.per_device_train_batch_size,
            "total_train_batch_size_per_node": train_batch_size // num_of_hosts,
            "total_train_batch_size": train_batch_size,
            "total_optimization_steps": f"{(data_args.num_train_steps - training_state['step']):,}",
            "starting_optimization_step": f"{training_state['step']:,}" if training_state['step'] > 0 else None,
            "finishing_optimization_step": f"{data_args.num_train_steps:,}",
            "num_train_dataset_workers": f"{num_workers}",
            "num_hosts": f"{num_of_hosts}",
            "total_num_training_examples": f"{data_args.num_train_steps * train_batch_size:,}",
            "steps_per_epoch": "_To be computed after first epoch_",
            "num_beams": model_args.num_beams,
        },
        # TODO: Adapt https://github.com/huggingface/transformers/blob/main/src/transformers/modelcard.py#L855
        # "hyperparameters": training_args.to_sanitized_dict()
    }   
        
    if training_args.gradient_accumulation_steps > 1:
        training_summary["hyperparameters"]["gradient_accumulation_steps"] = f"{training_args.gradient_accumulation_steps:,}"
        training_summary["hyperparameters"]["effective_total_train_batch_size"] = f"{train_batch_size * training_args.gradient_accumulation_steps:,}"
    
    if model_args.dropout or model_args.attention_dropout or model_args.activation_dropout or model_args.encoder_dropout or model_args.decoder_dropout:
        training_summary["hyperparameters"]["dropout"] = True
        if model_args.dropout:
            training_summary["hyperparameters"]["dropout_probability"] = model_args.dropout
        if model_args.attention_dropout:
            training_summary["hyperparameters"]["attention_dropout_probability"] = model_args.attention_dropout
        if model_args.activation_dropout:
            training_summary["hyperparameters"]["activation_dropout_probability"] = model_args.activation_dropout
        if model_args.encoder_dropout:
            training_summary["hyperparameters"]["encoder_dropout_probability"] = model_args.encoder_dropout
        if model_args.decoder_dropout:
            training_summary["hyperparameters"]["decoder_dropout_probability"] = model_args.decoder_dropout

    # Create README if it does not exist
    readme = output_dir / "README.md"
    if not readme.exists():
        readme.write_text(TrainingSummary(**training_summary).to_model_card())
    
    # ======================== Training ================================
    train_start = time.time()

    train_metrics = []
    epoch = 0
    if training_args.do_train:
        train_dataset = vectorized_datasets["train"].shuffle(seed=training_args.seed, buffer_size=data_args.shuffle_buffer_size)
        # Split by node
        train_dataset = split_dataset_by_node(train_dataset, rank=current_host_idx, world_size=num_of_hosts)   
    
        if train_dataset.n_shards < training_args.dataloader_num_workers:
            num_workers = train_dataset.n_shards

        logger.info(f"  Number of train dataset workers = {num_workers}")
        train_loader = data_loader(train_dataset, train_batch_size // num_of_hosts, num_workers=num_workers)

    if training_args.do_eval:
        eval_dataset = vectorized_datasets["eval"]
    
    if training_args.do_train and not training_args.ignore_data_skip and training_state["step"] > 0:
        logger.info(
            f"  Will skip the first {training_state['step']} steps. If this takes a lot of time,"
            " you can add the `--ignore_data_skip` flag to your launch command, but you will resume the"
            " training on data already seen by your model."
        )
        for step in tqdm(range(training_state["step"]), desc=f"Skipping data for {training_state['step']} steps...", position=1, leave=False):
            try:
                samples = next(train_loader)
            except StopIteration:
                epoch += 1
                train_dataset.set_epoch(epoch)
                train_loader = data_loader(train_dataset, train_batch_size // num_of_hosts, num_workers=num_workers)
                samples = next(train_loader)
            batch = data_collator(samples)
            # batch = shard(batch.data)
    
    for step in tqdm(range(data_args.num_train_steps), desc="Training...", position=1, leave=False):
        # Skip initial steps if these are specified. 
        if step < training_state["step"]:
            continue
        
        # =========================== Training ===========================
        if training_args.do_train:
            try:
                samples = next(train_loader)
            except StopIteration:
                epoch += 1
                train_dataset.set_epoch(epoch)
                train_loader = data_loader(train_dataset, train_batch_size // num_of_hosts, num_workers=num_workers)
                samples = next(train_loader)
                logger.info(
                    f"Completed epoch ({epoch} | Loss: {train_metric['loss']}, Learning Rate:"
                    f" {train_metric['learning_rate']})"
                )
                training_summary["hyperparameters"]["steps_per_epoch"] = step // epoch

            batch = data_collator(samples)
            batch = shard(batch.data)
                      
            state, train_metric = p_train_step(state, batch)
            train_metrics.append(train_metric)

            train_time += time.time() - train_start
            train_metric = unreplicate(train_metric)

        # ========================== Evaluating ==========================
        # Evaluate at each eval_steps, and at the end of training at num_train_steps
        if training_args.do_eval and (step % training_args.eval_steps == 0 or step == data_args.num_train_steps - 1):
            logger.info(
                f"Starting evaluation at step {step} of num_training_step {data_args.num_train_steps} steps. Planned evaluation every {training_args.eval_steps} steps." 
            )
            eval_metrics = []
            eval_preds = []
            eval_labels = []
            eval_loader = data_loader(eval_dataset, eval_batch_size, drop_last=False)
            if data_args.max_eval_samples:
                max_eval_steps_iter = range(1 + data_args.max_eval_samples // eval_batch_size)
            else:
                max_eval_steps_iter = itertools.repeat(None)
            for _ in tqdm(max_eval_steps_iter, desc="Evaluating...", position=2, leave=False):
                # Model forward
                try:
                    samples = next(eval_loader)
                except StopIteration:
                    break
                batch = data_collator(samples)
                
                labels = batch["labels"]

                metrics = pad_shard_unpad(p_eval_step, static_return=True)(
                    state.params, batch.data, min_device_batch=training_args.per_device_eval_batch_size
                )
                eval_metrics.append(metrics)

                # Generation
                if training_args.predict_with_generate:
                    generated_ids = pad_shard_unpad(
                        p_generate_step)(state.params, batch.data)
                    eval_preds.extend(jax.device_get(
                        generated_ids.reshape(-1, gen_kwargs["max_length"])))
                    eval_labels.extend(labels)

            # Normalize eval metrics
            eval_metrics = get_metrics(eval_metrics)
            eval_metrics = jax.tree_util.tree_map(jnp.mean, eval_metrics)

            # Compute metrics
            metric_desc = ""
            if training_args.predict_with_generate:
                metric_values = compute_metrics(
                    eval_preds, eval_labels, return_preds_labels=True
                )
                eval_metrics.update(metric_values)
                metric_desc = " | ".join(
                    [f"Eval {key}: {value}" for key, value in metric_values.items()])

            # Print metrics
            desc = f"Step: {step} | Epoch: {epoch} (Eval Loss: {eval_metrics['loss']} | {metric_desc})"
            logger.info(desc)

            # Update training state
            training_state = update_training_state(
                training_state,
                train_metrics,
                eval_metrics,
                step,
            )

            # Save metrics
            if has_tensorboard and current_host_idx == 0:
                write_metric(
                    summary_writer,
                    train_metrics,
                    eval_metrics,
                    train_time,
                    step,
                )

            # Save checkpoint at each eval_steps and push checkpoint to the hub
            if current_host_idx  == 0:
                params = jax.device_get(
                    jax.tree_util.tree_map(lambda x: x[0], state.params))
                model.save_pretrained(training_args.output_dir, params=params)
                tokenizer.save_pretrained(training_args.output_dir)
                # Report eval results if training is done
                if step == data_args.num_train_steps - 1:
                    training_summary["eval_results"] = training_state["eval_lines"][-1]
                else:
                    training_summary.update({"eval_lines": training_state["eval_lines"]})
                (output_dir / "training_state.bin").write_text(json.dumps(training_state))
                # Write model card
                readme.write_text(TrainingSummary(**training_summary).to_model_card())
                if training_args.push_to_hub:
                    repo.push_to_hub(
                        commit_message=f"Saving weights and logs of step {step} - epoch {epoch}", blocking=False)

    # ======================== Prediction loop ==============================
    if training_args.do_predict:
        logger.info("***** Runing prediction *****")
        predict_dataset = vectorized_datasets["test"]
        
        pred_metrics = []
        pred_preds = []
        pred_labels = []
        pred_loader = data_loader(predict_dataset, eval_batch_size, drop_last=False)
        if data_args.max_predict_samples:
            max_pred_steps_iter = range(1 + data_args.max_predict_samples // eval_batch_size)
        else:
            max_pred_steps_iter = itertools.repeat(None)
        for _ in tqdm(max_pred_steps_iter, desc="Predicting...", position=2, leave=False):
            # Model forward
            try:
                samples = next(pred_loader)
            except StopIteration:
                break
            batch = data_collator(samples)
            
            labels = batch["labels"]

            metrics = pad_shard_unpad(p_eval_step, static_return=True)(
                state.params, batch.data, min_device_batch=training_args.per_device_eval_batch_size
            )
            pred_metrics.append(metrics)

            # Generation
            if training_args.predict_with_generate:
                generated_ids = pad_shard_unpad(
                    p_generate_step)(state.params, batch.data)
                pred_preds.extend(jax.device_get(
                    generated_ids.reshape(-1, gen_kwargs["max_length"])))
                pred_labels.extend(labels)

        # Normalize eval metrics
        pred_metrics = get_metrics(pred_metrics)
        pred_metrics = jax.tree_util.tree_map(jnp.mean, pred_metrics)

        # Compute metrics
        metric_desc = ""
        if training_args.predict_with_generate:
            metric_values = compute_metrics(
                pred_preds, pred_labels, return_preds_labels=True
            )
            pred_metrics.update(metric_values)
            metric_desc = " | ".join(
                [f"Predict {key}: {value}" for key, value in metric_values.items()])

        # Print metrics
        desc = f"Predict Loss: {pred_metrics['loss']} | {metric_desc})"
        logger.info(desc)

        # Save metrics
        if has_tensorboard and current_host_idx == 0:
            write_metric(summary_writer, [], pred_metrics, 0, 0, do_predict=True)

        # Save final metrics in json
        if current_host_idx == 0:
            pred_metrics = {f"test_{metric_name}": value for metric_name, value in metric_values.items()}
            (output_dir / "test_results.json").write_text(
                json.dumps(pred_metrics, indent=4, sort_keys=True)
            )
            if training_args.push_to_hub:
                repo.push_to_hub(
                    commit_message=f"Saving test results", blocking=False)


if __name__ == "__main__":
    main()