run_table2text.py

#!/usr/bin/env python
# coding=utf-8
# Copyright 2021 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.
"""
# You can also adapt this script on your own sequence to sequence task. Pointers for this are left as comments.

import logging
import os
import sys
sys.path.append('YOUR_PATH_TO_PROJECT')
from dataclasses import dataclass, field
from typing import Optional

import datasets
import nltk  # Here to have a nice missing dependency error message early on
import numpy as np
from pathlib import Path
import torch
from datasets import load_dataset, load_metric
from filelock import FileLock

import transformers
from transformers import (
    AutoConfig,
    AutoModelForSeq2SeqLM,
    AutoTokenizer,
    DataCollatorForSeq2Seq,
    HfArgumentParser,

    Seq2SeqTrainer,
    Seq2SeqTrainingArguments,
    set_seed,
)

from transformers.trainer_callback import EarlyStoppingCallback

from transformers.trainer_utils import get_last_checkpoint
from transformers.utils import check_min_version, is_offline_mode
import re

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


logger = logging.getLogger(__name__)

try:
    nltk.data.find("tokenizers/punkt")
except (LookupError, OSError):
    if is_offline_mode():
        raise LookupError(
            "Offline mode: run this script without TRANSFORMERS_OFFLINE first to download nltk data files"
        )
    with FileLock(".lock") as lock:
        nltk.download("punkt", quiet=True)

import unidecode

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

    model_name_or_path: str = field(
        metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
    )
    config_name: Optional[str] = field(
        default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
    )
    tokenizer_name: Optional[str] = field(
        default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
    )
    cache_dir: Optional[str] = field(
        default=None,
        metadata={"help": "Where to store the pretrained models downloaded from huggingface.co"},
    )
    use_fast_tokenizer: bool = field(
        default=True,
        metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
    )
    model_revision: str = field(
        default="main",
        metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."},
    )
    use_auth_token: bool = field(
        default=False,
        metadata={
            "help": (
                "Will use the token generated when running `huggingface-cli login` (necessary to use this script "
                "with private models)."
            )
        },
    )
    resize_position_embeddings: Optional[bool] = field(
        default=None,
        metadata={
            "help": (
                "Whether to automatically resize the position embeddings if `max_source_length` exceeds "
                "the model's position embeddings."
            )
        },
    )


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

    lang: Optional[str] = field(default=None, metadata={"help": "Language id for summarization."})

    dataset_name: Optional[str] = field(
        default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
    )
    dataset_config_name: Optional[str] = field(
        default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
    )
    dataset_cache_dir: Optional[str] = field(
        default=None, metadata={"help": "The path to the cached datasets."}
    )

    use_watermark: bool = field(
        default=False, metadata={"help": "Whether to use watermark."}
    )
    use_input_enhance: bool = field(
        default=False, metadata={"help": "Whether to use input enhance during watermark."}
    )
    use_cluster_enhance: bool = field(
        default=False, metadata={"help": "Whether to use input&cluster enhance during watermark."}
    )
    save_similar_matrix: bool = field(
        default=False, metadata={"help": "Whether to save similar matrix mode!"}
    )
    similar_matrix_path: str =field(
        default=None, metadata={"help": "path to save word similar matrix."}
    )
    similar_matrix_per_size: int = field(
        default=1000, metadata={"help": "preprocess similar number per time."}
    )

    watermark_gamma: float = field(
        default=0.25, metadata={"help": "green list size."}
    )
    watermark_delta: float = field(
        default=2, metadata={"help": "green list aug"}
    )
    watermark_cluster_k: int = field(
        default=100, metadata={"help": "cluster number."}
    )
    watermark_temperature: float = field(
        default=1.0, metadata={"help": "cluster softmax temperature."}
    )
    early_stopping_patience: int = field(
        default=10, metadata={"help": "early stop patience."}
    )

    text_column: Optional[str] = field(
        default=None,
        metadata={"help": "The name of the column in the datasets containing the full texts (for summarization)."},
    )
    summary_column: Optional[str] = field(
        default=None,
        metadata={"help": "The name of the column in the datasets containing the summaries (for summarization)."},
    )

    train_file: Optional[str] = field(
        default=None, metadata={"help": "The input training data file (a jsonlines or csv file)."}
    )
    validation_file: Optional[str] = field(
        default=None,
        metadata={
            "help": (
                "An optional input evaluation data file to evaluate the metrics (rouge) on (a jsonlines or csv file)."
            )
        },
    )
    test_file: Optional[str] = field(
        default=None,
        metadata={
            "help": "An optional input test data file to evaluate the metrics (rouge) on (a jsonlines or csv file)."
        },
    )
    overwrite_cache: bool = field(
        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
    )
    preprocessing_num_workers: Optional[int] = field(
        default=None,
        metadata={"help": "The number of processes to use for the preprocessing."},
    )
    max_source_length: Optional[int] = field(
        default=1024,
        metadata={
            "help": (
                "The maximum total input sequence length after tokenization. Sequences longer "
                "than this will be truncated, sequences shorter will be padded."
            )
        },
    )
    max_target_length: Optional[int] = field(
        default=128,
        metadata={
            "help": (
                "The maximum total sequence length for target text after tokenization. Sequences longer "
                "than this will be truncated, sequences shorter will be padded."
            )
        },
    )
    val_max_target_length: Optional[int] = field(
        default=None,
        metadata={
            "help": (
                "The maximum total sequence length for validation target text after tokenization. Sequences longer "
                "than this will be truncated, sequences shorter will be padded. Will default to `max_target_length`."
                "This argument is also used to override the ``max_length`` param of ``model.generate``, which is used "
                "during ``evaluate`` and ``predict``."
            )
        },
    )
    pad_to_max_length: bool = field(
        default=False,
        metadata={
            "help": (
                "Whether to pad all samples to model maximum sentence length. "
                "If False, will pad the samples dynamically when batching to the maximum length in the batch. More "
                "efficient on GPU but very bad for TPU."
            )
        },
    )
    max_train_samples: Optional[int] = field(
        default=None,
        metadata={
            "help": (
                "For debugging purposes or quicker training, truncate the number of training examples to this "
                "value if set."
            )
        },
    )
    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": (
                "For debugging purposes or quicker training, truncate the number of prediction examples to this "
                "value if set."
            )
        },
    )
    num_beams: Optional[int] = field(
        default=None,
        metadata={
            "help": (
                "Number of beams to use for evaluation. This argument will be passed to ``model.generate``, "
                "which is used during ``evaluate`` and ``predict``."
            )
        },
    )
    ignore_pad_token_for_loss: bool = field(
        default=True,
        metadata={
            "help": "Whether to ignore the tokens corresponding to padded labels in the loss computation or not."
        },
    )
    source_prefix: Optional[str] = field(
        default="", metadata={"help": "A prefix to add before every source text (useful for T5 models)."}
    )

    forced_bos_token: Optional[str] = field(
        default=None,
        metadata={
            "help": (
                "The token to force as the first generated token after the decoder_start_token_id."
                "Useful for multilingual models like mBART where the first generated token"
                "needs to be the target language token (Usually it is the target language token)"
            )
        },
    )

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




table2text_name_mapping = {
    "dart": ("tripleset", "annotations"),
    'web_nlg': ("modified_triple_sets", "lex"),
}

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

    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, Seq2SeqTrainingArguments))
    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
        # If we pass only one argument to the script and it's the path to a json file,
        # let's parse it to get our arguments.
        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
    else:
        model_args, data_args, training_args = parser.parse_args_into_dataclasses()

    # 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_summarization", model_args, data_args)

    # 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)],
    )

    if training_args.should_log:
        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
        transformers.utils.logging.set_verbosity_info()

    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()

    # Log on each process the small summary:
    # print(f'local_rank:{training_args.local_rank}', flush=True)
    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}")

    if data_args.source_prefix is None and model_args.model_name_or_path in [
        "t5-small",
        "t5-base",
        "t5-large",
        "t5-3b",
        "t5-11b",
    ]:
        logger.warning(
            "You're running a t5 model but didn't provide a source prefix, which is the expected, e.g. with "
            "`--source_prefix 'translate Graph to English: ' `"
        )

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

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

    # Get the datasets: you can either provide your own CSV/JSON training and evaluation files (see below)
    # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
    # (the dataset will be downloaded automatically from the datasets Hub).
    #
    # For CSV/JSON files this script will use the first column for the full texts and the second column for the
    # summaries (unless you specify column names for this with the `text_column` and `summary_column` arguments).
    #
    # In distributed training, the load_dataset function guarantee that only one local process can concurrently
    # download the dataset.
    
    train_split = "train"
    validation_split = "validation"
    test_split = "test"
    if data_args.dataset_name == 'web_nlg':
        validation_split = "dev"

    if data_args.dataset_name is not None:
        if data_args.dataset_cache_dir is not None:
            if data_args.dataset_name == 'dart':
                raw_datasets = load_dataset(
                    'dart', 
                    cache_dir=data_args.dataset_cache_dir
                    )
            elif data_args.dataset_name == 'web_nlg':
                raw_datasets = load_dataset(
                    'web_nlg',
                    name='webnlg_challenge_2017',
                    cache_dir=data_args.dataset_cache_dir,
                )
                logger.info('dataset web_nlg: filter data that do not have reference.')
                raw_datasets = raw_datasets.filter(lambda x: x['test_category'] == 'testdata_with_lex' or x['test_category'] == '')
        
            else:
                raise NotImplementedError
    else:
        data_files = {}
        if data_args.train_file is not None:
            data_files["train"] = data_args.train_file
            extension = data_args.train_file.split(".")[-1]
        if data_args.validation_file is not None:
            data_files["validation"] = data_args.validation_file
            extension = data_args.validation_file.split(".")[-1]
        if data_args.test_file is not None:
            data_files["test"] = data_args.test_file
            extension = data_args.test_file.split(".")[-1]
        raw_datasets = load_dataset(
            extension,
            data_files=data_files,
            cache_dir=model_args.cache_dir,
            use_auth_token=True if model_args.use_auth_token else None,
        )
    # See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at
    # https://huggingface.co/docs/datasets/loading_datasets.html.

    # Load pretrained model and tokenizer
    #
    # Distributed training:
    # The .from_pretrained methods guarantee that only one local process can concurrently
    # download model & vocab.
    config = AutoConfig.from_pretrained(
        model_args.config_name if model_args.config_name else model_args.model_name_or_path,
        cache_dir=model_args.cache_dir,
        revision=model_args.model_revision,
        use_auth_token=True if model_args.use_auth_token else None,
    )
    tokenizer = AutoTokenizer.from_pretrained(
        model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path,
        cache_dir=model_args.cache_dir,
        use_fast=model_args.use_fast_tokenizer,
        revision=model_args.model_revision,
        use_auth_token=True if model_args.use_auth_token else None,
    )
    model = AutoModelForSeq2SeqLM.from_pretrained(
        model_args.model_name_or_path,
        from_tf=bool(".ckpt" in model_args.model_name_or_path),
        config=config,
        cache_dir=model_args.cache_dir,
        revision=model_args.model_revision,
        use_auth_token=True if model_args.use_auth_token else None,
    )

    # We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
    # on a small vocab and want a smaller embedding size, remove this test.
    embedding_size = model.get_input_embeddings().weight.shape[0]
    if len(tokenizer) > embedding_size:
        model.resize_token_embeddings(len(tokenizer))

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

    if (
        hasattr(model.config, "max_position_embeddings")
        and model.config.max_position_embeddings < data_args.max_source_length
    ):
        if model_args.resize_position_embeddings is None:
            logger.warning(
                "Increasing the model's number of position embedding vectors from"
                f" {model.config.max_position_embeddings} to {data_args.max_source_length}."
            )
            model.resize_position_embeddings(data_args.max_source_length)
        elif model_args.resize_position_embeddings:
            model.resize_position_embeddings(data_args.max_source_length)
        else:
            raise ValueError(
                f"`--max_source_length` is set to {data_args.max_source_length}, but the model only has"
                f" {model.config.max_position_embeddings} position encodings. Consider either reducing"
                f" `--max_source_length` to {model.config.max_position_embeddings} or to automatically resize the"
                " model's position encodings by passing `--resize_position_embeddings`."
            )

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

    # Preprocessing the datasets.
    # We need to tokenize inputs and targets.

    if training_args.do_train:
        column_names = raw_datasets[train_split].column_names
    elif training_args.do_eval:
        column_names = raw_datasets[validation_split].column_names
    elif training_args.do_predict:
        column_names = raw_datasets[test_split].column_names
    else:
        logger.info("There is nothing to do. Please pass `do_train`, `do_eval` and/or `do_predict`.")
        return

    # Get the column names for input/target.
    dataset_columns = table2text_name_mapping.get(data_args.dataset_name, None)
    if data_args.text_column is None:
        text_column = dataset_columns[0] if dataset_columns is not None else column_names[0]
    else:
        text_column = data_args.text_column
        if text_column not in column_names:
            raise ValueError(
                f"--text_column' value '{data_args.text_column}' needs to be one of: {', '.join(column_names)}"
            )
    
    if data_args.summary_column is None:
        summary_column = dataset_columns[1] if dataset_columns is not None else column_names[2]
    else:
        summary_column = data_args.summary_column
        if summary_column not in column_names:
            raise ValueError(
                f"--summary_column' value '{data_args.summary_column}' needs to be one of: {', '.join(column_names)}"
            )

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

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


    def preprocess_function_valid(examples):
        # remove pairs where at least one record is None
        def get_nodes(n):
            n = n.strip()
            n = n.replace('(', '')
            n = n.replace('\"', '')
            n = n.replace(')', '')
            n = n.replace(',', ' ')
            n = n.replace('_', ' ')

            n = unidecode.unidecode(n)

            return n
        def get_relation(n, lower=True):
            n = n.replace('(', '')
            n = n.replace(')', '')
            n = n.strip()
            n = n.split()
            n = "_".join(n)
            if lower:
                n = n.lower()
            return n
        
        def camel_case_split(identifier):
            matches = re.finditer('.+?(?:(?<=[a-z])(?=[A-Z])|(?<=[A-Z])(?=[A-Z][a-z])|$)', identifier)
            d = [m.group(0) for m in matches]
            new_d = []
            for token in d:
                token = token.replace('(', '')
                token_split = token.split('_')
                for t in token_split:
                    #new_d.append(t.lower())
                    new_d.append(t)
            return new_d
        
        inputs, targets = [], []
        for i in range(len(examples[text_column])):
            if examples[text_column][i] and examples[summary_column][i]:
                
                input = []
                if data_args.dataset_name == 'dart':
                    tripleset = examples[text_column][i]
                    for triplet in tripleset:
                        cur_input = ' (H) ' + get_nodes(triplet[0])  + ' (R) ' + ' '.join(camel_case_split(get_relation(triplet[1]))) + ' (T) ' + get_nodes(triplet[2])
                        input.append(cur_input)
                elif data_args.dataset_name == 'web_nlg':
                    tripleset = examples[text_column][i]['mtriple_set'][0]
                    for triplet in tripleset:
                        node1, node2, node3 = triplet.strip().split(' | ')
                        cur_input = ' (H) ' + get_nodes(node1)  + ' (R) ' + ' '.join(camel_case_split(get_relation(node2))) + ' (T) ' + get_nodes(node3)
                        input.append(cur_input)
                else:
                    raise NotImplementedError
                
                inputs.append(' '.join(input).strip())
                targets.append(examples[summary_column][i]['text'][0].strip().replace('\n', ''))

            else:
                continue
    
        inputs = [prefix + inp for inp in inputs]
        model_inputs = tokenizer(inputs, max_length=data_args.max_source_length, padding=padding, truncation=True)

        # Tokenize targets with the `text_target` keyword argument
        labels = tokenizer(text_target=targets, max_length=max_target_length, padding=padding, truncation=True)

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

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

    def preprocess_function(examples):
        # remove pairs where at least one record is None

        def get_nodes(n):
            n = n.strip()
            n = n.replace('(', '')
            n = n.replace('\"', '')
            n = n.replace(')', '')
            n = n.replace(',', ' ')
            n = n.replace('_', ' ')

            n = unidecode.unidecode(n)

            return n
        def get_relation(n, lower=True):
            n = n.replace('(', '')
            n = n.replace(')', '')
            n = n.strip()
            n = n.split()
            n = "_".join(n)
            if lower:
                n = n.lower()
            return n

        def camel_case_split(identifier):
            matches = re.finditer('.+?(?:(?<=[a-z])(?=[A-Z])|(?<=[A-Z])(?=[A-Z][a-z])|$)', identifier)
            d = [m.group(0) for m in matches]
            new_d = []
            for token in d:
                token = token.replace('(', '')
                token_split = token.split('_')
                for t in token_split:
                    #new_d.append(t.lower())
                    new_d.append(t)
            return new_d
        
        inputs, targets = [], []
        for i in range(len(examples[text_column])):
            if examples[text_column][i] and examples[summary_column][i]:
                
                input = []
                if data_args.dataset_name == 'dart':
                    tripleset = examples[text_column][i]
                    for triplet in tripleset:
                        cur_input = ' (H) ' + get_nodes(triplet[0])  + ' (R) ' + ' '.join(camel_case_split(get_relation(triplet[1]))) + ' (T) ' + get_nodes(triplet[2])
                        input.append(cur_input)
                elif data_args.dataset_name == 'web_nlg':
                    tripleset = examples[text_column][i]['mtriple_set'][0]
                    for triplet in tripleset:
                        node1, node2, node3 = triplet.strip().split(' | ')
                        cur_input = ' (H) ' + get_nodes(node1)  + ' (R) ' + ' '.join(camel_case_split(get_relation(node2))) + ' (T) ' + get_nodes(node3)
                        input.append(cur_input)
                else:
                    raise NotImplementedError

                for target in examples[summary_column][i]['text']:
                    inputs.append(' '.join(input).strip())
                    targets.append(target.strip().replace('\n', ''))

            else:
                continue

        inputs = [prefix + inp for inp in inputs]
        model_inputs = tokenizer(inputs, max_length=data_args.max_source_length, padding=padding, truncation=True)

        # Tokenize targets with the `text_target` keyword argument
        labels = tokenizer(text_target=targets, max_length=max_target_length, padding=padding, truncation=True)

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

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

    if training_args.do_train:
        if train_split not in raw_datasets:
            raise ValueError("--do_train requires a train dataset")
        train_dataset = raw_datasets[train_split]
        if data_args.max_train_samples is not None:
            max_train_samples = min(len(train_dataset), data_args.max_train_samples)
            train_dataset = train_dataset.select(range(max_train_samples))
        with training_args.main_process_first(desc="train dataset map pre-processing"):
            train_dataset = train_dataset.map(
                preprocess_function,
                batched=True,
                num_proc=data_args.preprocessing_num_workers,
                remove_columns=column_names,
                load_from_cache_file=not data_args.overwrite_cache,
                desc="Running tokenizer on train dataset",
            )

    if training_args.do_eval:
        max_target_length = data_args.val_max_target_length
        if validation_split not in raw_datasets:
            raise ValueError("--do_eval requires a validation dataset")
        eval_dataset = raw_datasets[validation_split]
        if data_args.max_eval_samples is not None:
            max_eval_samples = min(len(eval_dataset), data_args.max_eval_samples)
            eval_dataset = eval_dataset.select(range(max_eval_samples))
        with training_args.main_process_first(desc="validation dataset map pre-processing"):
            eval_dataset = eval_dataset.map(
                preprocess_function_valid,
                batched=True,
                num_proc=data_args.preprocessing_num_workers,
                remove_columns=column_names,
                load_from_cache_file=not data_args.overwrite_cache,
                desc="Running tokenizer on validation dataset",
            )

    if training_args.do_predict:
        max_target_length = data_args.val_max_target_length
        if test_split not in raw_datasets:
            raise ValueError("--do_predict requires a test dataset")
        

        predict_dataset = raw_datasets[test_split]
        if data_args.max_predict_samples is not None:
            max_predict_samples = min(len(predict_dataset), data_args.max_predict_samples)
            predict_dataset = predict_dataset.select(range(max_predict_samples))
        

        with training_args.main_process_first(desc="prediction dataset map pre-processing"):
            predict_dataset = predict_dataset.map(
                preprocess_function_valid,
                batched=True,
                num_proc=data_args.preprocessing_num_workers,
                remove_columns=column_names,
                load_from_cache_file=not data_args.overwrite_cache,
                desc="Running tokenizer on prediction dataset",
            )

    # Data collator
    label_pad_token_id = -100 if data_args.ignore_pad_token_for_loss else tokenizer.pad_token_id
    data_collator = DataCollatorForSeq2Seq(
        tokenizer,
        model=model,
        label_pad_token_id=label_pad_token_id,
        pad_to_multiple_of=8 if training_args.fp16 else None,
    )

    # Metric
    metric = load_metric("sacrebleu")
    
    def postprocess_text(preds, labels):
        preds = [pred.strip() for pred in preds]
        labels = [[label.strip()] for label in labels]

        return preds, labels

    def compute_metrics(eval_preds):
        preds, labels = eval_preds
        if isinstance(preds, tuple):
            preds = preds[0]
        decoded_preds = tokenizer.batch_decode(preds, skip_special_tokens=True)
        if data_args.ignore_pad_token_for_loss:
            # Replace -100 in the labels as we can't decode them.
            labels = np.where(labels != -100, labels, tokenizer.pad_token_id)
        decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)

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

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

        prediction_lens = [np.count_nonzero(pred != tokenizer.pad_token_id) for pred in preds]
        result["gen_len"] = np.mean(prediction_lens)
        result = {k: round(v, 4) for k, v in result.items()}
        return result

    # Override the decoding parameters of Seq2SeqTrainer
    training_args.generation_max_length = (
        training_args.generation_max_length
        if training_args.generation_max_length is not None
        else data_args.val_max_target_length
    )
    training_args.generation_num_beams = (
        data_args.num_beams if data_args.num_beams is not None else training_args.generation_num_beams
    )

    callbacks = []
    if 't5' in model_args.model_name_or_path:
        callbacks = [EarlyStoppingCallback(early_stopping_patience=data_args.early_stopping_patience)]
    # Initialize our Trainer
    trainer = Seq2SeqTrainer(
        model=model,
        args=training_args,
        train_dataset=train_dataset if training_args.do_train else None,
        eval_dataset=eval_dataset if training_args.do_eval else None,
        tokenizer=tokenizer,
        data_collator=data_collator,
        compute_metrics=compute_metrics if training_args.predict_with_generate else None,
        callbacks=callbacks
    )

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

        metrics = train_result.metrics
        max_train_samples = (
            data_args.max_train_samples if data_args.max_train_samples is not None else len(train_dataset)
        )
        metrics["train_samples"] = min(max_train_samples, len(train_dataset))

        trainer.log_metrics("train", metrics)
        trainer.save_metrics("train", metrics)
        trainer.save_state()

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

        metrics = trainer.evaluate(metric_key_prefix="eval")
        max_eval_samples = data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)
        metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset))

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

    if training_args.do_predict:
        example = raw_datasets[test_split][0]
        logger.info(f'first example: {example}')
        text = tokenizer.decode(predict_dataset[0]['input_ids'])
        logger.info(f'first example: {text}')
        if data_args.save_similar_matrix:
            
            if os.path.exists(data_args.similar_matrix_path):
                mode = 'r'
            else:
                mode = 'w+'
                Path(data_args.similar_matrix_path).parent.mkdir(parents=True, exist_ok=True)

            word_vector = model.get_input_embeddings().weight.clone().detach()
            norm_vector = (word_vector ** 2).sum(axis=1)

            similar_matrix = np.memmap(data_args.similar_matrix_path + '_index.npy', 
                            dtype=np.int32, mode=mode, shape=(embedding_size, embedding_size))

            num_batches = embedding_size // data_args.similar_matrix_per_size + int(embedding_size % data_args.similar_matrix_per_size != 0)

            matrix_pre_index = 0
            from tqdm import tqdm
            for i in tqdm(range(num_batches)):
                matrix_cur_index = min(similar_matrix.shape[1], matrix_pre_index + data_args.similar_matrix_per_size)

                query = word_vector[matrix_pre_index:matrix_cur_index,:]
                norm_query = norm_vector[matrix_pre_index:matrix_cur_index]
                
                distances, indices = torch.sort((norm_query.reshape(-1, 1) + norm_vector - 2 * query @ word_vector.T), dim=1, descending=False)
            
                similar_matrix[matrix_pre_index:matrix_cur_index, :] = indices.cpu().numpy()
                matrix_pre_index = matrix_cur_index
            
            logger.info(f'successful save word similar matrix(L2 distance) according to the word embedding.')
            logger.info(f'similar matrix path:{data_args.similar_matrix_path}')
            logger.info(f'similar matrix size:{embedding_size} * {embedding_size}')

            return None
            
        logger.info("*** Predict ***")
        #! add_watermark
        if data_args.use_watermark:
            from watermark_processor import WatermarkLogitsProcessor, WatermarkDetector
            from transformers import LogitsProcessorList
            watermark_processor = WatermarkLogitsProcessor(
                                                vocab=list(tokenizer.get_vocab().values()),
                                                gamma=data_args.watermark_gamma,
                                                delta=data_args.watermark_delta,
                                                seeding_scheme="simple_1")
            setattr(watermark_processor, 'use_input_enhance', data_args.use_input_enhance)
            setattr(watermark_processor, 'use_cluster_enhance', data_args.use_cluster_enhance)
            if data_args.use_cluster_enhance:
                if not os.path.exists('/'.join(data_args.similar_matrix_path.split('/')[:-1])):
                    logger.info(f'use word cluster to enhance, Please provide the similar-matrix path!')
                    raise ValueError
                else:
                    similar_matrix = np.memmap(data_args.similar_matrix_path + "_index.npy", dtype=np.int32, shape=(embedding_size, embedding_size))
                    
                    setattr(watermark_processor, 'similar_matrix', similar_matrix)
                    setattr(watermark_processor, 'tokenizer', tokenizer)
                    setattr(watermark_processor, 'cluster_k', data_args.watermark_cluster_k)

            logger.info(f'watermark use input enhance: {data_args.use_input_enhance}')
            logger.info(f'watermark use cluster enhance: {data_args.use_cluster_enhance}')
            logger.info(f'watermark cluster size: {data_args.watermark_cluster_k} default:100')
            gen_kwargs = {'logits_processor': LogitsProcessorList([watermark_processor])}
            predict_results = trainer.predict(predict_dataset, metric_key_prefix="predict", **gen_kwargs)
        else:
            gen_kwargs = {}
            predict_results = trainer.predict(predict_dataset, metric_key_prefix="predict")
        print(gen_kwargs)
        #! add_watermark

        metrics = predict_results.metrics
        max_predict_samples = (
            data_args.max_predict_samples if data_args.max_predict_samples is not None else len(predict_dataset)
        )
        metrics["predict_samples"] = min(max_predict_samples, len(predict_dataset))

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

        if trainer.is_world_process_zero():
            if training_args.predict_with_generate:
                predictions = tokenizer.batch_decode(
                    predict_results.predictions, skip_special_tokens=True, clean_up_tokenization_spaces=True
                )
                predictions = [pred.strip() for pred in predictions]
                output_prediction_file = os.path.join(training_args.output_dir, "generated_predictions.txt")
                with open(output_prediction_file, "w") as writer:
                    writer.write("\n".join(predictions))
                
                #! save generated results
                output_single_file = os.path.join(training_args.output_dir, "generate_predictions_start_single.txt")
                with open(output_single_file, 'w') as writer:
                    for prediction in predictions:
                        writer.write(prediction.replace('\n', '/n') + '\n')

    kwargs = {"finetuned_from": model_args.model_name_or_path, "tasks": "table2text"}
    if data_args.dataset_name is not None:
        kwargs["dataset_tags"] = data_args.dataset_name
        if data_args.dataset_config_name is not None:
            kwargs["dataset_args"] = data_args.dataset_config_name
            kwargs["dataset"] = f"{data_args.dataset_name} {data_args.dataset_config_name}"
        else:
            kwargs["dataset"] = data_args.dataset_name

    if data_args.lang is not None:
        kwargs["language"] = data_args.lang

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

    return results


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


if __name__ == "__main__":
    main()