save_tokenized_data.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 masked language modeling (BERT, ALBERT, RoBERTa...) with whole word masking on a
text file or a dataset.

Here is the full list of checkpoints on the hub that can be fine-tuned by this script:
https://huggingface.co/models?filter=masked-lm
"""
import logging
import os
import sys
import time
from dataclasses import dataclass, field

# You can also adapt this script on your own masked language modeling task. Pointers for this are left as comments.
from pathlib import Path
from typing import Dict, List, Optional, Tuple

import numpy as np
from datasets import load_dataset
from tqdm import tqdm

import flax
import jax
import jax.numpy as jnp
import optax
from flax import jax_utils, traverse_util
from flax.training import train_state
from flax.training.common_utils import get_metrics, onehot, shard
from transformers import (
    CONFIG_MAPPING,
    FLAX_MODEL_FOR_MASKED_LM_MAPPING,
    AutoConfig,
    AutoTokenizer,
    FlaxAutoModelForMaskedLM,
    HfArgumentParser,
    PreTrainedTokenizerBase,
    TensorType,
    TrainingArguments,
    is_tensorboard_available,
    set_seed,
)
import json
from flax.training import checkpoints
from flax.jax_utils import unreplicate
from flax.training.checkpoints import save_checkpoint, restore_checkpoint
from importlib.util import find_spec


MODEL_CONFIG_CLASSES = list(FLAX_MODEL_FOR_MASKED_LM_MAPPING.keys())
MODEL_TYPES = tuple(conf.model_type for conf in MODEL_CONFIG_CLASSES)


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

    model_name_or_path: Optional[str] = field(
        default=None,
        metadata={
            "help": "The model checkpoint for weights initialization."
            "Don't set if you want to train a model from scratch."
        },
    )
    model_type: Optional[str] = field(
        default=None,
        metadata={"help": "If training from scratch, pass a model type from the list: " + ", ".join(MODEL_TYPES)},
    )
    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 do you want to store the pretrained models downloaded from s3"}
    )
    use_fast_tokenizer: bool = field(
        default=True,
        metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
    )
    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]`."
        },
    )


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

    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)."}
    )
    train_ref_file: Optional[str] = field(
        default=None,
        metadata={"help": "An optional input train ref data file for whole word masking in Chinese."},
    )
    validation_ref_file: Optional[str] = field(
        default=None,
        metadata={"help": "An optional input validation ref data file for whole word masking in Chinese."},
    )
    overwrite_cache: bool = field(
        default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
    )



    validation_split_percentage: Optional[int] = field(
        default=5,
        metadata={
            "help": "The percentage of the train set used as validation set in case there's no validation split"
        },
    )
    max_seq_length: Optional[int] = field(
        default=None,
        metadata={
            "help": "The maximum total input sequence length after tokenization. Sequences longer "
            "than this will be truncated. Default to the max input length of the model."
        },
    )
    preprocessing_num_workers: Optional[int] = field(
        default=None,
        metadata={"help": "The number of processes to use for the preprocessing."},
    )
    mlm_probability: float = field(
        default=0.15, metadata={"help": "Ratio of tokens to mask for masked language modeling loss"}
    )
    pad_to_max_length: bool = field(
        default=False,
        metadata={
            "help": "Whether to pad all samples to `max_seq_length`. "
            "If False, will pad the samples dynamically when batching to the maximum length in the batch."
        },
    )
    line_by_line: bool = field(
        default=False,
        metadata={"help": "Whether distinct lines of text in the dataset are to be handled as distinct sequences."},
    )
    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."
        },
    )

    




@flax.struct.dataclass
class FlaxDataCollatorForLanguageModeling:
    """
    Data collator used for language modeling. Inputs are dynamically padded to the maximum length of a batch if they
    are not all of the same length.

    Args:
        tokenizer (:class:`~transformers.PreTrainedTokenizer` or :class:`~transformers.PreTrainedTokenizerFast`):
            The tokenizer used for encoding the data.
        mlm_probability (:obj:`float`, `optional`, defaults to 0.15):
            The probability with which to (randomly) mask tokens in the input.

    .. note::

        For best performance, this data collator should be used with a dataset having items that are dictionaries or
        BatchEncoding, with the :obj:`"special_tokens_mask"` key, as returned by a
        :class:`~transformers.PreTrainedTokenizer` or a :class:`~transformers.PreTrainedTokenizerFast` with the
        argument :obj:`return_special_tokens_mask=True`.
    """

    tokenizer: PreTrainedTokenizerBase
    mlm_probability: float = 0.15

    def __post_init__(self):
        if self.tokenizer.mask_token is None:
            raise ValueError(
                "This tokenizer does not have a mask token which is necessary for masked language modeling. "
                "You should pass `mlm=False` to train on causal language modeling instead."
            )

    def __call__(self, examples: List[Dict[str, np.ndarray]], pad_to_multiple_of: int) -> Dict[str, np.ndarray]:
        # Handle dict or lists with proper padding and conversion to tensor.
        batch = self.tokenizer.pad(examples, pad_to_multiple_of=pad_to_multiple_of, return_tensors=TensorType.NUMPY)

        # If special token mask has been preprocessed, pop it from the dict.
        special_tokens_mask = batch.pop("special_tokens_mask", None)

        batch["input_ids"], batch["labels"] = self.mask_tokens(
            batch["input_ids"], special_tokens_mask=special_tokens_mask
        )
        return batch

    def mask_tokens(
        self, inputs: np.ndarray, special_tokens_mask: Optional[np.ndarray]
    ) -> Tuple[jnp.ndarray, jnp.ndarray]:
        """
        Prepare masked tokens inputs/labels for masked language modeling: 80% MASK, 10% random, 10% original.
        """
        labels = inputs.copy()
        # We sample a few tokens in each sequence for MLM training (with probability `self.mlm_probability`)
        probability_matrix = np.full(labels.shape, self.mlm_probability)
        special_tokens_mask = special_tokens_mask.astype("bool")

        probability_matrix[special_tokens_mask] = 0.0
        masked_indices = np.random.binomial(1, probability_matrix).astype("bool")
        labels[~masked_indices] = -100  # We only compute loss on masked tokens

        # 80% of the time, we replace masked input tokens with tokenizer.mask_token ([MASK])
        indices_replaced = np.random.binomial(1, np.full(labels.shape, 0.8)).astype("bool") & masked_indices
        inputs[indices_replaced] = self.tokenizer.convert_tokens_to_ids(self.tokenizer.mask_token)

        # 10% of the time, we replace masked input tokens with random word
        indices_random = np.random.binomial(1, np.full(labels.shape, 0.5)).astype("bool")
        indices_random &= masked_indices & ~indices_replaced

        random_words = np.random.randint(self.tokenizer.vocab_size, size=labels.shape, dtype="i4")
        inputs[indices_random] = random_words[indices_random]

        # The rest of the time (10% of the time) we keep the masked input tokens unchanged
        return inputs, labels


def generate_batch_splits(samples_idx: jnp.ndarray, batch_size: int) -> jnp.ndarray:
    num_samples = len(samples_idx)
    samples_to_remove = num_samples % batch_size

    if samples_to_remove != 0:
        samples_idx = samples_idx[:-samples_to_remove]
    sections_split = num_samples // batch_size
    batch_idx = np.split(samples_idx, sections_split)
    return batch_idx


def write_train_metric(summary_writer, train_metrics, train_time, step):
    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)


def write_eval_metric(summary_writer, eval_metrics, step):
    for metric_name, value in eval_metrics.items():
        summary_writer.scalar(f"eval_{metric_name}", value, step)


if __name__ == "__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, TrainingArguments))
    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()

    if (
        os.path.exists(training_args.output_dir)
        and 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."
        )

    # Setup logging
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
        level="NOTSET",
        datefmt="[%X]",
    )

    # Log on each process the small summary:
    logger = logging.getLogger(__name__)

    # Set the verbosity to info of the Transformers logger (on main process only):
    logger.info(f"Training/evaluation parameters {training_args}")

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

    # Get the datasets: you can either provide your own CSV/JSON/TXT 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 column called 'text' or the first column if no column called
    # 'text' is found. You can easily tweak this behavior (see below).
    #
    # In distributed training, the load_dataset function guarantees that only one local process can concurrently
    # download the dataset.
    if data_args.dataset_name is not None:
        # Downloading and loading a dataset from the hub.
        datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name, cache_dir=model_args.cache_dir)

        if "validation" not in datasets.keys():
            datasets["validation"] = load_dataset(
                data_args.dataset_name,
                data_args.dataset_config_name,
                split=f"train[:{data_args.validation_split_percentage}%]",
                cache_dir=model_args.cache_dir,
            )
            datasets["train"] = load_dataset(
                data_args.dataset_name,
                data_args.dataset_config_name,
                split=f"train[{data_args.validation_split_percentage}%:]",
                cache_dir=model_args.cache_dir,
            )
    else:
        import glob
        import random
        data_files = []
        def add_jsonlines_dir(path, filespec):
            global data_files
            data_files += glob.glob(f"{path}/{filespec}")
            data_files = list(set(data_files))
            print(f"Number of files {len(data_files)} after adding {path} glob {filespec}")
        #add_jsonlines_dir(f"/data/c4_cleaned2", "*.gz")
        #add_jsonlines_dir(f"/data/nrc_uniq_cleaned_20210223", "*.gz")
        add_jsonlines_dir(f"/data/nu_uniq_cleaned_20210225", "*.gz")
        random.Random(42).shuffle(data_files)
        total = len(data_files)
        print(total)
        perc = 0.05
        val_size = int(perc * total)
        train_size = total - val_size
        train = data_files[5:8]
        val = data_files[1:3]
        print(f"Got {len(train)} training files and {perc * 100} % {len(val)} validation files")
        assert list(set(train) & set(val)) == [], "Train overlaps with test"
        datasets = load_dataset('json', data_files={'train': train, 'validation': val},cache_dir="/home/dat/.cache/huggingface/datasets/json/default-9add402b38836560/0.0.0/f92a4de297ac644ad9781979b79064b0e222b3af766f8ea3bee32390dca23723")

    #from datasets import Dataset

    #dataset = Dataset.from_file("/home/dat/.cache/huggingface/datasets/json/default-9add402b38836560/0.0.0/f92a4de297ac644ad9781979b79064b0e222b3af766f8ea3bee32390dca23723/json-train.arrow") 
    #dataset = Dataset.from_file("/home/dat/.cache/huggingface/datasets/json/default-9add402b38836560/0.0.0/f92a4de297ac644ad9781979b79064b0e222b3af766f8ea3bee32390dca23723/json-validation.arrow") 


    def mb_item(x):
        return x.item() if hasattr(x, "item") else x

    # 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.
    if model_args.config_name:
        config = AutoConfig.from_pretrained(model_args.config_name, cache_dir=model_args.cache_dir)
    elif model_args.model_name_or_path:
        config = AutoConfig.from_pretrained(model_args.model_name_or_path, cache_dir=model_args.cache_dir)
    else:
        config = CONFIG_MAPPING[model_args.model_type]()
        logger.warning("You are instantiating a new config instance from scratch.")

    if model_args.tokenizer_name:
        tokenizer = AutoTokenizer.from_pretrained(
            model_args.tokenizer_name, cache_dir=model_args.cache_dir, use_fast=model_args.use_fast_tokenizer
        )
    elif model_args.model_name_or_path:
        tokenizer = AutoTokenizer.from_pretrained(
            model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_fast=model_args.use_fast_tokenizer
        )
    else:
        raise ValueError(
            "You are instantiating a new tokenizer from scratch. This is not supported by this script."
            "You can do it from another script, save it, and load it from here, using --tokenizer_name."
        )

    # Preprocessing the datasets.
    # First we tokenize all the texts.
    if training_args.do_train:
        column_names = datasets["train"].column_names
    else:
        column_names = datasets["validation"].column_names
    text_column_name = "text" if "text" in column_names else column_names[0]

    max_seq_length = min(data_args.max_seq_length, tokenizer.model_max_length)

    if data_args.line_by_line:
        # When using line_by_line, we just tokenize each nonempty line.
        padding = "max_length" if data_args.pad_to_max_length else False

        def tokenize_function(examples):
            # Remove empty lines
            examples = [line for line in examples if len(line) > 0 and not line.isspace()]
            return tokenizer(
                examples,
                return_special_tokens_mask=True,
                padding=padding,
                truncation=True,
                max_length=max_seq_length,
            )

        tokenized_datasets = datasets.map(
            tokenize_function,
            input_columns=[text_column_name],
            batched=True,
            num_proc=data_args.preprocessing_num_workers,
            remove_columns=column_names,
            load_from_cache_file=not data_args.overwrite_cache,
        )
        tokenized_datasets.save_to_disk("/data/tokenized_data")
        print ("save data")
    else:
        # Otherwise, we tokenize every text, then concatenate them together before splitting them in smaller parts.
        # We use `return_special_tokens_mask=True` because DataCollatorForLanguageModeling (see below) is more
        # efficient when it receives the `special_tokens_mask`.
        def tokenize_function(examples):
            return tokenizer(examples[text_column_name], return_special_tokens_mask=True)

        tokenized_datasets = datasets.map(
            tokenize_function,
            batched=True,
            num_proc=data_args.preprocessing_num_workers,
            remove_columns=column_names,
            load_from_cache_file=not data_args.overwrite_cache,
        )

        # Main data processing function that will concatenate all texts from our dataset and generate chunks of
        # max_seq_length.
        def group_texts(examples):
            # Concatenate all texts.
            concatenated_examples = {k: sum(examples[k], []) for k in examples.keys()}
            total_length = len(concatenated_examples[list(examples.keys())[0]])
            # We drop the small remainder, we could add padding if the model supported it instead of this drop, you can
            # customize this part to your needs.
            if total_length >= max_seq_length:
                total_length = (total_length // max_seq_length) * max_seq_length
            # Split by chunks of max_len.
            result = {
                k: [t[i : i + max_seq_length] for i in range(0, total_length, max_seq_length)]
                for k, t in concatenated_examples.items()
            }
            return result

        # Note that with `batched=True`, this map processes 1,000 texts together, so group_texts throws away a
        # remainder for each of those groups of 1,000 texts. You can adjust that batch_size here but a higher value
        # might be slower to preprocess.
        #
        # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
        # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.map
        tokenized_datasets = tokenized_datasets.map(
            group_texts,
            batched=True,
            num_proc=data_args.preprocessing_num_workers,
            load_from_cache_file=not data_args.overwrite_cache,
        )

        tokenized_datasets.save_to_disk("/data/tokenized_data")
        print ("save data")