run_xnli.py

#!/usr/bin/env python
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
# Copyright (c) 2018, NVIDIA CORPORATION.  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.
""" Finetuning multi-lingual models on XNLI (e.g. Bert, DistilBERT, XLM).
    Adapted from `examples/text-classification/run_glue.py`"""

import logging
import os, pdb
import random
import sys
import warnings
from dataclasses import dataclass, field
from typing import Optional

import datasets
import evaluate
import numpy as np
from datasets import load_dataset
from tqdm import tqdm
import pandas as pd
import transformers
import json
from transformers import (
    AutoConfig,
    DebertaConfig,
    AutoModelForSequenceClassification,
    AutoTokenizer,
    DebertaTokenizer, 
    DebertaForSequenceClassification,
    DataCollatorWithPadding,
    EvalPrediction,
    HfArgumentParser,
    Trainer,
    TrainingArguments,
    default_data_collator,
    set_seed,
)
from transformers.trainer_utils import get_last_checkpoint
from transformers.utils import check_min_version, send_example_telemetry
from transformers.utils.versions import require_version
import torch
from torch import nn
import torch.nn.functional as F
import torchmetrics as tm
from torchmetrics.classification import MulticlassCalibrationError

# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
check_min_version("4.37.0.dev0")

require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/text-classification/requirements.txt")

logger = logging.getLogger(__name__)


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

    Using `HfArgumentParser` we can turn this class
    into argparse arguments to be able to specify them on
    the command line.
    """

    max_seq_length: Optional[int] = field(
        default=128,
        metadata={
            "help": (
                "The maximum total input sequence length after tokenization. Sequences longer "
                "than this will be truncated, sequences shorter will be padded."
            )
        },
    )
    overwrite_cache: bool = field(
        default=False, metadata={"help": "Overwrite the cached preprocessed datasets or not."}
    )
    pad_to_max_length: bool = field(
        default=True,
        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."
            )
        },
    )
    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."
            )
        },
    )


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

    model_name_or_path: str = field(
        default=None, metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
    )
    language: str = field(
        default=None, metadata={"help": "Evaluation language. Also train language if `train_language` is set to None."}
    )
    tau: float = field(
        default = 0.0
        )

    train_language: Optional[str] = field(
        default=None, metadata={"help": "Train language if it is different from the evaluation language."}
    )
    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 huggingface.co"},
    )
    do_lower_case: Optional[bool] = field(
        default=False,
        metadata={"help": "arg to indicate if tokenizer should do lower case in AutoTokenizer.from_pretrained()"},
    )
    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)."},
    )
    token: str = field(
        default=None,
        metadata={
            "help": (
                "The token to use as HTTP bearer authorization for remote files. If not specified, will use the token "
                "generated when running `huggingface-cli login` (stored in `~/.huggingface`)."
            )
        },
    )
    use_auth_token: bool = field(
        default=None,
        metadata={
            "help": "The `use_auth_token` argument is deprecated and will be removed in v4.34. Please use `token` instead."
        },
    )
    trust_remote_code: bool = field(
        default=False,
        metadata={
            "help": (
                "Whether or not to allow for custom models defined on the Hub in their own modeling files. This option"
                "should only be set to `True` for repositories you trust and in which you have read the code, as it will "
                "execute code present on the Hub on your local machine."
            )
        },
    )
    ignore_mismatched_sizes: bool = field(
        default=False,
        metadata={"help": "Will enable to load a pretrained model whose head dimensions are different."},
    )


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, TrainingArguments))
    model_args, data_args, training_args = parser.parse_args_into_dataclasses()

    if model_args.use_auth_token is not None:
        warnings.warn(
            "The `use_auth_token` argument is deprecated and will be removed in v4.34. Please use `token` instead.",
            FutureWarning,
        )
        if model_args.token is not None:
            raise ValueError("`token` and `use_auth_token` are both specified. Please set only the argument `token`.")
        model_args.token = model_args.use_auth_token

    # 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_xnli", model_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:
    logger.warning(
        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}, "
        + f"distributed training: {training_args.parallel_mode.value == 'distributed'}, 16-bits training: {training_args.fp16}"
    )
    logger.info(f"Training/evaluation parameters {training_args}")

    # 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:
            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)

    # In distributed training, the load_dataset function guarantees that only one local process can concurrently
    # download the dataset.
    # Downloading and loading xnli dataset from the hub.
    # if training_args.do_train:
    #     if model_args.train_language is None:
    #         train_dataset = load_dataset(
    #             "xnli",
    #             model_args.language,
    #             split="train",
    #             cache_dir=model_args.cache_dir,
    #             token=model_args.token,
    #         )
    #     else:
    #         train_dataset = load_dataset(
    #             "xnli",
    #             model_args.train_language,
    #             split="train",
    #             cache_dir=model_args.cache_dir,
    #             token=model_args.token,
    #         )
    #     label_list = train_dataset.features["label"].names
    # pdb.set_trace()

    # if training_args.do_eval:
    #     eval_dataset = load_dataset(
    #         "xnli",
    #         model_args.language,
    #         split="validation",
    #         cache_dir=model_args.cache_dir,
    #         token=model_args.token,
    #     )
    #     label_list = eval_dataset.features["label"].names

    # if training_args.do_predict:
    #     predict_dataset = load_dataset(
    #         "xnli",
    #         model_args.language,
    #         split="test",
    #         cache_dir=model_args.cache_dir,
    #         token=model_args.token,
    #     )
    #     label_list = predict_dataset.features["label"].names


    label2int = {"contradiction": 0, "entailment": 1, "neutral": 2}
    label_list = ["contradiction", "entailment", "neutral"]
    id2label={i: label for i, label in enumerate(label_list)}
    nli_dataset_path = "tmp_data/train_en_es.tsv"
    sts_dataset_path = "tmp_data/dev_en_es.tsv"
    if training_args.do_train:
        train_samples = []
        with open(nli_dataset_path) as f:
            #test_dict = json.load(f)
            l = f.readline()
            #pdb.set_trace()
            for item in tqdm(f):
                premise, hypothesis, label, _ = item.strip().split("\t")
                label_id = label2int[label]
                train_samples.append({'premise': premise, 'hypothesis':hypothesis, 'label':label2int[label]})

        
        train_dataset = datasets.Dataset.from_pandas(pd.DataFrame(data=train_samples))
    # pdb.set_trace()
    # label_list = train_dataset.features["label"].names
    # Read STSbenchmark dataset and use it as development set
    logging.info("Read STSbenchmark dev dataset")
    
    if training_args.do_eval:
        dev_samples = []
        with open(sts_dataset_path) as f:
            #test_dict = json.load(f)
            l = f.readline()
            #pdb.set_trace()
            for item in tqdm(f):
                premise, hypothesis, label, _ = item.strip().split("\t")
                label_id = label2int[label]
                dev_samples.append({'premise': premise, 'hypothesis':hypothesis, 'label':label2int[label]})
        # Labels
        eval_dataset = datasets.Dataset.from_pandas(pd.DataFrame(data=dev_samples))

    # Load pretrained model and tokenizer
    # In distributed training, the .from_pretrained methods guarantee that only one local process can concurrently
    # download model & vocab.
    #pdb.set_trace()
    test_dataset_path = "data/processed/test_"+model_args.language+".tsv"
    if training_args.do_predict:
        test_samples = []
        with open(test_dataset_path) as f:
            #test_dict = json.load(f)
            l = f.readline()
            #pdb.set_trace()
            for item in tqdm(f):
                premise, hypothesis, label, _ = item.strip().split("\t")
                label_id = label2int[label]
                test_samples.append({'premise': premise, 'hypothesis':hypothesis, 'label':label2int[label]})

        
        predict_dataset = datasets.Dataset.from_pandas(pd.DataFrame(data=test_samples))

    num_labels = len(label_list)
    config = DebertaConfig.from_pretrained(
        model_args.config_name if model_args.config_name else model_args.model_name_or_path,
        num_labels=num_labels,
        id2label={str(i): label for i, label in enumerate(label_list)},
        label2id={label: i for i, label in enumerate(label_list)},
        finetuning_task="xnli",
        cache_dir=model_args.cache_dir,
        revision=model_args.model_revision,
        token=model_args.token,
        trust_remote_code=model_args.trust_remote_code,
    )
    tokenizer = DebertaTokenizer.from_pretrained(
        model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path,
        do_lower_case=model_args.do_lower_case,
        cache_dir=model_args.cache_dir,
        use_fast=model_args.use_fast_tokenizer,
        revision=model_args.model_revision,
        token=model_args.token,
        trust_remote_code=model_args.trust_remote_code,
    )
    model = DebertaForSequenceClassification.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,
        token=model_args.token,
        trust_remote_code=model_args.trust_remote_code,
        ignore_mismatched_sizes=model_args.ignore_mismatched_sizes,
    )

    # Preprocessing the datasets
    # Padding strategy
    if data_args.pad_to_max_length:
        padding = "max_length"
    else:
        # We will pad later, dynamically at batch creation, to the max sequence length in each batch
        padding = False

    def preprocess_function(examples):
        # Tokenize the texts
        return tokenizer(
            examples["premise"],
            examples["hypothesis"],
            padding=padding,
            max_length=data_args.max_seq_length,
            truncation=True,
        )

    if training_args.do_train:
        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,
                load_from_cache_file=not data_args.overwrite_cache,
                desc="Running tokenizer on train dataset",
            )
        # Log a few random samples from the training set:
        for index in random.sample(range(len(train_dataset)), 3):
            logger.info(f"Sample {index} of the training set: {train_dataset[index]}.")

    if training_args.do_eval:
        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,
                batched=True,
                load_from_cache_file=not data_args.overwrite_cache,
                desc="Running tokenizer on validation dataset",
            )

    if training_args.do_predict:
        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,
                batched=True,
                load_from_cache_file=not data_args.overwrite_cache,
                desc="Running tokenizer on prediction dataset",
            )

    # Get the metric function
    metric = evaluate.load("xnli")

    # You can define your custom compute_metrics function. It takes an `EvalPrediction` object (a namedtuple with a
    # predictions and label_ids field) and has to return a dictionary string to float.
    def compute_metrics(p: EvalPrediction):
        preds = p.predictions[0] if isinstance(p.predictions, tuple) else p.predictions
        preds = np.argmax(preds, axis=1)
        return metric.compute(predictions=preds, references=p.label_ids)

    # Data collator will default to DataCollatorWithPadding, so we change it if we already did the padding.
    if data_args.pad_to_max_length:
        data_collator = default_data_collator
    elif training_args.fp16:
        data_collator = DataCollatorWithPadding(tokenizer, pad_to_multiple_of=8)
    else:
        data_collator = None

    # Initialize our Trainer
    trainer = Trainer(
        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,
        compute_metrics=compute_metrics,
        tokenizer=tokenizer,
        data_collator=data_collator,
    )

    # 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)
        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.save_model()  # Saves the tokenizer too for easy upload

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

    # Evaluation
    if training_args.do_eval:
        logger.info("*** Evaluate ***")
        metrics = trainer.evaluate(eval_dataset=eval_dataset)

        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)

    # Prediction
    if training_args.do_predict:
        tau = model_args.tau
        logger.info("*** Predict ***")
        predictions, labels, metrics = trainer.predict(predict_dataset, metric_key_prefix="predict")
        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)
        #pdb.set_trace()
        probs = F.softmax(torch.tensor(predictions))
        max_probs = [max(prob).item() for prob in probs]
        #pdb.set_trace()
        f_w=open("deberta_ft_outputs/" + model_args.language + "_max_probs.json", "w")
        json.dump(max_probs, f_w)
        targets = torch.tensor(labels)

        ece_metric = MulticlassCalibrationError(num_classes=3, n_bins=10, norm='l1')
        ece_value = ece_metric(probs, targets)

        print(f"ECE L1: {ece_value.item()}")
        #pdb.set_trace()
        ece_metric = MulticlassCalibrationError(num_classes=3, n_bins=10, norm='max')
        ece_value = ece_metric(probs, targets)
        print(f"ECE max: {ece_value.item()}")
        predictions = np.argmax(predictions, axis=1)
        #pdb.set_trace()
        output_predict_file = os.path.join("deberta_ft_outputs", model_args.language+"_predictions.json")
        with open(output_predict_file, "w") as writer:
            write_preds = []
            for pred, gold in zip(predictions.tolist(), labels.tolist()):
                #pdb.set_trace()
                write_preds.append(id2label[pred])
            json.dump(write_preds, writer)


        corr=[1 if p==g else 0 for p,g in zip(predictions,labels)]
        corr_tau = [corr[j] for j,it in enumerate(max_probs) if it >= tau]
        print("Fraction of more than ", str(tau), "is: - ", sum(corr_tau)*100/len(corr_tau))

if __name__ == "__main__":
    main()