run_summarization.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.

from dataclasses import dataclass, field
from typing import Optional

import logging
import os
import sys
import nltk  # Here to have a nice missing dependency error message early on
import numpy as np
import torch
from filelock import FileLock
import seaborn as sns
import pandas as pd
import matplotlib.pyplot as plt
import copy
import datasets
import evaluate
import transformers
from sklearn.metrics import f1_score
from datasets import load_dataset
from peft import LoraConfig, TaskType, get_peft_model, PeftModel
from transformers import (
    AutoConfig,
    AutoModelForSeq2SeqLM,
    AutoTokenizer,
    DataCollatorForSeq2Seq,
    HfArgumentParser,
    Seq2SeqTrainingArguments,
    set_seed,
)
from transformers.trainer_utils import get_last_checkpoint
from transformers.utils import check_min_version, is_offline_mode, send_example_telemetry
from transformers.utils.versions import require_version

from sum_lib import (
    ModelArguments,
    DataTrainingArguments,
    SumTrainer,
    adjust_training_args,
)
from models import (
    T5ForConditionalGeneration,
    DeployT5ForConditionalGeneration,
    LongT5ForConditionalGeneration,
    DeployLongT5ForConditionalGeneration,
)
from util import (
    AdditionalArguments,
    additional_args,
    update_autoconfig,
)

import wandb

# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
# check_min_version("4.28.0.dev0")
# require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/summarization/requirements.txt")

logger = logging.getLogger(__name__)

TESTING = True

summarization_name_mapping = {
    "amazon_reviews_multi": ("review_body", "review_title"),
    "big_patent": ("description", "abstract"),
    "cnn_dailymail": ("article", "highlights"),
    "orange_sum": ("text", "summary"),
    "pn_summary": ("article", "summary"),
    "psc": ("extract_text", "summary_text"),
    "samsum": ("dialogue", "summary"),
    "thaisum": ("body", "summary"),
    "xglue": ("news_body", "news_title"),
    "xsum": ("document", "summary"),
    "wiki_summary": ("article", "highlights"),
    "multi_news": ("document", "summary"),
}

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)


def main(model_args, data_args, training_args, additional_args, model_cls, trainer_cls, jupyter=False):
    # 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:
    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 'summarize: ' `"
        )

    # 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.
    if data_args.dataset_name is not None:
        # Downloading and loading a dataset from the hub.
        raw_datasets = load_dataset(
            data_args.dataset_name,
            data_args.dataset_config_name,
            cache_dir=model_args.cache_dir,
            use_auth_token=True if model_args.use_auth_token else None,
        )
    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.
    if not additional_args.use_lora or training_args.do_train:
        config_name = model_args.config_name if model_args.config_name else model_args.model_name_or_path
        tokenizer_name = model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path
        model_name = model_args.model_name_or_path
    else:
        lora_config = LoraConfig.from_pretrained(model_args.model_name_or_path)
        config_name = tokenizer_name = model_name = lora_config.base_model_name_or_path

    config = AutoConfig.from_pretrained(
        config_name,
        cache_dir=model_args.cache_dir,
        revision=model_args.model_revision,
        use_auth_token=True if model_args.use_auth_token else None,
    )
    config = update_autoconfig(
        config,
        additional_args,
        max_answer_length=data_args.max_target_length
    )
    
    tokenizer = AutoTokenizer.from_pretrained(
        tokenizer_name,
        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 = model_cls.from_pretrained(
        model_name,
        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,
    )
        
    if additional_args.use_lora:
        if training_args.do_train:
            lora_config = LoraConfig(
                task_type=TaskType.SEQ_2_SEQ_LM, r=additional_args.lora_rank, 
                lora_alpha=additional_args.lora_alpha, lora_dropout=additional_args.lora_dropout,
                target_modules=additional_args.lora_target_modules,
            )
            model = get_peft_model(model, lora_config)
            model.print_trainable_parameters()
        else:
            model = PeftModel.from_pretrained(model, model_args.model_name_or_path, config=lora_config)
            model = model.merge_and_unload()

    # 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:
        if "train" not in raw_datasets:
            raise ValueError("--do_train requires a train dataset")
        column_names = raw_datasets["train"].column_names
    elif training_args.do_eval:
        if "validation" not in raw_datasets:
            raise ValueError("--do_eval requires a validation dataset")
        column_names = raw_datasets["validation"].column_names
    elif training_args.do_predict:
        if "test" not in raw_datasets:
            raise ValueError("--do_predict requires a test dataset")
        column_names = raw_datasets["test"].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 = summarization_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[1]
    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)}"
            )

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

        inputs, targets = [], []
        for i in range(len(examples[text_column])):
            if examples[text_column][i] and examples[summary_column][i]:
                inputs.append(examples[text_column][i])
                targets.append(examples[summary_column][i])

        inputs = [prefix + inp for inp in inputs]
        max_target_length = data_args.max_target_length
        # Encoder-Decoder Language Models

        padding = "max_length" if data_args.pad_to_max_length else False

        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:
        train_dataset = raw_datasets["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,
                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
        eval_dataset = raw_datasets["validation"]
        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,
                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
        predict_dataset = raw_datasets["test"]
        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,
                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 = evaluate.load("rouge")

    def postprocess_text(preds, labels):
        preds = [pred.strip() for pred in preds]
        labels = [label.strip() for label in labels]

        # rougeLSum expects newline after each sentence
        preds = ["\n".join(nltk.sent_tokenize(pred)) for pred in preds]
        labels = ["\n".join(nltk.sent_tokenize(label)) for label in labels]

        return preds, labels

    def compute_metrics(eval_preds, compute_metrics=True):
        if compute_metrics:
            preds, labels, _ = eval_preds
        else:
            preds, labels = eval_preds
        if isinstance(preds, tuple):
            preds = preds[0]
            
        try:
            decoded_preds = tokenizer.batch_decode(preds, skip_special_tokens=True)
        except:
            
            decoded_preds = tokenizer.batch_decode(np.where(preds != -100, preds, tokenizer.pad_token_id), 
                                                   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)
            
        try:
            decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)
        except:
            decoded_labels = tokenizer.batch_decode(np.where(labels != -100, labels, tokenizer.pad_token_id), 
                                                    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, use_stemmer=True)
        result = {k: round(v * 100, 4) for k, v in result.items()}
        prediction_lens = [np.count_nonzero(pred != tokenizer.pad_token_id) for pred in preds]
        result["gen_len"] = np.mean(prediction_lens)
        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
    )    
    # adjust training arguments
    training_args = adjust_training_args(training_args, additional_args)

    # Initialize our Trainer
    trainer = trainer_cls(
        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
    )

    # 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)
        if not additional_args.use_lora: 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()

        if additional_args.use_lora:
            model.save_pretrained(training_args.output_dir)  # save adapter_config.json
            model.base_model.save_pretrained(training_args.output_dir)  # save config.json

    # Evaluation
    results = {}
    if training_args.do_eval:
        logger.info("*** Evaluate ***")
        if training_args.include_inputs_for_metrics:
            output = trainer.evaluate(metric_key_prefix="eval")
            metrics = output.metrics
            if additional_args.count_flops:
                final_flops = model.decoder.flop_counter/model.decoder.count_passes
                wandb.log({"FLOP/Token": final_flops})
                flops_sample = model.decoder.flop_counter/len(eval_dataset)
                wandb.log({"FLOP/Sample": flops_sample})
                total_flops = model.decoder.flop_counter
                wandb.log({"Total FLOP": total_flops})
        else:
            metrics = trainer.evaluate(metric_key_prefix="eval")

        if additional_args.plotting_logits:
            data = model.decoder.graph_top_k_list
            max_length = max(len(arr) for arr in data)

            # Pad arrays with NaNs to ensure they are all the same length
            padded_data = [np.pad(np.array(arr, dtype=float),  # Convert array to float
                        (0, max_length - len(arr)),
                        mode='constant',
                        constant_values=np.nan)
                for arr in data]
            


            # Convert the list of arrays into a single NumPy array
            padded_array = np.array(padded_data)

            # Converting the array to a DataFrame for easier handling in seaborn
            df = pd.DataFrame(padded_array)
            table = wandb.Table(dataframe=df)
            wandb.log({'plotting_logits': table})           

            # Creating a boxplot
            plt.figure(figsize=(12, 8))
            sns.boxplot(data=df)
            plt.title('Rank of the final predicted token at each layer', fontsize=20)
            plt.xlabel('Layer', fontsize=16)
            plt.ylabel('Rank of the final predicted token', fontsize=16)
            plt.grid(True)
            file_path = "plots/boxplot_topk_rank_eval" + data_args.dataset_name.replace("/", "_") + "_" + model_args.model_name_or_path.replace("/", "_") + ".png"
            plt.savefig(file_path)
            wandb.log({"Boxplot": wandb.Image(file_path)})
        
        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:
        logger.info("*** Predict ***")

        predict_results = trainer.predict(predict_dataset, metric_key_prefix="predict")
        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))

    kwargs = {"finetuned_from": model_args.model_name_or_path, "tasks": "summarization"}
    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)

    if not jupyter:
        return results, metrics
    else:
        return trainer, metrics


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.
    #os.environ["WANDB_DISABLED"] = "true"
    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, Seq2SeqTrainingArguments, AdditionalArguments))
    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, additional_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
    else:
        model_args, data_args, training_args, additional_args = parser.parse_args_into_dataclasses()
    
    if 't5' in model_args.model_name_or_path:
        if data_args.dataset_name in ["cnn_dailymail", "xsum", "samsum"]:
            model_cls = T5ForConditionalGeneration if not additional_args.deploy_scenario \
                else DeployT5ForConditionalGeneration
        elif data_args.dataset_name in ["multi_news", "big_patent"]:
            model_cls = LongT5ForConditionalGeneration if not additional_args.deploy_scenario \
                else DeployLongT5ForConditionalGeneration
    else:
        model_cls = T5ForConditionalGeneration if not additional_args.deploy_scenario \
            else DeployT5ForConditionalGeneration

    trainer_cls = SumTrainer
    training_args.include_inputs_for_metrics = True

    try:
        wandb.login()

        wandb.init(
            # set the wandb project where this run will be logged
            project="ProjectName",
            entity="ProjectDirectory",
            # track hyperparameters and run metadata
            config={
                "dataset": data_args.dataset_name,
                "model": model_args.model_name_or_path, 
                "exit_conf_type": additional_args.exit_conf_type,
                "exit_conf_threshold": additional_args.exit_conf_threshold,
                "exit_min_layer": additional_args.exit_min_layer,
                "type_vocab_reduct": additional_args.type_vocab_reduct,
                },
            mode="disabled" if False else "online",
        )
    except:
        print("Wandb repository is not available")
        pass

    main(model_args, data_args, training_args, additional_args, model_cls, trainer_cls)

    try:
        wandb.finish()
    except:
        pass