unilm/adalm/finetune/run_pico.py

# 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 the library models for sequence classification on GLUE (Bert, XLM, XLNet, RoBERTa)."""

from __future__ import absolute_import, division, print_function

import argparse
import glob
import logging
import os
import random
import json

import numpy as np
import torch
from sklearn.metrics import matthews_corrcoef, f1_score
from sklearn.metrics import cohen_kappa_score, precision_score, recall_score, precision_recall_fscore_support
from torch.utils.data import (DataLoader, RandomSampler, SequentialSampler,
                              TensorDataset)
from torch.utils.data.distributed import DistributedSampler
from torch.nn import CrossEntropyLoss

try:
    from torch.utils.tensorboard import SummaryWriter
except:
    from tensorboardX import SummaryWriter

from tqdm import tqdm, trange

from transformers import (
    WEIGHTS_NAME,
    AdamW,
    BertConfig,
    BertForTokenClassification,
    BertTokenizer,
    DistilBertConfig,
    DistilBertForTokenClassification,
    DistilBertTokenizer,
    RobertaConfig,
    RobertaForTokenClassification,
    RobertaTokenizer,
    XLMRobertaConfig,
    XLMRobertaForTokenClassification,
    XLMRobertaTokenizer
)

from transformers import AdamW, get_linear_schedule_with_warmup

from utils_ner import convert_examples_to_features, get_labels, read_examples_from_file

logger = logging.getLogger(__name__)


ALL_MODELS = sum(
    (
        tuple(conf.pretrained_config_archive_map.keys())
        for conf in (BertConfig, RobertaConfig, DistilBertConfig, XLMRobertaConfig)
    ),
    (),
)


MODEL_CLASSES = {
    "bert": (BertConfig, BertForTokenClassification, BertTokenizer),
    "roberta": (RobertaConfig, RobertaForTokenClassification, RobertaTokenizer),
    "distilbert": (DistilBertConfig, DistilBertForTokenClassification, DistilBertTokenizer),
    "xlmroberta": (XLMRobertaConfig, XLMRobertaForTokenClassification, XLMRobertaTokenizer),
}

TOKENIZER_ARGS = ["do_lower_case", "strip_accents", "keep_accents", "use_fast"]

def set_seed(args):
    random.seed(args.seed)
    np.random.seed(args.seed)
    torch.manual_seed(args.seed)
    if args.n_gpu > 0:
        torch.cuda.manual_seed_all(args.seed)

def get_f1(prec, rec):
  return 2*prec*rec/(prec+rec)

def token_f1(true, pred, labels):

  print(true[:30])
  print(pred[:30])
  print(labels)
  total_f1 = 0.0
  class_scores = zip(labels, precision_score(true,pred,labels,average=None), recall_score(true,pred,labels,average=None))
  for label, prec, rec in class_scores:
    print('Label: %s' %label)
    if label != 'O':
        total_f1 += get_f1(prec, rec)
    print('\tf1        = %f' %get_f1(prec, rec))
    print('\tprecision = %f' %prec)
    print('\trecall    = %f' %rec)
  return total_f1/3

def train(args, train_dataset, model, tokenizer, labels, pad_token_label_id):
    """ Train the model """
    if args.local_rank in [-1, 0]:
        tb_writer = SummaryWriter()

    args.train_batch_size = args.per_gpu_train_batch_size * max(1, args.n_gpu)
    train_sampler = RandomSampler(train_dataset) if args.local_rank == -1 else DistributedSampler(train_dataset)
    train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size)

    if args.max_steps > 0:
        t_total = args.max_steps
        args.num_train_epochs = args.max_steps // (len(train_dataloader) // args.gradient_accumulation_steps) + 1
    else:
        t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs


    # Prepare optimizer and schedule (linear warmup and decay)
    if args.warmup_ratio > 0:
        args.warmup_steps = int(t_total*args.warmup_ratio)


    # Prepare optimizer and schedule (linear warmup and decay)
    no_decay = ["bias", "LayerNorm.weight"]
    optimizer_grouped_parameters = [
        {
            "params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)],
            "weight_decay": args.weight_decay,
        },
        {"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
    ]
    optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
    scheduler = get_linear_schedule_with_warmup(
        optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
    )


    # Check if saved optimizer or scheduler states exist
    if os.path.isfile(os.path.join(args.model_name_or_path, "optimizer.pt")) and os.path.isfile(
        os.path.join(args.model_name_or_path, "scheduler.pt")
    ):
        # Load in optimizer and scheduler states
        optimizer.load_state_dict(torch.load(os.path.join(args.model_name_or_path, "optimizer.pt")))
        scheduler.load_state_dict(torch.load(os.path.join(args.model_name_or_path, "scheduler.pt")))

    if args.fp16:
        try:
            from apex import amp
        except ImportError:
            raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.")
        model, optimizer = amp.initialize(model, optimizer, opt_level=args.fp16_opt_level)

    # multi-gpu training (should be after apex fp16 initialization)
    if args.n_gpu > 1:
        model = torch.nn.DataParallel(model)

    # Distributed training (should be after apex fp16 initialization)
    if args.local_rank != -1:
        model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.local_rank],
                                                          output_device=args.local_rank,
                                                          find_unused_parameters=True)

    # Train!
    logger.info("***** Running training *****")
    logger.info("  Num examples = %d", len(train_dataset))
    logger.info("  Num Epochs = %d", args.num_train_epochs)
    logger.info("  Instantaneous batch size per GPU = %d", args.per_gpu_train_batch_size)
    logger.info("  Total train batch size (w. parallel, distributed & accumulation) = %d",
                   args.train_batch_size * args.gradient_accumulation_steps * (torch.distributed.get_world_size() if args.local_rank != -1 else 1))
    logger.info("  Gradient Accumulation steps = %d", args.gradient_accumulation_steps)
    logger.info("  Total optimization steps = %d", t_total)

    metric_for_best = args.metric_for_choose_best_checkpoint
    best_performance = None
    best_epoch = None
    global_step = 0
    tr_loss, logging_loss = 0.0, 0.0
    model.zero_grad()
    train_iterator = trange(int(args.num_train_epochs), desc="Epoch", disable=args.local_rank not in [-1, 0])
    set_seed(args)  # Added here for reproductibility (even between python 2 and 3)
    for _ in train_iterator:
        if args.disable_tqdm:
            epoch_iterator = train_dataloader
        else:
            epoch_iterator = tqdm(train_dataloader, desc="Iteration", disable=args.local_rank not in [-1, 0])
        for step, batch in enumerate(epoch_iterator):
            model.train()
            batch = tuple(t.to(args.device) for t in batch)
            inputs = {'input_ids':      batch[0],
                      'attention_mask': batch[1],
                      'labels':         batch[3]}
            if args.model_type != 'distilbert':
                inputs['token_type_ids'] = batch[2] if args.model_type in ['bert', 'xlnet', 'unilm', 'adapterbert'] else None  # XLM, DistilBERT and RoBERTa don't use segment_ids
            outputs = model(**inputs)
            loss = outputs[0]  # model outputs are always tuple in transformers (see doc)

            if args.n_gpu > 1:
                loss = loss.mean() # mean() to average on multi-gpu parallel training
            if args.gradient_accumulation_steps > 1:
                loss = loss / args.gradient_accumulation_steps

            if args.fp16:
                with amp.scale_loss(loss, optimizer) as scaled_loss:
                    scaled_loss.backward()
            else:
                loss.backward()

            tr_loss += loss.item()
            if (step + 1) % args.gradient_accumulation_steps == 0:
                if args.max_grad_norm > 0:
                    if args.fp16:
                        torch.nn.utils.clip_grad_norm_(amp.master_params(optimizer), args.max_grad_norm)
                    else:
                        torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_grad_norm)

                optimizer.step()
                scheduler.step()  # Update learning rate schedule
                model.zero_grad()
                global_step += 1
                epoch_iterator.set_description('Iter (loss=%5.3f) lr=%9.7f' % (loss.item(), scheduler.get_lr()[0]))
                if args.local_rank in [-1, 0] and args.logging_steps > 0 and global_step % args.logging_steps == 0:
                    logs = {}
                    loss_scalar = (tr_loss - logging_loss) / args.logging_steps
                    learning_rate_scalar = scheduler.get_lr()[0]
                    logs['learning_rate'] = learning_rate_scalar
                    logs['loss'] = loss_scalar
                    logging_loss = tr_loss

                    for key, value in logs.items():
                        tb_writer.add_scalar(key, value, global_step)
                    logger.info(json.dumps({**logs, **{'step': global_step}}))

            if args.max_steps > 0 and global_step > args.max_steps:
                if not args.disable_tqdm:
                    epoch_iterator.close()
                break

        if args.local_rank in [-1, 0]:
            logs = {}
            if args.local_rank == -1 and args.evaluate_during_training:  # Only evaluate when single GPU otherwise metrics may not average well
                results = evaluate(args, model, tokenizer, prefix='epoch-{}'.format(_ + 1))
                for key, value in results.items():
                    eval_key = 'eval_{}'.format(key)
                    logs[eval_key] = value
                    if metric_for_best is None:
                        metric_for_best = key
                if best_epoch is None or best_performance[metric_for_best] < results[metric_for_best]:
                    best_epoch = 'epoch-{}'.format(_ + 1)
                    best_performance = results

            loss_scalar = (tr_loss - logging_loss) / args.logging_steps
            learning_rate_scalar = scheduler.get_lr()[0]
            logs['learning_rate'] = learning_rate_scalar
            logs['loss'] = loss_scalar
            logging_loss = tr_loss

            for key, value in logs.items():
                tb_writer.add_scalar(key, value, global_step)
            print(json.dumps({**logs, **{'step': global_step}}))

            # Save model checkpoint
            output_dir = os.path.join(args.output_dir, 'epoch-{}'.format(_ + 1))
            if not os.path.exists(output_dir):
                os.makedirs(output_dir)
            model_to_save = model.module if hasattr(model, 'module') else model  # Take care of distributed/parallel training
            model_to_save.save_pretrained(output_dir)
            torch.save(args, os.path.join(output_dir, 'training_args.bin'))
            logger.info("Saving model checkpoint to %s", output_dir)

        if args.max_steps > 0 and global_step > args.max_steps:
            train_iterator.close()
            break

    if args.local_rank in [-1, 0]:
        tb_writer.close()

    if best_epoch is not None:
        logger.info(" ***************** Best checkpoint: {}, choosed by {} *****************".format(
            best_epoch, metric_for_best))
        logger.info("Best performance = %s" % json.dumps(best_performance))
        save_best_result(best_epoch, best_performance, args.output_dir)

    return global_step, tr_loss / global_step


def save_best_result(best_epoch, best_performance, output_dir):
    best_performance["checkpoint"] = best_epoch
    with open(os.path.join(output_dir, "best_performance.json"), mode="w") as writer:
        writer.write(json.dumps(best_performance, indent=2))


def evaluate(args, model, tokenizer, labels, pad_token_label_id, mode, prefix=""):
    eval_dataset = load_and_cache_examples(args, tokenizer, labels, pad_token_label_id, mode=mode)

    args.eval_batch_size = args.per_gpu_eval_batch_size * max(1, args.n_gpu)
    # Note that DistributedSampler samples randomly
    eval_sampler = SequentialSampler(eval_dataset) if args.local_rank == -1 else DistributedSampler(eval_dataset)
    eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size)

    # multi-gpu evaluate
    if args.n_gpu > 1:
        model = torch.nn.DataParallel(model)

    # Eval!
    logger.info("***** Running evaluation %s *****", prefix)
    logger.info("  Num examples = %d", len(eval_dataset))
    logger.info("  Batch size = %d", args.eval_batch_size)
    eval_loss = 0.0
    nb_eval_steps = 0
    preds = None
    out_label_ids = None
    model.eval()
    for batch in tqdm(eval_dataloader, desc="Evaluating"):
        batch = tuple(t.to(args.device) for t in batch)

        with torch.no_grad():
            inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels": batch[3]}
            if args.model_type != "distilbert":
                inputs["token_type_ids"] = (
                    batch[2] if args.model_type in ["bert", "xlnet", "adapterbert"] else None
                )  # XLM and RoBERTa don"t use segment_ids
            outputs = model(**inputs)
            tmp_eval_loss, logits = outputs[:2]

            if args.n_gpu > 1:
                tmp_eval_loss = tmp_eval_loss.mean()  # mean() to average on multi-gpu parallel evaluating

            eval_loss += tmp_eval_loss.item()
        nb_eval_steps += 1
        if preds is None:
            preds = logits.detach().cpu().numpy()
            out_label_ids = inputs["labels"].detach().cpu().numpy()
        else:
            preds = np.append(preds, logits.detach().cpu().numpy(), axis=0)
            out_label_ids = np.append(out_label_ids, inputs["labels"].detach().cpu().numpy(), axis=0)

    eval_loss = eval_loss / nb_eval_steps
    preds = np.argmax(preds, axis=2)

    label_map = {i: label for i, label in enumerate(labels)}

    out_label_list = [[] for _ in range(out_label_ids.shape[0])]
    preds_list = [[] for _ in range(out_label_ids.shape[0])]

    for i in range(out_label_ids.shape[0]):
        for j in range(out_label_ids.shape[1]):
            if out_label_ids[i, j] != pad_token_label_id:
                out_label_list[i].append(label_map[out_label_ids[i][j]])
                preds_list[i].append(label_map[preds[i][j]])
    out_labels = [i for item in out_label_list for i in item]
    preds_labels = [i for item in preds_list for i in item]
    results = {
        "loss": eval_loss,
        "f1": token_f1(true = out_labels,pred = preds_labels, labels = labels),
    }

    logger.info("***** Eval results %s *****", prefix)
    for key in sorted(results.keys()):
        logger.info("  %s = %s", key, str(results[key]))

    output_file = os.path.join(args.output_dir, "eval_out.txt")
    with open(output_file, "w+", encoding="utf-8") as f:
        for line in tqdm(preds_list):
            line = " ".join(line) + "\n"
            f.write(line)

    return results, preds_list

def test(args, model, tokenizer, labels, pad_token_label_id, mode, prefix=""):
    test_dataset = load_and_cache_examples(args, tokenizer, labels, pad_token_label_id, mode="test")

    args.test_batch_size = args.per_gpu_eval_batch_size * max(1, args.n_gpu)

    test_sampler = SequentialSampler(test_dataset) if args.local_rank == -1 else DistributedSampler(test_dataset)

    test_dataloader = DataLoader(test_dataset, sampler=test_sampler, batch_size=args.test_batch_size)

    if args.n_gpu > 1:
        model = torch.nn.DataParallel(model)

    logger.info("***** Running Prediction %s *****", prefix)
    logger.info("  Num examples = %d", len(test_dataset))
    logger.info("  Batch size = %d", args.test_batch_size)
    eval_loss = 0.0
    nb_eval_steps = 0
    preds = None
    out_label_ids = None
    model.eval()
    for batch in tqdm(test_dataloader, desc="Prediction"):
        batch = tuple(t.to(args.device) for t in batch)

        with torch.no_grad():
            inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels": batch[3]}
            if args.model_type != "distilbert":
                inputs["token_type_ids"] = (
                    batch[2] if args.model_type in ["bert", "xlnet", "adapterbert"] else None
                )  # XLM and RoBERTa don"t use segment_ids
            outputs = model(**inputs)
            tmp_eval_loss, logits = outputs[:2]

            if args.n_gpu > 1:
                tmp_eval_loss = tmp_eval_loss.mean()  # mean() to average on multi-gpu parallel evaluating

            eval_loss += tmp_eval_loss.item()
        nb_eval_steps += 1
        if preds is None:
            preds = logits.detach().cpu().numpy()
            out_label_ids = inputs["labels"].detach().cpu().numpy()
        else:
            preds = np.append(preds, logits.detach().cpu().numpy(), axis=0)
            out_label_ids = np.append(out_label_ids, inputs["labels"].detach().cpu().numpy(), axis=0)

    eval_loss = eval_loss / nb_eval_steps
    preds = np.argmax(preds, axis=2)

    label_map = {i: label for i, label in enumerate(labels)}

    out_label_list = [[] for _ in range(out_label_ids.shape[0])]
    preds_list = [[] for _ in range(out_label_ids.shape[0])]

    for i in range(out_label_ids.shape[0]):
        for j in range(out_label_ids.shape[1]):
            if out_label_ids[i, j] != pad_token_label_id:
                out_label_list[i].append(label_map[out_label_ids[i][j]])
                preds_list[i].append(label_map[preds[i][j]])
    
    out_file = os.path.join(args.output_dir, "predict.txt")
    
    out_labels = [i for item in out_label_list for i in item]
    preds_labels = [i for item in preds_list for i in item]
    results = {
        "loss": eval_loss,
        "f1": token_f1(true = out_labels,pred = preds_labels, labels = labels),
    }
    print(out_label_list[0])
    print(preds_list[0])
    logger.info("write results into {}".format(out_file))
    output_eval_file = os.path.join(args.output_dir, "predict_results.txt")
    with open(output_eval_file, "w") as writer:
        logger.info("***** Predict results {} *****".format(prefix))
        writer.write(json.dumps(results, indent=2))
        logger.info("Result = %s" % json.dumps(results, indent=2))
    with open(out_file, "w+", encoding="utf-8") as f:
        for line in preds_list:
            line = " ".join(line) + "\n"
            f.write(line)
    
    for key in sorted(results.keys()):
        logger.info("  %s = %s", key, str(results[key]))
    
    return results, preds_list

def load_and_cache_examples(args, tokenizer, labels, pad_token_label_id, mode):
    if args.local_rank not in [-1, 0] and not evaluate:
        torch.distributed.barrier()  # Make sure only the first process in distributed training process the dataset, and the others will use the cache

    # Load data features from cache or dataset file
    cached_features_file = os.path.join(
        args.data_dir,
        "cached_{}_{}_{}".format(
            mode, list(filter(None, args.model_name_or_path.split("/"))).pop(), str(args.max_seq_length)
        ),
    )
    if os.path.exists(cached_features_file) and not args.overwrite_cache:
        logger.info("Loading features from cached file %s", cached_features_file)
        features = torch.load(cached_features_file)
    else:
        logger.info("Creating features from dataset file at %s", args.data_dir)
        examples = read_examples_from_file(args.data_dir, mode)
        features = convert_examples_to_features(
            examples,
            labels,
            args.max_seq_length,
            tokenizer,
            cls_token_at_end=bool(args.model_type in ["xlnet"]),
            # xlnet has a cls token at the end
            cls_token=tokenizer.cls_token,
            cls_token_segment_id=2 if args.model_type in ["xlnet"] else 0,
            sep_token=tokenizer.sep_token,
            sep_token_extra=bool(args.model_type in ["roberta"]),
            # roberta uses an extra separator b/w pairs of sentences, cf. github.com/pytorch/fairseq/commit/1684e166e3da03f5b600dbb7855cb98ddfcd0805
            pad_on_left=bool(args.model_type in ["xlnet"]),
            # pad on the left for xlnet
            pad_token=tokenizer.convert_tokens_to_ids([tokenizer.pad_token])[0],
            pad_token_segment_id=4 if args.model_type in ["xlnet"] else 0,
            pad_token_label_id=pad_token_label_id,
            mode=mode,
        )
        if args.local_rank in [-1, 0]:
            logger.info("Saving features into cached file %s", cached_features_file)
            torch.save(features, cached_features_file)

    if args.local_rank == 0 and not evaluate:
        torch.distributed.barrier()  # Make sure only the first process in distributed training process the dataset, and the others will use the cache

    # Convert to Tensors and build dataset
    all_input_ids = torch.tensor([f.input_ids for f in features], dtype=torch.long)
    all_input_mask = torch.tensor([f.input_mask for f in features], dtype=torch.long)
    all_segment_ids = torch.tensor([f.segment_ids for f in features], dtype=torch.long)
    all_label_ids = torch.tensor([f.label_ids for f in features], dtype=torch.long)

    dataset = TensorDataset(all_input_ids, all_input_mask, all_segment_ids, all_label_ids)
    return dataset


def main():
    parser = argparse.ArgumentParser()

    ## Required parameters
    parser.add_argument("--data_dir", default=None, type=str, required=True,
                        help="The input data dir. Should contain the .tsv files (or other data files) for the task.")
    parser.add_argument("--model_type", default="unilm", type=str, 
                        help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys()))
    parser.add_argument("--model_name_or_path", default=None, type=str, required=True,
                        help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS))
    parser.add_argument("--output_dir", default=None, type=str, required=True,
                        help="The output directory where the model predictions and checkpoints will be written.")
    parser.add_argument('--disable_tqdm', action='store_true',
                        help='Disable the tqdm bar. ')

    ## Other parameters
    parser.add_argument("--labels", default="", type=str,
                        help="Path to a file containing all labels. If not specified, CoNLL-2003 labels are used.")
    parser.add_argument("--config_name", default="", type=str,
                        help="Pretrained config name or path if not the same as model_name")
    parser.add_argument("--tokenizer_name", default="", type=str,
                        help="Pretrained tokenizer name or path if not the same as model_name")
    parser.add_argument("--cache_dir", default="", type=str,
                        help="Where do you want to store the pre-trained models downloaded from s3")
    parser.add_argument("--max_seq_length", default=128, type=int,
                        help="The maximum total input sequence length after tokenization. Sequences longer "
                             "than this will be truncated, sequences shorter will be padded.")
    parser.add_argument("--do_train", action='store_true',
                        help="Whether to run training.")
    parser.add_argument("--do_eval", action='store_true',
                        help="Whether to run eval on the dev set.")
    parser.add_argument("--do_predict", action="store_true", help="Whether to run predictions on the test set.")
    parser.add_argument("--evaluate_during_training", action='store_true',
                        help="Rul evaluation during training at each logging step.")
    parser.add_argument("--do_lower_case", action='store_true',
                        help="Set this flag if you are using an uncased model.")
    parser.add_argument(
        "--keep_accents", action="store_const", const=True, help="Set this flag if model is trained with accents."
    )
    parser.add_argument(
        "--strip_accents", action="store_const", const=True, help="Set this flag if model is trained without accents."
    )
    parser.add_argument("--use_fast", action="store_const", const=True, help="Set this flag to use fast tokenization.")
    parser.add_argument("--per_gpu_train_batch_size", default=8, type=int,
                        help="Batch size per GPU/CPU for training.")
    parser.add_argument("--per_gpu_eval_batch_size", default=8, type=int,
                        help="Batch size per GPU/CPU for evaluation.")
    parser.add_argument('--gradient_accumulation_steps', type=int, default=1,
                        help="Number of updates steps to accumulate before performing a backward/update pass.")     
    parser.add_argument("--learning_rate", default=5e-5, type=float,
                        help="The initial learning rate for Adam.")
    parser.add_argument("--weight_decay", default=0.0, type=float,
                        help="Weight decay if we apply some.")
    parser.add_argument("--adam_epsilon", default=1e-8, type=float,
                        help="Epsilon for Adam optimizer.")
    parser.add_argument("--max_grad_norm", default=1.0, type=float,
                        help="Max gradient norm.")
    parser.add_argument("--num_train_epochs", default=3.0, type=float,
                        help="Total number of training epochs to perform.")
    parser.add_argument("--max_steps", default=-1, type=int,
                        help="If > 0: set total number of training steps to perform. Override num_train_epochs.")
    parser.add_argument("--warmup_ratio", default=0.1, type=float,
                        help="Linear warmup over warmup_ratio.")

    parser.add_argument('--logging_steps', type=int, default=50,
                        help="Log every X updates steps.")
    parser.add_argument("--save_steps", type=int, default=500, help="Save checkpoint every X updates steps.")
    parser.add_argument("--eval_all_checkpoints", action='store_true',
                        help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number")
    parser.add_argument("--no_cuda", action='store_true',
                        help="Avoid using CUDA when available")
    parser.add_argument('--overwrite_output_dir', action='store_true',
                        help="Overwrite the content of the output directory")
    parser.add_argument('--overwrite_cache', action='store_true',
                        help="Overwrite the cached training and evaluation sets")
    parser.add_argument('--seed', type=int, default=42,
                        help="random seed for initialization")
    parser.add_argument('--metric_for_choose_best_checkpoint', type=str, default=None,
                        help="Set the metric to choose the best checkpoint")

    parser.add_argument('--fp16', action='store_true',
                        help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit")
    parser.add_argument('--fp16_opt_level', type=str, default='O1',
                        help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']."
                             "See details at https://nvidia.github.io/apex/amp.html")
    parser.add_argument("--local_rank", type=int, default=-1,
                        help="For distributed training: local_rank")
    parser.add_argument('--server_ip', type=str, default='', help="For distant debugging.")
    parser.add_argument('--server_port', type=str, default='', help="For distant debugging.")
    args = parser.parse_args()

    if os.path.exists(args.output_dir) and os.listdir(args.output_dir) and args.do_train and not args.overwrite_output_dir:
        raise ValueError("Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome.".format(args.output_dir))

    # Setup distant debugging if needed
    if args.server_ip and args.server_port:
        # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script
        import ptvsd
        print("Waiting for debugger attach")
        ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True)
        ptvsd.wait_for_attach()

    # Setup CUDA, GPU & distributed training
    if args.local_rank == -1 or args.no_cuda:
        device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu")
        args.n_gpu = torch.cuda.device_count()
    else:  # Initializes the distributed backend which will take care of sychronizing nodes/GPUs
        torch.cuda.set_device(args.local_rank)
        device = torch.device("cuda", args.local_rank)
        torch.distributed.init_process_group(backend='nccl')
        args.n_gpu = 1
    args.device = device

    # Setup logging
    logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s -   %(message)s',
                        datefmt = '%m/%d/%Y %H:%M:%S',
                        level = logging.INFO if args.local_rank in [-1, 0] else logging.WARN)
    logger.warning("Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s",
                    args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16)

    # Set seed
    set_seed(args)

    # Prepare CONLL-2003 task
    labels = get_labels(args.labels)
    num_labels = len(labels)
    # Use cross entropy ignore index as padding label id so that only real label ids contribute to the loss later
    pad_token_label_id = CrossEntropyLoss().ignore_index

    # Load pretrained model and tokenizer
    if args.local_rank not in [-1, 0]:
        torch.distributed.barrier()  # Make sure only the first process in distributed training will download model & vocab

    args.model_type = args.model_type.lower()
    config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type]
    config = config_class.from_pretrained(
        args.config_name if args.config_name else args.model_name_or_path,
        num_labels=num_labels,
        id2label={str(i): label for i, label in enumerate(labels)},
        label2id={label: i for i, label in enumerate(labels)},
        cache_dir=args.cache_dir if args.cache_dir else None,
    )
    tokenizer_args = {k: v for k, v in vars(args).items() if v is not None and k in TOKENIZER_ARGS}
    logger.info("Tokenizer arguments: %s", tokenizer_args)
    tokenizer_name = args.tokenizer_name if args.tokenizer_name else args.model_name_or_path
    tokenizer = tokenizer_class.from_pretrained(
        args.tokenizer_name if args.tokenizer_name else args.model_name_or_path,
        cache_dir=args.cache_dir if args.cache_dir else None,
        **tokenizer_args,
    )
    if not hasattr(config, 'need_pooler') or config.need_pooler is not True:
        setattr(config, 'need_pooler', True)
    model = model_class.from_pretrained(
        args.model_name_or_path, config=config, 
        cache_dir=args.cache_dir if args.cache_dir else None)

    if args.local_rank == 0:
        torch.distributed.barrier()  # Make sure only the first process in distributed training will download model & vocab

    model.to(args.device)

    logger.info("Training/evaluation parameters %s", args)


    # Training
    if args.do_train:
        train_dataset = load_and_cache_examples(args, tokenizer, labels, pad_token_label_id, mode="train")
        global_step, tr_loss = train(args, train_dataset, model, tokenizer, labels, pad_token_label_id)
        logger.info(" global_step = %s, average loss = %s", global_step, tr_loss)

        tokenizer.save_pretrained(args.output_dir)

    # Saving best-practices: if you use defaults names for the model, you can reload it using from_pretrained()
    if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0):
        # Create output directory if needed
        if not os.path.exists(args.output_dir) and args.local_rank in [-1, 0]:
            os.makedirs(args.output_dir)

        logger.info("Saving model checkpoint to %s", args.output_dir)
        # Save a trained model, configuration and tokenizer using `save_pretrained()`.
        # They can then be reloaded using `from_pretrained()`
        model_to_save = (
            model.module if hasattr(model, "module") else model
        )  # Take care of distributed/parallel training
        model_to_save.save_pretrained(args.output_dir)
        tokenizer.save_pretrained(args.output_dir)

        # Good practice: save your training arguments together with the trained model
        torch.save(args, os.path.join(args.output_dir, "training_args.bin"))




    # Evaluation
    if args.do_eval and args.local_rank in [-1, 0]:
        tokenizer = tokenizer_class.from_pretrained(args.output_dir, **tokenizer_args)

        checkpoints = list(os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + '/**/' + WEIGHTS_NAME, recursive=True)))
        logging.getLogger("transformers.modeling_utils").setLevel(logging.WARN)  # Reduce logging
        logger.info("Evaluate the following checkpoints: %s", checkpoints)

        metric_for_best = args.metric_for_choose_best_checkpoint
        best_performance = None
        best_epoch = None

        for checkpoint in checkpoints:
            prefix = checkpoint.split('/')[-1] if checkpoint.find('checkpoint') != -1 else ""
            checkpoint_config = config_class.from_pretrained(checkpoint)
            model = model_class.from_pretrained(checkpoint, config=checkpoint_config)
            model.to(args.device)
            result, _ = evaluate(args, model, tokenizer, labels, pad_token_label_id, mode="dev", prefix=global_step)

            if metric_for_best is None:
                metric_for_best = list(result.keys())[-1]
            if best_epoch is None:
                best_epoch = checkpoint
                best_performance = result
            else:
                if best_performance[metric_for_best] < result[metric_for_best]:
                    best_performance = result
                    best_epoch = checkpoint

        if best_epoch is not None:
            logger.info(" ***************** Best checkpoint: {}, choosed by {} *****************".format(
                best_epoch, metric_for_best))
            logger.info("Best performance = %s" % json.dumps(best_performance))

            save_best_result(best_epoch, best_performance, args.output_dir)
            checkpoint = best_epoch
            checkpoint_config = config_class.from_pretrained(checkpoint)
            model = model_class.from_pretrained(checkpoint, config=checkpoint_config)
            model.to(args.device)
            result, _ = test(args, model, tokenizer, labels, pad_token_label_id, mode="test", prefix=global_step)


if __name__ == "__main__":
    main()