run_classification.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, Albert, XLM-RoBERTa)."""


import argparse
from distutils.dir_util import copy_tree, remove_tree
import glob
import json
import logging
import os
import random

import numpy as np
import torch
from torch.utils.data import DataLoader, RandomSampler, SequentialSampler, TensorDataset
from torch.utils.data.distributed import DistributedSampler
from tqdm import tqdm, trange

from transformers import (
    WEIGHTS_NAME,
    AdamW,
    AlbertConfig,
    AlbertForSequenceClassification,
    AlbertTokenizer,
    BertConfig,
    BertForSequenceClassification,
    BertTokenizer,
    DistilBertConfig,
    DistilBertForSequenceClassification,
    DistilBertTokenizer,
    RobertaConfig,
    RobertaForSequenceClassification,
    RobertaTokenizer,
    XLMConfig,
    XLMForSequenceClassification,
    XLMRobertaConfig,
    XLMRobertaForSequenceClassification,
    XLMRobertaTokenizer,
    XLMTokenizer,
    XLNetConfig,
    XLNetForSequenceClassification,
    XLNetTokenizer,
    get_linear_schedule_with_warmup,
    RobertaModel
)
from classification.re_processors import compute_metrics
from classification.re_processors import convert_examples_to_features
from classification.re_processors import output_modes
from classification.re_processors import processors
from classification.re_processors import RELATION_MAPPING, TITLE_NAMES
from classification.re_config import SPECIAL_TOKENS

from models.mtb import RobertaForRelationClassification


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


logger = logging.getLogger(__name__)

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

MODEL_CLASSES = {
    "bert": (BertConfig, BertForSequenceClassification, BertTokenizer),
    "xlnet": (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer),
    "xlm": (XLMConfig, XLMForSequenceClassification, XLMTokenizer),
    "roberta": (RobertaConfig, RobertaForSequenceClassification, RobertaTokenizer),
    "roberta-rc": (RobertaConfig, RobertaForRelationClassification, RobertaTokenizer),
    "roberta-asaf": (RobertaConfig, RobertaModel, RobertaTokenizer),
    "distilbert": (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer),
    "albert": (AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer),
    "xlmroberta": (XLMRobertaConfig, XLMRobertaForSequenceClassification, XLMRobertaTokenizer),
}


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 train(args, train_dataset, model, tokenizer):
    """ 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)
    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)

    global_step = 0
    epochs_trained = 0
    steps_trained_in_current_epoch = 0
    # Check if continuing training from a checkpoint
    if os.path.exists(args.model_name_or_path):
        # set global_step to gobal_step of last saved checkpoint from model path
        global_step = int(args.model_name_or_path.split("-")[-1].split("/")[0])
        epochs_trained = global_step // (len(train_dataloader) // args.gradient_accumulation_steps)
        steps_trained_in_current_epoch = global_step % (len(train_dataloader) // args.gradient_accumulation_steps)

        logger.info("  Continuing training from checkpoint, will skip to saved global_step")
        logger.info("  Continuing training from epoch %d", epochs_trained)
        logger.info("  Continuing training from global step %d", global_step)
        logger.info("  Will skip the first %d steps in the first epoch", steps_trained_in_current_epoch)

    tr_loss, logging_loss = 0.0, 0.0
    model.zero_grad()
    train_iterator = trange(
        epochs_trained, int(args.num_train_epochs), desc="Epoch", disable=args.local_rank not in [-1, 0],
    )
    set_seed(args)  # Added here for reproductibility
    best_f1, n_no_improve = 0, 0
    patience_ended = False
    for _ in train_iterator:
        epoch_iterator = tqdm(train_dataloader, desc="Iteration", disable=args.local_rank not in [-1, 0])
        for step, batch in enumerate(epoch_iterator):

            # Skip past any already trained steps if resuming training
            if steps_trained_in_current_epoch > 0:
                steps_trained_in_current_epoch -= 1
                continue

            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", "albert"] else None
                )  # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids
            if args.task_name == "docred" or args.task_name == "tacred":
                inputs["markers_mask"] = batch[4]
            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 or ((step + 1) < args.gradient_accumulation_steps and (step + 1) == len(epoch_iterator)):
                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

                if args.local_rank in [-1, 0] and args.logging_steps > 0 and global_step % args.logging_steps == 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)
                        for key, value in results.items():
                            eval_key = "eval_{}".format(key)
                            logs[eval_key] = value

                    if args.patience > 0:
                        if results[args.save_best_wrt_metric] > best_f1:
                            best_f1 = results[args.save_best_wrt_metric]
                            n_no_improve = 0
                        else:
                            n_no_improve += 1
                        logger.info("number of not improved %s", str(n_no_improve))

                        if n_no_improve >= args.patience:
                            patience_ended = True

                    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}}))

                if args.local_rank in [-1, 0] and args.save_steps > 0 and global_step % args.save_steps == 0:
                    # Save model checkpoint
                    output_dir = os.path.join(args.output_dir, "checkpoint-{}".format(global_step))
                    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)
                    tokenizer.save_pretrained(output_dir)

                    torch.save(args, os.path.join(output_dir, "training_args.bin"))
                    logger.info("Saving model checkpoint to %s", output_dir)

                    torch.save(optimizer.state_dict(), os.path.join(output_dir, "optimizer.pt"))
                    torch.save(scheduler.state_dict(), os.path.join(output_dir, "scheduler.pt"))
                    assert args.logging_steps == args.save_steps
                    torch.save(results, os.path.join(output_dir, "results.bin"))
                    logger.info("Saving optimizer and scheduler states to %s", output_dir)

                    best_dir = os.path.join(args.output_dir, "best")
                    if not os.path.exists(best_dir):
                        logger.info("Writing first model to best dir.")
                        os.makedirs(best_dir)
                        copy_tree(output_dir, best_dir)
                        logger.info("Done writing first model to best dir.")
                    elif results[args.save_best_wrt_metric] > torch.load(os.path.join(best_dir, "results.bin"))[args.save_best_wrt_metric]:
                        logger.info("Writing new best model to best dir.")
                        remove_tree(best_dir)
                        os.makedirs(best_dir)
                        copy_tree(output_dir, best_dir)
                        logger.info("Done writing new best model to best dir.")

                    if args.save_only_best:
                        remove_tree(output_dir)

            if patience_ended:
                break

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

        if patience_ended:
            break

        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()

    return global_step, tr_loss / global_step


def evaluate(args, model, tokenizer, prefix="", set_type="full_dev_eval"):
    # Loop to handle MNLI double evaluation (matched, mis-matched)
    eval_task_names = ("mnli", "mnli-mm") if args.task_name == "mnli" else (args.task_name,)
    eval_outputs_dirs = (args.output_dir, args.output_dir + "-MM") if args.task_name == "mnli" else (args.output_dir,)

    results = {}
    for eval_task, eval_output_dir in zip(eval_task_names, eval_outputs_dirs):
        eval_dataset = load_and_cache_examples(args, eval_task, tokenizer, set_type)

        if not os.path.exists(eval_output_dir) and args.local_rank in [-1, 0]:
            os.makedirs(eval_output_dir)

        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)
        eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=args.eval_batch_size)

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

        # Eval!
        logger.info("***** Running evaluation {} *****".format(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
        titles = None
        relation_heads = None
        relation_tails = None
        for batch in tqdm(eval_dataloader, desc="Evaluating"):
            model.eval()
            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", "albert"] else None
                    )  # XLM, DistilBERT, RoBERTa, and XLM-RoBERTa don't use segment_ids
                if args.task_name == "docred" or args.task_name == "tacred":
                    inputs["markers_mask"] = batch[4]
                outputs = model(**inputs)
                tmp_eval_loss, logits = outputs[:2]

                eval_loss += tmp_eval_loss.mean().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)
            if (args.task_name == "docred" or args.task_name == "tacred") and 'full' in set_type:
                if titles is None:
                    titles = [TITLE_NAMES(args.task_name, set_type, t) for t in batch[5].detach().cpu().numpy()]
                    relation_heads = batch[6].detach().cpu().numpy()
                    relation_tails = batch[7].detach().cpu().numpy()
                else:
                    titles += [TITLE_NAMES(args.task_name, set_type, t) for t in batch[5].detach().cpu().numpy()]
                    relation_heads = np.append(relation_heads, batch[6].detach().cpu().numpy(), axis=0)
                    relation_tails = np.append(relation_tails, batch[7].detach().cpu().numpy(), axis=0)

        eval_loss = eval_loss / nb_eval_steps
        if args.output_mode == "classification":
            preds = np.argmax(preds, axis=1)
            normalized_preds = torch.from_numpy(preds).softmax(1)[preds]
        elif args.output_mode == "regression":
            normalized_preds = 1/(1 + np.exp(-preds))
            preds = np.squeeze(preds) >= 0.5

        positive_label_index = 1
        result = compute_metrics(eval_task, preds, out_label_ids, positive_label_index)
        results.update(result)

        output_eval_file = os.path.join(eval_output_dir, prefix, "eval_results.txt")
        with open(output_eval_file, "w") as writer:
            logger.info("***** Eval results {} *****".format(prefix))
            for key in sorted(result.keys()):
                logger.info("  %s = %s", key, str(result[key]))
                writer.write("%s = %s\n" % (key, str(result[key])))

        if (args.task_name == "docred" or args.task_name == "tacred") and 'full' in set_type:
            full_eval_results = []
            for title, h_idx, t_idx, pred, confidence in zip(titles, relation_heads, relation_tails, preds, normalized_preds):
                if (args.output_mode == "regression" and pred) or (args.output_mode == "classification" and pred == positive_label_index):
                    relation_id = RELATION_MAPPING[args.task_name][args.relation_name]['id']
                    full_eval_results.append({'title': title, 'h_idx': int(h_idx), 't_idx': int(t_idx), 'r': relation_id, 'c': confidence.item()})
            output_eval_file = os.path.join(eval_output_dir, prefix, f"{set_type}_results.json")
            json.dump(full_eval_results, open(output_eval_file, "w"))

    return results

def load_and_cache_examples(args, task, tokenizer, set_type="train"):
    if args.local_rank not in [-1, 0] and set_type == "train":
        torch.distributed.barrier()  # Make sure only the first process in distributed training process the dataset, and the others will use the cache

    processor = processors[task](args.relation_name, args.num_positive_examples, args.ratio_negative_examples, args.type_independent_neg_sample)
    output_mode = output_modes[task]
    # Load data features from cache or dataset file
    cached_features_file = os.path.join(
        args.data_dir,
        "cached_{}_{}_{}_{}_{}_{}_{}".format(
            set_type,
            list(filter(None, args.model_name_or_path.split("/"))).pop(),
            str(args.max_seq_length),
            str(task),
            str(args.relation_name),
            str(args.num_positive_examples),
            str(args.ratio_negative_examples),
        ),
    )
    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)
        label_list = processor.get_labels()
        if task in ["mnli", "mnli-mm"] and args.model_type in ["roberta", "xlmroberta", "roberta-rc"]:
            # HACK(label indices are swapped in RoBERTa pretrained model)
            label_list[1], label_list[2] = label_list[2], label_list[1]
        examples = processor.get_examples_by_set_type(set_type, args.data_dir)
        features = convert_examples_to_features(
            examples,
            tokenizer,
            task=task,
            label_list=label_list,
            max_length=args.max_seq_length,
            output_mode=output_mode,
            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,
        )
        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 set_type == "train":
        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_attention_mask = torch.tensor([f.attention_mask for f in features], dtype=torch.long)
    all_token_type_ids = torch.tensor([f.token_type_ids for f in features], dtype=torch.long)
    if output_mode == "classification":
        all_labels = torch.tensor([f.label for f in features], dtype=torch.long)
    elif output_mode == "regression":
        all_labels = torch.tensor([f.label for f in features], dtype=torch.float)
    if args.task_name == "docred" or args.task_name == "tacred":
        all_markers_mask = torch.tensor([f.markers_mask for f in features], dtype=torch.bool)
        all_titles = torch.tensor([f.title for f in features], dtype=torch.long)
        all_h = torch.tensor([f.h for f in features], dtype=torch.long)
        all_t = torch.tensor([f.t for f in features], dtype=torch.long)
        return TensorDataset(all_input_ids, all_attention_mask, all_token_type_ids, all_labels, all_markers_mask, all_titles, all_h, all_t)
    
    return TensorDataset(all_input_ids, all_attention_mask, all_token_type_ids, all_labels)


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=None,
        type=str,
        required=True,
        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(
        "--task_name",
        default=None,
        type=str,
        required=True,
        help="The name of the task to train selected in the processors_factory function",
    )
    parser.add_argument(
        "--relation_name",
        default=None,
        type=str,
        required=True,
        help="The relation name on which you want to do binary classification",
    )
    parser.add_argument(
        "--num_positive_examples",
        default=None,
        type=int,
        required=True,
        help="The number of positive examples allowed for classification",
    )
    parser.add_argument(
        "--ratio_negative_examples",
        default=None,
        type=int,
        required=True,
        help="The ratio of negative examples allowed for classification comparing to positive examples",
    )
    parser.add_argument(
        "--output_dir",
        default=None,
        type=str,
        required=True,
        help="The output directory where the model predictions and checkpoints will be written.",
    )

    # Other parameters
    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_distant_train", action="store_true", help="Whether to run training on distant supervision data.")
    parser.add_argument("--do_search_train", action="store_true", help="Whether to run training on search supervision data.")
    parser.add_argument("--do_search_from_generation_train", action="store_true", help="Whether to run training on search_from_generation supervision data.")
    parser.add_argument("--do_generation_train", action="store_true", help="Whether to run training on generation supervision data.")
    parser.add_argument(
        "--training_method",
        default=None,
        type=str
    )
    parser.add_argument("--do_full_dev_eval", action="store_true", help="Whether to run eval over all possible relations on train eval split.")
    parser.add_argument("--do_full_test_eval", action="store_true", help="Whether to run eval over all possible relations on dev.")
    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(
        "--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("--patience", default=-1, type=int, help="Patience for Early Stopping.")
    parser.add_argument("--warmup_steps", default=100, type=int, help="Linear warmup over warmup_steps.")

    parser.add_argument("--logging_steps", type=int, default=500, 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("--save_only_best", action="store_true", help="No need to keep all checkpoints, just best one.")
    parser.add_argument("--save_best_wrt_metric", type=str, default="f1", help="What is the metric comparing checkpoints to")
    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(
        "--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.")

    parser.add_argument("--type_independent_neg_sample",
                        action="store_true",
                        help="Whether negative examples should have same entity types")
    args = parser.parse_args()

    if (
        os.path.exists(args.output_dir)
        and os.listdir(args.output_dir)
        and (args.training_method is not None)
        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 GLUE task
    args.task_name = args.task_name.lower()
    processor = processors[args.task_name](args.relation_name, args.num_positive_examples, args.ratio_negative_examples, args.type_independent_neg_sample)
    args.output_mode = output_modes[args.task_name]
    label_list = processor.get_labels()
    num_labels = len(label_list)

    # 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,
        finetuning_task=args.task_name,
        cache_dir=args.cache_dir if args.cache_dir else None,
    )
    tokenizer = tokenizer_class.from_pretrained(
        args.tokenizer_name if args.tokenizer_name else args.model_name_or_path,
        do_lower_case=args.do_lower_case,
        cache_dir=args.cache_dir if args.cache_dir else None,
    )
    model = model_class.from_pretrained(
        args.model_name_or_path,
        from_tf=bool(".ckpt" in args.model_name_or_path),
        config=config,
        cache_dir=args.cache_dir if args.cache_dir else None,
    )
    tokenizer.add_special_tokens({'additional_special_tokens': SPECIAL_TOKENS})
    model.resize_token_embeddings(len(tokenizer))

    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.training_method is not None:
        train_dataset = load_and_cache_examples(args, args.task_name, tokenizer, args.training_method)
        global_step, tr_loss = train(args, train_dataset, model, tokenizer)
        logger.info(" global_step = %s, average loss = %s", global_step, tr_loss)

    # Saving best-practices: if you use defaults names for the model, you can reload it using from_pretrained()
    if args.training_method is not None 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)

        best_dir = os.path.join(args.output_dir, "best")
        if os.path.exists(best_dir):
            copy_tree(best_dir, args.output_dir)
        else:
            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"))

        # Load a trained model and vocabulary that you have fine-tuned
        model = model_class.from_pretrained(args.output_dir)
        tokenizer = tokenizer_class.from_pretrained(args.output_dir)
        tokenizer.add_special_tokens({'additional_special_tokens': SPECIAL_TOKENS})
        model.resize_token_embeddings(len(tokenizer))
        model.to(args.device)

    # Evaluation
    results = {}
    splits = ['full_dev_eval', 'full_test_eval']
    bools = [args.do_full_dev_eval, args.do_full_test_eval]
    splits_to_eval = [s for s, b in zip(splits, bools) if b]
    if len(splits_to_eval) > 0 and args.local_rank in [-1, 0]:
        tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case)
        tokenizer.add_special_tokens({'additional_special_tokens': SPECIAL_TOKENS})
        model.resize_token_embeddings(len(tokenizer))
        checkpoints = [args.output_dir]
        if args.eval_all_checkpoints:
            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)
        for checkpoint in checkpoints:
            global_step = checkpoint.split("-")[-1] if len(checkpoints) > 1 else ""
            prefix = checkpoint.split("/")[-1] if checkpoint.find("checkpoint") != -1 else ""

            model = model_class.from_pretrained(checkpoint)
            model.to(args.device)
            for split in splits_to_eval:
                result = evaluate(args, model, tokenizer, prefix=prefix, set_type=split)
                result = dict((k + "_{}".format(global_step), v) for k, v in result.items())
                results.update(result)

    return results


if __name__ == "__main__":
    main()