Source code for recbole.trainer.trainer

# @Time   : 2020/6/26
# @Author : Shanlei Mu
# @Email  : slmu@ruc.edu.cn

# UPDATE:
# @Time   : 2022/7/8, 2021/6/23, 2020/9/26, 2020/9/26, 2020/10/01, 2020/9/16
# @Author : Zhen Tian, Zihan Lin, Yupeng Hou, Yushuo Chen, Shanlei Mu, Xingyu Pan
# @Email  : chenyuwuxinn@gmail.com, zhlin@ruc.edu.cn, houyupeng@ruc.edu.cn, chenyushuo@ruc.edu.cn, slmu@ruc.edu.cn, panxy@ruc.edu.cn

# UPDATE:
# @Time   : 2020/10/8, 2020/10/15, 2020/11/20, 2021/2/20, 2021/3/3, 2021/3/5, 2021/7/18, 2022/7/11, 2023/2/11
# @Author : Hui Wang, Xinyan Fan, Chen Yang, Yibo Li, Lanling Xu, Haoran Cheng, Zhichao Feng, Lei Wang, Gaowei Zhang
# @Email  : hui.wang@ruc.edu.cn, xinyan.fan@ruc.edu.cn, 254170321@qq.com, 2018202152@ruc.edu.cn, xulanling_sherry@163.com, chenghaoran29@foxmail.com, fzcbupt@gmail.com, zxcptss@gmail.com, zgw2022101006@ruc.edu.cn

r"""
recbole.trainer.trainer
################################
"""

import os

from logging import getLogger
from time import time

import numpy as np
import torch
import torch.optim as optim
from torch.nn.utils.clip_grad import clip_grad_norm_
from tqdm import tqdm
import torch.cuda.amp as amp

from recbole.data.interaction import Interaction
from recbole.data.dataloader import FullSortEvalDataLoader
from recbole.evaluator import Evaluator, Collector
from recbole.utils import (
    ensure_dir,
    get_local_time,
    early_stopping,
    calculate_valid_score,
    dict2str,
    EvaluatorType,
    KGDataLoaderState,
    get_tensorboard,
    set_color,
    get_gpu_usage,
    WandbLogger,
)
from torch.nn.parallel import DistributedDataParallel


class AbstractTrainer(object):
    r"""Trainer Class is used to manage the training and evaluation processes of recommender system models.
    AbstractTrainer is an abstract class in which the fit() and evaluate() method should be implemented according
    to different training and evaluation strategies.
    """

    def __init__(self, config, model):
        self.config = config
        self.model = model
        if not config["single_spec"]:
            self.model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model)
            self.distributed_model = DistributedDataParallel(
                self.model, device_ids=[config["local_rank"]]
            )

    def fit(self, train_data):
        r"""Train the model based on the train data."""
        raise NotImplementedError("Method [next] should be implemented.")

    def evaluate(self, eval_data):
        r"""Evaluate the model based on the eval data."""

        raise NotImplementedError("Method [next] should be implemented.")

    def set_reduce_hook(self):
        r"""Call the forward function of 'distributed_model' to apply grads
        reduce hook to each parameter of its module.

        """
        t = self.model.forward
        self.model.forward = lambda x: x
        self.distributed_model(torch.LongTensor([0]).to(self.device))
        self.model.forward = t

    def sync_grad_loss(self):
        r"""Ensure that each parameter appears to the loss function to
        make the grads reduce sync in each node.

        """
        sync_loss = 0
        for params in self.model.parameters():
            sync_loss += torch.sum(params) * 0
        return sync_loss


class Trainer(AbstractTrainer):
    r"""The basic Trainer for basic training and evaluation strategies in recommender systems. This class defines common
    functions for training and evaluation processes of most recommender system models, including fit(), evaluate(),
    resume_checkpoint() and some other features helpful for model training and evaluation.

    Generally speaking, this class can serve most recommender system models, If the training process of the model is to
    simply optimize a single loss without involving any complex training strategies, such as adversarial learning,
    pre-training and so on.

    Initializing the Trainer needs two parameters: `config` and `model`. `config` records the parameters information
    for controlling training and evaluation, such as `learning_rate`, `epochs`, `eval_step` and so on.
    `model` is the instantiated object of a Model Class.

    """

    def __init__(self, config, model):
        super(Trainer, self).__init__(config, model)

        self.logger = getLogger()
        self.tensorboard = get_tensorboard(self.logger)
        self.wandblogger = WandbLogger(config)
        self.learner = config["learner"]
        self.learning_rate = config["learning_rate"]
        self.epochs = config["epochs"]
        self.eval_step = min(config["eval_step"], self.epochs)
        self.stopping_step = config["stopping_step"]
        self.clip_grad_norm = config["clip_grad_norm"]
        self.valid_metric = config["valid_metric"].lower()
        self.valid_metric_bigger = config["valid_metric_bigger"]
        self.test_batch_size = config["eval_batch_size"]
        self.gpu_available = torch.cuda.is_available() and config["use_gpu"]
        self.device = config["device"]
        self.checkpoint_dir = config["checkpoint_dir"]
        self.enable_amp = config["enable_amp"]
        self.enable_scaler = torch.cuda.is_available() and config["enable_scaler"]
        ensure_dir(self.checkpoint_dir)
        saved_model_file = "{}-{}.pth".format(self.config["model"], get_local_time())
        self.saved_model_file = os.path.join(self.checkpoint_dir, saved_model_file)
        self.weight_decay = config["weight_decay"]

        self.start_epoch = 0
        self.cur_step = 0
        self.best_valid_score = -np.inf if self.valid_metric_bigger else np.inf
        self.best_valid_result = None
        self.train_loss_dict = dict()
        self.optimizer = self._build_optimizer()
        self.eval_type = config["eval_type"]
        self.eval_collector = Collector(config)
        self.evaluator = Evaluator(config)
        self.item_tensor = None
        self.tot_item_num = None

    def _build_optimizer(self, **kwargs):
        r"""Init the Optimizer

        Args:
            params (torch.nn.Parameter, optional): The parameters to be optimized.
                Defaults to ``self.model.parameters()``.
            learner (str, optional): The name of used optimizer. Defaults to ``self.learner``.
            learning_rate (float, optional): Learning rate. Defaults to ``self.learning_rate``.
            weight_decay (float, optional): The L2 regularization weight. Defaults to ``self.weight_decay``.

        Returns:
            torch.optim: the optimizer
        """
        params = kwargs.pop("params", self.model.parameters())
        learner = kwargs.pop("learner", self.learner)
        learning_rate = kwargs.pop("learning_rate", self.learning_rate)
        weight_decay = kwargs.pop("weight_decay", self.weight_decay)

        if (
            self.config["reg_weight"]
            and weight_decay
            and weight_decay * self.config["reg_weight"] > 0
        ):
            self.logger.warning(
                "The parameters [weight_decay] and [reg_weight] are specified simultaneously, "
                "which may lead to double regularization."
            )

        if learner.lower() == "adam":
            optimizer = optim.Adam(params, lr=learning_rate, weight_decay=weight_decay)
        elif learner.lower() == "adamw":
            optimizer = optim.AdamW(params, lr=learning_rate, weight_decay=weight_decay)
        elif learner.lower() == "sgd":
            optimizer = optim.SGD(params, lr=learning_rate, weight_decay=weight_decay)
        elif learner.lower() == "adagrad":
            optimizer = optim.Adagrad(
                params, lr=learning_rate, weight_decay=weight_decay
            )
        elif learner.lower() == "rmsprop":
            optimizer = optim.RMSprop(
                params, lr=learning_rate, weight_decay=weight_decay
            )
        elif learner.lower() == "sparse_adam":
            optimizer = optim.SparseAdam(params, lr=learning_rate)
            if weight_decay > 0:
                self.logger.warning(
                    "Sparse Adam cannot argument received argument [{weight_decay}]"
                )
        else:
            self.logger.warning(
                "Received unrecognized optimizer, set default Adam optimizer"
            )
            optimizer = optim.Adam(params, lr=learning_rate)
        return optimizer

    def _train_epoch(self, train_data, epoch_idx, loss_func=None, show_progress=False):
        r"""Train the model in an epoch

        Args:
            train_data (DataLoader): The train data.
            epoch_idx (int): The current epoch id.
            loss_func (function): The loss function of :attr:`model`. If it is ``None``, the loss function will be
                :attr:`self.model.calculate_loss`. Defaults to ``None``.
            show_progress (bool): Show the progress of training epoch. Defaults to ``False``.

        Returns:
            float/tuple: The sum of loss returned by all batches in this epoch. If the loss in each batch contains
            multiple parts and the model return these multiple parts loss instead of the sum of loss, it will return a
            tuple which includes the sum of loss in each part.
        """
        self.model.train()
        loss_func = loss_func or self.model.calculate_loss
        total_loss = None
        iter_data = (
            tqdm(
                train_data,
                total=len(train_data),
                ncols=100,
                desc=set_color(f"Train {epoch_idx:>5}", "pink"),
            )
            if show_progress
            else train_data
        )

        if not self.config["single_spec"] and train_data.shuffle:
            train_data.sampler.set_epoch(epoch_idx)

        scaler = amp.GradScaler(enabled=self.enable_scaler)
        for batch_idx, interaction in enumerate(iter_data):
            interaction = interaction.to(self.device)
            self.optimizer.zero_grad()
            sync_loss = 0
            if not self.config["single_spec"]:
                self.set_reduce_hook()
                sync_loss = self.sync_grad_loss()

            with torch.autocast(device_type=self.device.type, enabled=self.enable_amp):
                losses = loss_func(interaction)

            if isinstance(losses, tuple):
                loss = sum(losses)
                loss_tuple = tuple(per_loss.item() for per_loss in losses)
                total_loss = (
                    loss_tuple
                    if total_loss is None
                    else tuple(map(sum, zip(total_loss, loss_tuple)))
                )
            else:
                loss = losses
                total_loss = (
                    losses.item() if total_loss is None else total_loss + losses.item()
                )
            self._check_nan(loss)
            scaler.scale(loss + sync_loss).backward()
            if self.clip_grad_norm:
                clip_grad_norm_(self.model.parameters(), **self.clip_grad_norm)
            scaler.step(self.optimizer)
            scaler.update()
            if self.gpu_available and show_progress:
                iter_data.set_postfix_str(
                    set_color("GPU RAM: " + get_gpu_usage(self.device), "yellow")
                )
        return total_loss

    def _valid_epoch(self, valid_data, show_progress=False):
        r"""Valid the model with valid data

        Args:
            valid_data (DataLoader): the valid data.
            show_progress (bool): Show the progress of evaluate epoch. Defaults to ``False``.

        Returns:
            float: valid score
            dict: valid result
        """
        valid_result = self.evaluate(
            valid_data, load_best_model=False, show_progress=show_progress
        )
        valid_score = calculate_valid_score(valid_result, self.valid_metric)
        return valid_score, valid_result

    def _save_checkpoint(self, epoch, verbose=True, **kwargs):
        r"""Store the model parameters information and training information.

        Args:
            epoch (int): the current epoch id

        """
        if not self.config["single_spec"] and self.config["local_rank"] != 0:
            return
        saved_model_file = kwargs.pop("saved_model_file", self.saved_model_file)
        state = {
            "config": self.config,
            "epoch": epoch,
            "cur_step": self.cur_step,
            "best_valid_score": self.best_valid_score,
            "state_dict": self.model.state_dict(),
            "other_parameter": self.model.other_parameter(),
            "optimizer": self.optimizer.state_dict(),
        }
        torch.save(state, saved_model_file, pickle_protocol=4)
        if verbose:
            self.logger.info(
                set_color("Saving current", "blue") + f": {saved_model_file}"
            )

    def resume_checkpoint(self, resume_file):
        r"""Load the model parameters information and training information.

        Args:
            resume_file (file): the checkpoint file

        """
        resume_file = str(resume_file)
        self.saved_model_file = resume_file
        checkpoint = torch.load(resume_file, map_location=self.device)
        self.start_epoch = checkpoint["epoch"] + 1
        self.cur_step = checkpoint["cur_step"]
        self.best_valid_score = checkpoint["best_valid_score"]

        # load architecture params from checkpoint
        if checkpoint["config"]["model"].lower() != self.config["model"].lower():
            self.logger.warning(
                "Architecture configuration given in config file is different from that of checkpoint. "
                "This may yield an exception while state_dict is being loaded."
            )
        self.model.load_state_dict(checkpoint["state_dict"])
        self.model.load_other_parameter(checkpoint.get("other_parameter"))

        # load optimizer state from checkpoint only when optimizer type is not changed
        self.optimizer.load_state_dict(checkpoint["optimizer"])
        message_output = "Checkpoint loaded. Resume training from epoch {}".format(
            self.start_epoch
        )
        self.logger.info(message_output)

    def _check_nan(self, loss):
        if torch.isnan(loss):
            raise ValueError("Training loss is nan")

    def _generate_train_loss_output(self, epoch_idx, s_time, e_time, losses):
        des = self.config["loss_decimal_place"] or 4
        train_loss_output = (
            set_color("epoch %d training", "green")
            + " ["
            + set_color("time", "blue")
            + ": %.2fs, "
        ) % (epoch_idx, e_time - s_time)
        if isinstance(losses, tuple):
            des = set_color("train_loss%d", "blue") + ": %." + str(des) + "f"
            train_loss_output += ", ".join(
                des % (idx + 1, loss) for idx, loss in enumerate(losses)
            )
        else:
            des = "%." + str(des) + "f"
            train_loss_output += set_color("train loss", "blue") + ": " + des % losses
        return train_loss_output + "]"

    def _add_train_loss_to_tensorboard(self, epoch_idx, losses, tag="Loss/Train"):
        if isinstance(losses, tuple):
            for idx, loss in enumerate(losses):
                self.tensorboard.add_scalar(tag + str(idx), loss, epoch_idx)
        else:
            self.tensorboard.add_scalar(tag, losses, epoch_idx)

    def _add_hparam_to_tensorboard(self, best_valid_result):
        # base hparam
        hparam_dict = {
            "learner": self.config["learner"],
            "learning_rate": self.config["learning_rate"],
            "train_batch_size": self.config["train_batch_size"],
        }
        # unrecorded parameter
        unrecorded_parameter = {
            parameter
            for parameters in self.config.parameters.values()
            for parameter in parameters
        }.union({"model", "dataset", "config_files", "device"})
        # other model-specific hparam
        hparam_dict.update(
            {
                para: val
                for para, val in self.config.final_config_dict.items()
                if para not in unrecorded_parameter
            }
        )
        for k in hparam_dict:
            if hparam_dict[k] is not None and not isinstance(
                hparam_dict[k], (bool, str, float, int)
            ):
                hparam_dict[k] = str(hparam_dict[k])

        self.tensorboard.add_hparams(
            hparam_dict, {"hparam/best_valid_result": best_valid_result}
        )

    def fit(
        self,
        train_data,
        valid_data=None,
        verbose=True,
        saved=True,
        show_progress=False,
        callback_fn=None,
    ):
        r"""Train the model based on the train data and the valid data.

        Args:
            train_data (DataLoader): the train data
            valid_data (DataLoader, optional): the valid data, default: None.
                                               If it's None, the early_stopping is invalid.
            verbose (bool, optional): whether to write training and evaluation information to logger, default: True
            saved (bool, optional): whether to save the model parameters, default: True
            show_progress (bool): Show the progress of training epoch and evaluate epoch. Defaults to ``False``.
            callback_fn (callable): Optional callback function executed at end of epoch.
                                    Includes (epoch_idx, valid_score) input arguments.

        Returns:
             (float, dict): best valid score and best valid result. If valid_data is None, it returns (-1, None)
        """
        if saved and self.start_epoch >= self.epochs:
            self._save_checkpoint(-1, verbose=verbose)

        self.eval_collector.data_collect(train_data)
        if self.config["train_neg_sample_args"].get("dynamic", False):
            train_data.get_model(self.model)
        valid_step = 0

        for epoch_idx in range(self.start_epoch, self.epochs):
            # train
            training_start_time = time()
            train_loss = self._train_epoch(
                train_data, epoch_idx, show_progress=show_progress
            )
            self.train_loss_dict[epoch_idx] = (
                sum(train_loss) if isinstance(train_loss, tuple) else train_loss
            )
            training_end_time = time()
            train_loss_output = self._generate_train_loss_output(
                epoch_idx, training_start_time, training_end_time, train_loss
            )
            if verbose:
                self.logger.info(train_loss_output)
            self._add_train_loss_to_tensorboard(epoch_idx, train_loss)
            self.wandblogger.log_metrics(
                {"epoch": epoch_idx, "train_loss": train_loss, "train_step": epoch_idx},
                head="train",
            )

            # eval
            if self.eval_step <= 0 or not valid_data:
                if saved:
                    self._save_checkpoint(epoch_idx, verbose=verbose)
                continue
            if (epoch_idx + 1) % self.eval_step == 0:
                valid_start_time = time()
                valid_score, valid_result = self._valid_epoch(
                    valid_data, show_progress=show_progress
                )

                (
                    self.best_valid_score,
                    self.cur_step,
                    stop_flag,
                    update_flag,
                ) = early_stopping(
                    valid_score,
                    self.best_valid_score,
                    self.cur_step,
                    max_step=self.stopping_step,
                    bigger=self.valid_metric_bigger,
                )
                valid_end_time = time()
                valid_score_output = (
                    set_color("epoch %d evaluating", "green")
                    + " ["
                    + set_color("time", "blue")
                    + ": %.2fs, "
                    + set_color("valid_score", "blue")
                    + ": %f]"
                ) % (epoch_idx, valid_end_time - valid_start_time, valid_score)
                valid_result_output = (
                    set_color("valid result", "blue") + ": \n" + dict2str(valid_result)
                )
                if verbose:
                    self.logger.info(valid_score_output)
                    self.logger.info(valid_result_output)
                self.tensorboard.add_scalar("Vaild_score", valid_score, epoch_idx)
                self.wandblogger.log_metrics(
                    {**valid_result, "valid_step": valid_step}, head="valid"
                )

                if update_flag:
                    if saved:
                        self._save_checkpoint(epoch_idx, verbose=verbose)
                    self.best_valid_result = valid_result

                if callback_fn:
                    callback_fn(epoch_idx, valid_score)

                if stop_flag:
                    stop_output = "Finished training, best eval result in epoch %d" % (
                        epoch_idx - self.cur_step * self.eval_step
                    )
                    if verbose:
                        self.logger.info(stop_output)
                    break

                valid_step += 1

        self._add_hparam_to_tensorboard(self.best_valid_score)
        return self.best_valid_score, self.best_valid_result

    def _full_sort_batch_eval(self, batched_data):
        interaction, history_index, positive_u, positive_i = batched_data
        try:
            # Note: interaction without item ids
            scores = self.model.full_sort_predict(interaction.to(self.device))
        except NotImplementedError:
            inter_len = len(interaction)
            new_inter = interaction.to(self.device).repeat_interleave(self.tot_item_num)
            batch_size = len(new_inter)
            new_inter.update(self.item_tensor.repeat(inter_len))
            if batch_size <= self.test_batch_size:
                scores = self.model.predict(new_inter)
            else:
                scores = self._spilt_predict(new_inter, batch_size)

        scores = scores.view(-1, self.tot_item_num)
        scores[:, 0] = -np.inf
        if history_index is not None:
            scores[history_index] = -np.inf
        return interaction, scores, positive_u, positive_i

    def _neg_sample_batch_eval(self, batched_data):
        interaction, row_idx, positive_u, positive_i = batched_data
        batch_size = interaction.length
        if batch_size <= self.test_batch_size:
            origin_scores = self.model.predict(interaction.to(self.device))
        else:
            origin_scores = self._spilt_predict(interaction, batch_size)

        if self.config["eval_type"] == EvaluatorType.VALUE:
            return interaction, origin_scores, positive_u, positive_i
        elif self.config["eval_type"] == EvaluatorType.RANKING:
            col_idx = interaction[self.config["ITEM_ID_FIELD"]]
            batch_user_num = positive_u[-1] + 1
            scores = torch.full(
                (batch_user_num, self.tot_item_num), -np.inf, device=self.device
            )
            scores[row_idx, col_idx] = origin_scores
            return interaction, scores, positive_u, positive_i

    @torch.no_grad()
    def evaluate(
        self, eval_data, load_best_model=True, model_file=None, show_progress=False
    ):
        r"""Evaluate the model based on the eval data.

        Args:
            eval_data (DataLoader): the eval data
            load_best_model (bool, optional): whether load the best model in the training process, default: True.
                                              It should be set True, if users want to test the model after training.
            model_file (str, optional): the saved model file, default: None. If users want to test the previously
                                        trained model file, they can set this parameter.
            show_progress (bool): Show the progress of evaluate epoch. Defaults to ``False``.

        Returns:
            collections.OrderedDict: eval result, key is the eval metric and value in the corresponding metric value.
        """
        if not eval_data:
            return

        if load_best_model:
            checkpoint_file = model_file or self.saved_model_file
            checkpoint = torch.load(checkpoint_file, map_location=self.device)
            self.model.load_state_dict(checkpoint["state_dict"])
            self.model.load_other_parameter(checkpoint.get("other_parameter"))
            message_output = "Loading model structure and parameters from {}".format(
                checkpoint_file
            )
            self.logger.info(message_output)

        self.model.eval()

        if isinstance(eval_data, FullSortEvalDataLoader):
            eval_func = self._full_sort_batch_eval
            if self.item_tensor is None:
                self.item_tensor = eval_data._dataset.get_item_feature().to(self.device)
        else:
            eval_func = self._neg_sample_batch_eval
        if self.config["eval_type"] == EvaluatorType.RANKING:
            self.tot_item_num = eval_data._dataset.item_num

        iter_data = (
            tqdm(
                eval_data,
                total=len(eval_data),
                ncols=100,
                desc=set_color(f"Evaluate   ", "pink"),
            )
            if show_progress
            else eval_data
        )

        num_sample = 0
        for batch_idx, batched_data in enumerate(iter_data):
            num_sample += len(batched_data)
            interaction, scores, positive_u, positive_i = eval_func(batched_data)
            if self.gpu_available and show_progress:
                iter_data.set_postfix_str(
                    set_color("GPU RAM: " + get_gpu_usage(self.device), "yellow")
                )
            self.eval_collector.eval_batch_collect(
                scores, interaction, positive_u, positive_i
            )
        self.eval_collector.model_collect(self.model)
        struct = self.eval_collector.get_data_struct()
        result = self.evaluator.evaluate(struct)
        if not self.config["single_spec"]:
            result = self._map_reduce(result, num_sample)
        self.wandblogger.log_eval_metrics(result, head="eval")
        return result

    def _map_reduce(self, result, num_sample):
        gather_result = {}
        total_sample = [
            torch.zeros(1).to(self.device) for _ in range(self.config["world_size"])
        ]
        torch.distributed.all_gather(
            total_sample, torch.Tensor([num_sample]).to(self.device)
        )
        total_sample = torch.cat(total_sample, 0)
        total_sample = torch.sum(total_sample).item()
        for key, value in result.items():
            result[key] = torch.Tensor([value * num_sample]).to(self.device)
            gather_result[key] = [
                torch.zeros_like(result[key]).to(self.device)
                for _ in range(self.config["world_size"])
            ]
            torch.distributed.all_gather(gather_result[key], result[key])
            gather_result[key] = torch.cat(gather_result[key], dim=0)
            gather_result[key] = round(
                torch.sum(gather_result[key]).item() / total_sample,
                self.config["metric_decimal_place"],
            )
        return gather_result

    def _spilt_predict(self, interaction, batch_size):
        spilt_interaction = dict()
        for key, tensor in interaction.interaction.items():
            spilt_interaction[key] = tensor.split(self.test_batch_size, dim=0)
        num_block = (batch_size + self.test_batch_size - 1) // self.test_batch_size
        result_list = []
        for i in range(num_block):
            current_interaction = dict()
            for key, spilt_tensor in spilt_interaction.items():
                current_interaction[key] = spilt_tensor[i]
            result = self.model.predict(
                Interaction(current_interaction).to(self.device)
            )
            if len(result.shape) == 0:
                result = result.unsqueeze(0)
            result_list.append(result)
        return torch.cat(result_list, dim=0)


class KGTrainer(Trainer):
    r"""KGTrainer is designed for Knowledge-aware recommendation methods. Some of these models need to train the
    recommendation related task and knowledge related task alternately.

    """

    def __init__(self, config, model):
        super(KGTrainer, self).__init__(config, model)

        self.train_rec_step = config["train_rec_step"]
        self.train_kg_step = config["train_kg_step"]

    def _train_epoch(self, train_data, epoch_idx, loss_func=None, show_progress=False):
        if self.train_rec_step is None or self.train_kg_step is None:
            interaction_state = KGDataLoaderState.RSKG
        elif (
            epoch_idx % (self.train_rec_step + self.train_kg_step) < self.train_rec_step
        ):
            interaction_state = KGDataLoaderState.RS
        else:
            interaction_state = KGDataLoaderState.KG
        if not self.config["single_spec"]:
            train_data.knowledge_shuffle(epoch_idx)
        train_data.set_mode(interaction_state)
        if interaction_state in [KGDataLoaderState.RSKG, KGDataLoaderState.RS]:
            return super()._train_epoch(
                train_data, epoch_idx, show_progress=show_progress
            )
        elif interaction_state in [KGDataLoaderState.KG]:
            return super()._train_epoch(
                train_data,
                epoch_idx,
                loss_func=self.model.calculate_kg_loss,
                show_progress=show_progress,
            )
        return None


class KGATTrainer(Trainer):
    r"""KGATTrainer is designed for KGAT, which is a knowledge-aware recommendation method."""

    def __init__(self, config, model):
        super(KGATTrainer, self).__init__(config, model)

    def _train_epoch(self, train_data, epoch_idx, loss_func=None, show_progress=False):
        # train rs
        if not self.config["single_spec"]:
            train_data.knowledge_shuffle(epoch_idx)
        train_data.set_mode(KGDataLoaderState.RS)
        rs_total_loss = super()._train_epoch(
            train_data, epoch_idx, show_progress=show_progress
        )

        # train kg
        train_data.set_mode(KGDataLoaderState.KG)
        kg_total_loss = super()._train_epoch(
            train_data,
            epoch_idx,
            loss_func=self.model.calculate_kg_loss,
            show_progress=show_progress,
        )

        # update A
        self.model.eval()
        with torch.no_grad():
            self.model.update_attentive_A()

        return rs_total_loss, kg_total_loss


class PretrainTrainer(Trainer):
    r"""PretrainTrainer is designed for pre-training.
    It can be inherited by the trainer which needs pre-training and fine-tuning.
    """

    def __init__(self, config, model):
        super(PretrainTrainer, self).__init__(config, model)
        self.pretrain_epochs = self.config["pretrain_epochs"]
        self.save_step = self.config["save_step"]

    def save_pretrained_model(self, epoch, saved_model_file):
        r"""Store the model parameters information and training information.

        Args:
            epoch (int): the current epoch id
            saved_model_file (str): file name for saved pretrained model

        """
        state = {
            "config": self.config,
            "epoch": epoch,
            "state_dict": self.model.state_dict(),
            "optimizer": self.optimizer.state_dict(),
            "other_parameter": self.model.other_parameter(),
        }
        torch.save(state, saved_model_file)
        self.saved_model_file = saved_model_file

    def pretrain(self, train_data, verbose=True, show_progress=False):
        for epoch_idx in range(self.start_epoch, self.pretrain_epochs):
            # train
            training_start_time = time()
            train_loss = self._train_epoch(
                train_data, epoch_idx, show_progress=show_progress
            )
            self.train_loss_dict[epoch_idx] = (
                sum(train_loss) if isinstance(train_loss, tuple) else train_loss
            )
            training_end_time = time()
            train_loss_output = self._generate_train_loss_output(
                epoch_idx, training_start_time, training_end_time, train_loss
            )
            if verbose:
                self.logger.info(train_loss_output)
            self._add_train_loss_to_tensorboard(epoch_idx, train_loss)

            if (epoch_idx + 1) % self.save_step == 0:
                saved_model_file = os.path.join(
                    self.checkpoint_dir,
                    "{}-{}-{}.pth".format(
                        self.config["model"], self.config["dataset"], str(epoch_idx + 1)
                    ),
                )
                self.save_pretrained_model(epoch_idx, saved_model_file)
                update_output = (
                    set_color("Saving current", "blue") + ": %s" % saved_model_file
                )
                if verbose:
                    self.logger.info(update_output)

        return self.best_valid_score, self.best_valid_result


class S3RecTrainer(PretrainTrainer):
    r"""S3RecTrainer is designed for S3Rec, which is a self-supervised learning based sequential recommenders.
    It includes two training stages: pre-training ang fine-tuning.

    """

    def __init__(self, config, model):
        super(S3RecTrainer, self).__init__(config, model)

    def fit(
        self,
        train_data,
        valid_data=None,
        verbose=True,
        saved=True,
        show_progress=False,
        callback_fn=None,
    ):
        if self.model.train_stage == "pretrain":
            return self.pretrain(train_data, verbose, show_progress)
        elif self.model.train_stage == "finetune":
            return super().fit(
                train_data, valid_data, verbose, saved, show_progress, callback_fn
            )
        else:
            raise ValueError(
                "Please make sure that the 'train_stage' is 'pretrain' or 'finetune'!"
            )


class MKRTrainer(Trainer):
    r"""MKRTrainer is designed for MKR, which is a knowledge-aware recommendation method."""

    def __init__(self, config, model):
        super(MKRTrainer, self).__init__(config, model)
        self.kge_interval = config["kge_interval"]

    def _train_epoch(self, train_data, epoch_idx, loss_func=None, show_progress=False):
        rs_total_loss, kg_total_loss = 0.0, 0.0

        # train rs
        self.logger.info("Train RS")
        train_data.set_mode(KGDataLoaderState.RS)
        rs_total_loss = super()._train_epoch(
            train_data,
            epoch_idx,
            loss_func=self.model.calculate_rs_loss,
            show_progress=show_progress,
        )

        # train kg
        if epoch_idx % self.kge_interval == 0:
            self.logger.info("Train KG")
            train_data.set_mode(KGDataLoaderState.KG)
            kg_total_loss = super()._train_epoch(
                train_data,
                epoch_idx,
                loss_func=self.model.calculate_kg_loss,
                show_progress=show_progress,
            )

        return rs_total_loss, kg_total_loss


class TraditionalTrainer(Trainer):
    r"""TraditionalTrainer is designed for Traditional model(Pop,ItemKNN), which set the epoch to 1 whatever the config."""

    def __init__(self, config, model):
        super(TraditionalTrainer, self).__init__(config, model)
        self.epochs = 1  # Set the epoch to 1 when running memory based model


class DecisionTreeTrainer(AbstractTrainer):
    """DecisionTreeTrainer is designed for DecisionTree model."""

    def __init__(self, config, model):
        super(DecisionTreeTrainer, self).__init__(config, model)

        self.logger = getLogger()
        self.tensorboard = get_tensorboard(self.logger)
        self.label_field = config["LABEL_FIELD"]
        self.convert_token_to_onehot = self.config["convert_token_to_onehot"]

        # evaluator
        self.eval_type = config["eval_type"]
        self.epochs = config["epochs"]
        self.eval_step = min(config["eval_step"], self.epochs)
        self.valid_metric = config["valid_metric"].lower()
        self.eval_collector = Collector(config)
        self.evaluator = Evaluator(config)

        # model saved
        self.checkpoint_dir = config["checkpoint_dir"]
        ensure_dir(self.checkpoint_dir)
        temp_file = "{}-{}-temp.pth".format(self.config["model"], get_local_time())
        self.temp_file = os.path.join(self.checkpoint_dir, temp_file)

        temp_best_file = "{}-{}-temp-best.pth".format(
            self.config["model"], get_local_time()
        )
        self.temp_best_file = os.path.join(self.checkpoint_dir, temp_best_file)

        saved_model_file = "{}-{}.pth".format(self.config["model"], get_local_time())
        self.saved_model_file = os.path.join(self.checkpoint_dir, saved_model_file)

        self.stopping_step = config["stopping_step"]
        self.valid_metric_bigger = config["valid_metric_bigger"]
        self.cur_step = 0
        self.best_valid_score = -np.inf if self.valid_metric_bigger else np.inf
        self.best_valid_result = None

    def _interaction_to_sparse(self, dataloader):
        r"""Convert data format from interaction to sparse or numpy

        Args:
            dataloader (DecisionTreeDataLoader): DecisionTreeDataLoader dataloader.
        Returns:
            cur_data (sparse or numpy): data.
            interaction_np[self.label_field] (numpy): label.
        """
        interaction = dataloader._dataset[:]
        interaction_np = interaction.numpy()
        cur_data = np.array([])
        columns = []
        for key, value in interaction_np.items():
            value = np.resize(value, (value.shape[0], 1))
            if key != self.label_field:
                columns.append(key)
                if cur_data.shape[0] == 0:
                    cur_data = value
                else:
                    cur_data = np.hstack((cur_data, value))

        if self.convert_token_to_onehot:
            from scipy import sparse
            from scipy.sparse import dok_matrix

            convert_col_list = dataloader._dataset.convert_col_list
            hash_count = dataloader._dataset.hash_count

            new_col = cur_data.shape[1] - len(convert_col_list)
            for key, values in hash_count.items():
                new_col = new_col + values
            onehot_data = dok_matrix((cur_data.shape[0], new_col))

            cur_j = 0
            new_j = 0

            for key in columns:
                if key in convert_col_list:
                    for i in range(cur_data.shape[0]):
                        onehot_data[i, int(new_j + cur_data[i, cur_j])] = 1
                    new_j = new_j + hash_count[key] - 1
                else:
                    for i in range(cur_data.shape[0]):
                        onehot_data[i, new_j] = cur_data[i, cur_j]
                cur_j = cur_j + 1
                new_j = new_j + 1

            cur_data = sparse.csc_matrix(onehot_data)

        return cur_data, interaction_np[self.label_field]

    def _interaction_to_lib_datatype(self, dataloader):
        pass

    def _valid_epoch(self, valid_data):
        r"""

        Args:
            valid_data (DecisionTreeDataLoader): DecisionTreeDataLoader, which is the same with GeneralDataLoader.
        """
        valid_result = self.evaluate(valid_data, load_best_model=False)
        valid_score = calculate_valid_score(valid_result, self.valid_metric)
        return valid_score, valid_result

    def _save_checkpoint(self, epoch):
        r"""Store the model parameters information and training information.

        Args:
            epoch (int): the current epoch id

        """
        state = {
            "config": self.config,
            "epoch": epoch,
            "cur_step": self.cur_step,
            "best_valid_score": self.best_valid_score,
            "state_dict": self.temp_best_file,
            "other_parameter": None,
        }
        torch.save(state, self.saved_model_file)

    def fit(
        self, train_data, valid_data=None, verbose=True, saved=True, show_progress=False
    ):
        for epoch_idx in range(self.epochs):
            self._train_at_once(train_data, valid_data)

            if (epoch_idx + 1) % self.eval_step == 0:
                # evaluate
                valid_start_time = time()
                valid_score, valid_result = self._valid_epoch(valid_data)

                (
                    self.best_valid_score,
                    self.cur_step,
                    stop_flag,
                    update_flag,
                ) = early_stopping(
                    valid_score,
                    self.best_valid_score,
                    self.cur_step,
                    max_step=self.stopping_step,
                    bigger=self.valid_metric_bigger,
                )

                valid_end_time = time()
                valid_score_output = (
                    set_color("epoch %d evaluating", "green")
                    + " ["
                    + set_color("time", "blue")
                    + ": %.2fs, "
                    + set_color("valid_score", "blue")
                    + ": %f]"
                ) % (epoch_idx, valid_end_time - valid_start_time, valid_score)
                valid_result_output = (
                    set_color("valid result", "blue") + ": \n" + dict2str(valid_result)
                )
                if verbose:
                    self.logger.info(valid_score_output)
                    self.logger.info(valid_result_output)
                self.tensorboard.add_scalar("Vaild_score", valid_score, epoch_idx)

                if update_flag:
                    if saved:
                        self.model.save_model(self.temp_best_file)
                        self._save_checkpoint(epoch_idx)
                    self.best_valid_result = valid_result

                if stop_flag:
                    stop_output = "Finished training, best eval result in epoch %d" % (
                        epoch_idx - self.cur_step * self.eval_step
                    )
                    if self.temp_file:
                        os.remove(self.temp_file)
                    if verbose:
                        self.logger.info(stop_output)
                    break

        return self.best_valid_score, self.best_valid_result

    def evaluate(
        self, eval_data, load_best_model=True, model_file=None, show_progress=False
    ):
        raise NotImplementedError

    def _train_at_once(self, train_data, valid_data):
        raise NotImplementedError


class XGBoostTrainer(DecisionTreeTrainer):
    """XGBoostTrainer is designed for XGBOOST."""

    def __init__(self, config, model):
        super(XGBoostTrainer, self).__init__(config, model)

        self.xgb = __import__("xgboost")
        self.boost_model = config["xgb_model"]
        self.silent = config["xgb_silent"]
        self.nthread = config["xgb_nthread"]

        # train params
        self.params = config["xgb_params"]
        self.num_boost_round = config["xgb_num_boost_round"]
        self.evals = ()
        self.early_stopping_rounds = config["xgb_early_stopping_rounds"]
        self.evals_result = {}
        self.verbose_eval = config["xgb_verbose_eval"]
        self.callbacks = None
        self.deval = None
        self.eval_pred = self.eval_true = None

    def _interaction_to_lib_datatype(self, dataloader):
        r"""Convert data format from interaction to DMatrix

        Args:
            dataloader (DecisionTreeDataLoader): xgboost dataloader.
        Returns:
            DMatrix: Data in the form of 'DMatrix'.
        """
        data, label = self._interaction_to_sparse(dataloader)
        return self.xgb.DMatrix(
            data=data, label=label, silent=self.silent, nthread=self.nthread
        )

    def _train_at_once(self, train_data, valid_data):
        r"""

        Args:
            train_data (DecisionTreeDataLoader): DecisionTreeDataLoader, which is the same with GeneralDataLoader.
            valid_data (DecisionTreeDataLoader): DecisionTreeDataLoader, which is the same with GeneralDataLoader.
        """
        self.dtrain = self._interaction_to_lib_datatype(train_data)
        self.dvalid = self._interaction_to_lib_datatype(valid_data)
        self.evals = [(self.dtrain, "train"), (self.dvalid, "valid")]
        self.model = self.xgb.train(
            self.params,
            self.dtrain,
            self.num_boost_round,
            self.evals,
            early_stopping_rounds=self.early_stopping_rounds,
            evals_result=self.evals_result,
            verbose_eval=self.verbose_eval,
            xgb_model=self.boost_model,
            callbacks=self.callbacks,
        )

        self.model.save_model(self.temp_file)
        self.boost_model = self.temp_file

    def evaluate(
        self, eval_data, load_best_model=True, model_file=None, show_progress=False
    ):
        if load_best_model:
            if model_file:
                checkpoint_file = model_file
            else:
                checkpoint_file = self.temp_best_file
            self.model.load_model(checkpoint_file)

        self.deval = self._interaction_to_lib_datatype(eval_data)
        self.eval_true = torch.Tensor(self.deval.get_label())
        self.eval_pred = torch.Tensor(self.model.predict(self.deval))

        self.eval_collector.eval_collect(self.eval_pred, self.eval_true)
        result = self.evaluator.evaluate(self.eval_collector.get_data_struct())
        return result


class LightGBMTrainer(DecisionTreeTrainer):
    """LightGBMTrainer is designed for LightGBM."""

    def __init__(self, config, model):
        super(LightGBMTrainer, self).__init__(config, model)

        self.lgb = __import__("lightgbm")

        # train params
        self.params = config["lgb_params"]
        self.num_boost_round = config["lgb_num_boost_round"]
        self.evals = ()
        self.deval_data = self.deval_label = None
        self.eval_pred = self.eval_true = None

    def _interaction_to_lib_datatype(self, dataloader):
        r"""Convert data format from interaction to Dataset

        Args:
            dataloader (DecisionTreeDataLoader): xgboost dataloader.
        Returns:
            dataset(lgb.Dataset): Data in the form of 'lgb.Dataset'.
        """
        data, label = self._interaction_to_sparse(dataloader)
        return self.lgb.Dataset(data=data, label=label)

    def _train_at_once(self, train_data, valid_data):
        r"""

        Args:
            train_data (DecisionTreeDataLoader): DecisionTreeDataLoader, which is the same with GeneralDataLoader.
            valid_data (DecisionTreeDataLoader): DecisionTreeDataLoader, which is the same with GeneralDataLoader.
        """
        self.dtrain = self._interaction_to_lib_datatype(train_data)
        self.dvalid = self._interaction_to_lib_datatype(valid_data)
        self.evals = [self.dtrain, self.dvalid]
        self.model = self.lgb.train(
            self.params, self.dtrain, self.num_boost_round, self.evals
        )

        self.model.save_model(self.temp_file)
        self.boost_model = self.temp_file

    def evaluate(
        self, eval_data, load_best_model=True, model_file=None, show_progress=False
    ):
        if load_best_model:
            if model_file:
                checkpoint_file = model_file
            else:
                checkpoint_file = self.temp_best_file
            self.model = self.lgb.Booster(model_file=checkpoint_file)

        self.deval_data, self.deval_label = self._interaction_to_sparse(eval_data)
        self.eval_true = torch.Tensor(self.deval_label)
        self.eval_pred = torch.Tensor(self.model.predict(self.deval_data))

        self.eval_collector.eval_collect(self.eval_pred, self.eval_true)
        result = self.evaluator.evaluate(self.eval_collector.get_data_struct())
        return result


class RaCTTrainer(PretrainTrainer):
    r"""RaCTTrainer is designed for RaCT, which is an actor-critic reinforcement learning based general recommenders.
    It includes three training stages: actor pre-training, critic pre-training and actor-critic training.

    """

    def __init__(self, config, model):
        super(RaCTTrainer, self).__init__(config, model)

    def fit(
        self,
        train_data,
        valid_data=None,
        verbose=True,
        saved=True,
        show_progress=False,
        callback_fn=None,
    ):
        if self.model.train_stage == "actor_pretrain":
            return self.pretrain(train_data, verbose, show_progress)
        elif self.model.train_stage == "critic_pretrain":
            return self.pretrain(train_data, verbose, show_progress)
        elif self.model.train_stage == "finetune":
            return super().fit(
                train_data, valid_data, verbose, saved, show_progress, callback_fn
            )
        else:
            raise ValueError(
                "Please make sure that the 'train_stage' is "
                "'actor_pretrain', 'critic_pretrain' or 'finetune'!"
            )


class RecVAETrainer(Trainer):
    r"""RecVAETrainer is designed for RecVAE, which is a general recommender."""

    def __init__(self, config, model):
        super(RecVAETrainer, self).__init__(config, model)
        self.n_enc_epochs = config["n_enc_epochs"]
        self.n_dec_epochs = config["n_dec_epochs"]

        self.optimizer_encoder = self._build_optimizer(
            params=self.model.encoder.parameters()
        )
        self.optimizer_decoder = self._build_optimizer(
            params=self.model.decoder.parameters()
        )

    def _train_epoch(self, train_data, epoch_idx, loss_func=None, show_progress=False):
        self.optimizer = self.optimizer_encoder
        encoder_loss_func = lambda data: self.model.calculate_loss(
            data, encoder_flag=True
        )
        for epoch in range(self.n_enc_epochs):
            super()._train_epoch(
                train_data,
                epoch_idx,
                loss_func=encoder_loss_func,
                show_progress=show_progress,
            )

        self.model.update_prior()
        loss = 0.0
        self.optimizer = self.optimizer_decoder
        decoder_loss_func = lambda data: self.model.calculate_loss(
            data, encoder_flag=False
        )
        for epoch in range(self.n_dec_epochs):
            loss += super()._train_epoch(
                train_data,
                epoch_idx,
                loss_func=decoder_loss_func,
                show_progress=show_progress,
            )
        return loss


class NCLTrainer(Trainer):
    def __init__(self, config, model):
        super(NCLTrainer, self).__init__(config, model)

        self.num_m_step = config["m_step"]
        assert self.num_m_step is not None

    def fit(
        self,
        train_data,
        valid_data=None,
        verbose=True,
        saved=True,
        show_progress=False,
        callback_fn=None,
    ):
        r"""Train the model based on the train data and the valid data.

        Args:
            train_data (DataLoader): the train data.
            valid_data (DataLoader, optional): the valid data, default: None.
                                               If it's None, the early_stopping is invalid.
            verbose (bool, optional): whether to write training and evaluation information to logger, default: True
            saved (bool, optional): whether to save the model parameters, default: True
            show_progress (bool): Show the progress of training epoch and evaluate epoch. Defaults to ``False``.
            callback_fn (callable): Optional callback function executed at end of epoch.
                                    Includes (epoch_idx, valid_score) input arguments.

        Returns:
             (float, dict): best valid score and best valid result. If valid_data is None, it returns (-1, None)
        """
        if saved and self.start_epoch >= self.epochs:
            self._save_checkpoint(-1)

        self.eval_collector.data_collect(train_data)

        for epoch_idx in range(self.start_epoch, self.epochs):
            # only differences from the original trainer
            if epoch_idx % self.num_m_step == 0:
                self.logger.info("Running E-step ! ")
                self.model.e_step()
            # train
            training_start_time = time()
            train_loss = self._train_epoch(
                train_data, epoch_idx, show_progress=show_progress
            )
            self.train_loss_dict[epoch_idx] = (
                sum(train_loss) if isinstance(train_loss, tuple) else train_loss
            )
            training_end_time = time()
            train_loss_output = self._generate_train_loss_output(
                epoch_idx, training_start_time, training_end_time, train_loss
            )
            if verbose:
                self.logger.info(train_loss_output)
            self._add_train_loss_to_tensorboard(epoch_idx, train_loss)

            # eval
            if self.eval_step <= 0 or not valid_data:
                if saved:
                    self._save_checkpoint(epoch_idx)
                    update_output = (
                        set_color("Saving current", "blue")
                        + ": %s" % self.saved_model_file
                    )
                    if verbose:
                        self.logger.info(update_output)
                continue
            if (epoch_idx + 1) % self.eval_step == 0:
                valid_start_time = time()
                valid_score, valid_result = self._valid_epoch(
                    valid_data, show_progress=show_progress
                )

                (
                    self.best_valid_score,
                    self.cur_step,
                    stop_flag,
                    update_flag,
                ) = early_stopping(
                    valid_score,
                    self.best_valid_score,
                    self.cur_step,
                    max_step=self.stopping_step,
                    bigger=self.valid_metric_bigger,
                )
                valid_end_time = time()
                valid_score_output = (
                    set_color("epoch %d evaluating", "green")
                    + " ["
                    + set_color("time", "blue")
                    + ": %.2fs, "
                    + set_color("valid_score", "blue")
                    + ": %f]"
                ) % (epoch_idx, valid_end_time - valid_start_time, valid_score)
                valid_result_output = (
                    set_color("valid result", "blue") + ": \n" + dict2str(valid_result)
                )
                if verbose:
                    self.logger.info(valid_score_output)
                    self.logger.info(valid_result_output)
                self.tensorboard.add_scalar("Vaild_score", valid_score, epoch_idx)

                if update_flag:
                    if saved:
                        self._save_checkpoint(epoch_idx)
                        update_output = (
                            set_color("Saving current best", "blue")
                            + ": %s" % self.saved_model_file
                        )
                        if verbose:
                            self.logger.info(update_output)
                    self.best_valid_result = valid_result

                if callback_fn:
                    callback_fn(epoch_idx, valid_score)

                if stop_flag:
                    stop_output = "Finished training, best eval result in epoch %d" % (
                        epoch_idx - self.cur_step * self.eval_step
                    )
                    if verbose:
                        self.logger.info(stop_output)
                    break
        self._add_hparam_to_tensorboard(self.best_valid_score)
        return self.best_valid_score, self.best_valid_result

    def _train_epoch(self, train_data, epoch_idx, loss_func=None, show_progress=False):
        r"""Train the model in an epoch
        Args:
            train_data (DataLoader): The train data.
            epoch_idx (int): The current epoch id.
            loss_func (function): The loss function of :attr:`model`. If it is ``None``, the loss function will be
                :attr:`self.model.calculate_loss`. Defaults to ``None``.
            show_progress (bool): Show the progress of training epoch. Defaults to ``False``.
        Returns:
            float/tuple: The sum of loss returned by all batches in this epoch. If the loss in each batch contains
            multiple parts and the model return these multiple parts loss instead of the sum of loss, it will return a
            tuple which includes the sum of loss in each part.
        """
        self.model.train()
        loss_func = loss_func or self.model.calculate_loss
        total_loss = None
        iter_data = (
            tqdm(
                train_data,
                total=len(train_data),
                ncols=100,
                desc=set_color(f"Train {epoch_idx:>5}", "pink"),
            )
            if show_progress
            else train_data
        )
        scaler = amp.GradScaler(enabled=self.enable_scaler)

        if not self.config["single_spec"] and train_data.shuffle:
            train_data.sampler.set_epoch(epoch_idx)

        for batch_idx, interaction in enumerate(iter_data):
            interaction = interaction.to(self.device)
            self.optimizer.zero_grad()
            sync_loss = 0
            if not self.config["single_spec"]:
                self.set_reduce_hook()
                sync_loss = self.sync_grad_loss()

            with amp.autocast(enabled=self.enable_amp):
                losses = loss_func(interaction)

            if isinstance(losses, tuple):
                if epoch_idx < self.config["warm_up_step"]:
                    losses = losses[:-1]
                loss = sum(losses)
                loss_tuple = tuple(per_loss.item() for per_loss in losses)
                total_loss = (
                    loss_tuple
                    if total_loss is None
                    else tuple(map(sum, zip(total_loss, loss_tuple)))
                )
            else:
                loss = losses
                total_loss = (
                    losses.item() if total_loss is None else total_loss + losses.item()
                )
            self._check_nan(loss)
            scaler.scale(loss + sync_loss).backward()

            if self.clip_grad_norm:
                clip_grad_norm_(self.model.parameters(), **self.clip_grad_norm)
            scaler.step(self.optimizer)
            scaler.update()
            if self.gpu_available and show_progress:
                iter_data.set_postfix_str(
                    set_color("GPU RAM: " + get_gpu_usage(self.device), "yellow")
                )
        return total_loss