如何将标记器输出转换为Huggingface Transformers中Trainer所需的train_dataset？

from transformers import Trainer, TrainingArguments
from torch.utils.data import Dataset

class CustomDataset(Dataset):
    def __init__(self, tokenized_inputs, labels):
        self.tokenized_inputs = tokenized_inputs
        self.labels = labels

    def __len__(self):
        return len(self.labels)

    def __getitem__(self, idx):
        return {
            'input_ids': self.tokenized_inputs['input_ids'][idx],
            'attention_mask': self.tokenized_inputs['attention_mask'][idx],
            'labels': self.labels[idx]
        }

tokenized_inputs = ...  # 标记器的输出，包含input_ids和attention_mask等字段
labels = ...  # 标签数据

dataset = CustomDataset(tokenized_inputs, labels)

training_args = TrainingArguments(
    output_dir='./results',  # 训练结果的输出目录
    num_train_epochs=3,  # 训练的轮数
    per_device_train_batch_size=16,  # 每个设备的训练批次大小
    learning_rate=2e-5,  # 学习率
)

trainer = Trainer(
    model=model,  # 模型
    args=training_args,  # 训练参数
    train_dataset=dataset,  # 自定义的Dataset
)

trainer.train()