def model_train 训练模型并保存至modelpth train_data_list test_data_list = get_data_list train_data_list test_data_list = data_preprocesstrain_data_list test_data_list train_data_loader valid_data_

The function model_train() trains a multimodal model using the data provided by get_data_list(), after preprocessing it with data_preprocess(). It then creates data loaders using get_data_loader() and initializes the model with the bert-base-uncased pre-trained weights.

The optimizer used is AdamW with learning rate args.lr and weight decay args.weight_decay. The loss function used is CrossEntropyLoss().

The function then trains the model for args.epochs epochs, printing training and validation accuracy rates, as well as weighted accuracy, weighted F1 score, macro accuracy and macro F1 score.

If the validation accuracy rate is better than the previously saved best rate, the model is saved to model.pth.

The function returns nothing