The tensor size does not match when I am training a Chinese-English-Mixed model

[busyyang]

I tried to train a Chinese-English-Mixed model by our own dataset. The training precess worked, but with some tensor size errors as below.

The expanded size of the tensor (32) must match the existing size (0) at non-singleton dimension 1.  Target sizes: [192, 32].  Tensor sizes: [192, 0]

And I noticed, there is a try .... catch part in training process.

    ......
    for epoch in range(epoch_str, hps.train.epochs + 1):
        try:
            if rank == 0:
                train_and_evaluate(
                    rank,
                    epoch,
                    hps,
                    [net_g, net_d, net_dur_disc],
                    [optim_g, optim_d, optim_dur_disc],
                    [scheduler_g, scheduler_d, scheduler_dur_disc],
                    scaler,
                    [train_loader, eval_loader],
                    logger,
                    [writer, writer_eval],
                )
            else:
                train_and_evaluate(
                    rank,
                    epoch,
                    hps,
                    [net_g, net_d, net_dur_disc],
                    [optim_g, optim_d, optim_dur_disc],
                    [scheduler_g, scheduler_d, scheduler_dur_disc],
                    scaler,
                    [train_loader, None],
                    None,
                    None,
                )
        except Exception as e:
            print(e)
            torch.cuda.empty_cache()
        scheduler_g.step()
        scheduler_d.step()
        if net_dur_disc is not None:
            scheduler_dur_disc.step()

Why should we add the try ....catch code for errors, and why the size of tensor may be not match during the training?