关于训练的问题

[BrianWayland]

您好，我训练时使用chinese_cleaners2来处理中文文本，然后在训练时输出了以下的异常信息：

Traceback (most recent call last):
  File "/home/featurize/data/vits-main/train_ms.py", line 119, in run
    train_and_evaluate(rank, epoch, hps, [net_g, net_d], [optim_g, optim_d], [scheduler_g, scheduler_d], scaler, [train_loader, eval_loader], logger, [writer, writer_eval])
  File "/home/featurize/data/vits-main/train_ms.py", line 147, in train_and_evaluate
    (z, z_p, m_p, logs_p, m_q, logs_q) = net_g(x, x_lengths, spec, spec_lengths, speakers)
  File "/home/featurize/data/vits-main/models.py", line 467, in forward
    z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
  File "/home/featurize/data/vits-main/models.py", line 237, in forward
    x = self.enc(x, x_mask, g=g)
  File "/environment/miniconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/featurize/data/vits-main/modules.py", line 166, in forward
    n_channels_tensor)
RuntimeError: The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
  File "/home/featurize/data/vits-main/commons.py", line 103, in fused_add_tanh_sigmoid_multiply
def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
  n_channels_int = n_channels[0]
  in_act = input_a + input_b
           ~~~~~~~~~~~~~~~~~ <--- HERE

在打印了input_a和input_b的维度后，发现input_a的形状是[64, 384, 500]，input_b的形状是[64, 384, 1]这两种形状的张量是可以通过广播进行相加的，不存在不能够相加的问题。此外，有时候输出异常信息时会提示后验编码器的卷积层计算错误。
请问您遇到了这种情况吗，又是如何解决的呢。我认为这种问题应该跟多卡训练有关，但问题是我使用的只有一张卡，也就是创建的只有一个训练进程，按理来说不会发生数据冲突之类的问题。期待您的回复。