Multiple TracerWarning on every epoch

I am running FastSpeech2 and every epoch gives me this:

/home/perry/PycharmProjects/espnet/espnet/nets/pytorch_backend/nets_utils.py:154: TracerWarning: Converting a tensor to a Python list might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  lengths = lengths.long().tolist()
/home/perry/PycharmProjects/espnet/espnet/nets/pytorch_backend/nets_utils.py:172: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert xs.size(0) == bs, (xs.size(0), bs)
/home/perry/PycharmProjects/espnet/espnet2/tts/feats_extract/log_mel_fbank.py:92: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert input_stft.shape[-1] == 2, input_stft.shape
/home/perry/PycharmProjects/espnet/espnet2/tts/fastspeech2/fastspeech2.py:533: TracerWarning: Iterating over a tensor might cause the trace to be incorrect. Passing a tensor of different shape won't change the number of iterations executed (and might lead to errors or silently give incorrect results).
  for i, l in enumerate(text_lengths):
/home/perry/PycharmProjects/espnet/espnet/nets/pytorch_backend/transformer/embedding.py:61: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if self.pe.size(1) >= x.size(1):
/home/perry/PycharmProjects/espnet/espnet/nets/pytorch_backend/transformer/attention.py:81: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
  numpy.finfo(torch.tensor(0, dtype=scores.dtype).numpy().dtype).min
/home/perry/PycharmProjects/espnet/espnet/nets/pytorch_backend/transformer/attention.py:81: TracerWarning: Converting a tensor to a NumPy array might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  numpy.finfo(torch.tensor(0, dtype=scores.dtype).numpy().dtype).min
/home/perry/PycharmProjects/espnet/espnet/nets/pytorch_backend/fastspeech/length_regulator.py:56: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if ds.sum() == 0:
/home/perry/PycharmProjects/espnet/espnet/nets/pytorch_backend/fastspeech/length_regulator.py:66: TracerWarning: Iterating over a tensor might cause the trace to be incorrect. Passing a tensor of different shape won't change the number of iterations executed (and might lead to errors or silently give incorrect results).
  repeat = [torch.repeat_interleave(x, d, dim=0) for x, d in zip(xs, ds)]
/home/perry/PycharmProjects/espnet/espnet/nets/pytorch_backend/nets_utils.py:55: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  max_len = max(x.size(0) for x in xs)
/home/perry/PycharmProjects/espnet/espnet2/tts/fastspeech2/fastspeech2.py:582: TracerWarning: Converting a tensor to a Python number might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  l1_loss=l1_loss.item(),
/home/perry/PycharmProjects/espnet/espnet2/tts/fastspeech2/fastspeech2.py:583: TracerWarning: Converting a tensor to a Python number might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  duration_loss=duration_loss.item(),
/home/perry/PycharmProjects/espnet/espnet2/tts/fastspeech2/fastspeech2.py:584: TracerWarning: Converting a tensor to a Python number might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  pitch_loss=pitch_loss.item(),
/home/perry/PycharmProjects/espnet/espnet2/tts/fastspeech2/fastspeech2.py:585: TracerWarning: Converting a tensor to a Python number might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  energy_loss=energy_loss.item(),
/home/perry/PycharmProjects/espnet/espnet2/tts/fastspeech2/fastspeech2.py:591: TracerWarning: Converting a tensor to a Python number might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  encoder_alpha=self.encoder.embed[-1].alpha.data.item(),
/home/perry/PycharmProjects/espnet/espnet2/tts/fastspeech2/fastspeech2.py:595: TracerWarning: Converting a tensor to a Python number might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  decoder_alpha=self.decoder.embed[-1].alpha.data.item(),
/home/perry/PycharmProjects/espnet/espnet2/tts/fastspeech2/fastspeech2.py:599: TracerWarning: Converting a tensor to a Python number might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  stats.update(loss=loss.item())
/home/perry/PycharmProjects/espnet/espnet2/torch_utils/device_funcs.py:64: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
  return torch.tensor([data], dtype=torch.float, device=device)

Is there some way to solve it?

[kan-bayashi]

It does not affect the training results, but a little bit annoying. Let us think how we should deal with this function.

[kan-bayashi]

I changed the graph creation to option and set to False as a default in #4551. I will close this issue.