tts_infer.py return segmentation fault

[ly1984]

Ubuntu 18.04
Python3.6.8,
Pytorch 1.3
GPU:1080ti

I've downloaded the tacotron2 and waveglow model form ngc
tacotron2 model：https://ngc.nvidia.com/catalog/models/nvidia:tacotron2_ljspeech
waveglow model：https://ngc.nvidia.com/catalog/models/nvidia:waveglow_ljspeech

NeMo/examples/tts/tts_infer.py

And run the below command，and got segmentation fault。
python3 tts_infer.py --spec_model=tacotron2 --spec_model_config=configs/tacotron2.yaml --spec_model_load_dir=tacotron2_checkopints/ --vocoder=waveglow --vocoder_model_config=configs/waveglow.yaml --vocoder_model_load_dir=waveglow_checkopints/ --save_dir=wav_files/ --eval_dataset=test.json

[blisc]

Can you please try again inside the NeMo container: https://ngc.nvidia.com/catalog/containers/nvidia:nemo

Can you also please run in it in float32 mode using the argument --amp_opt_level=O0?

[ly1984]

I've tested on Google Colab.and everything goes well. It should be my problem ,I will check my own environment later.