synthesis error

[BuaaAlban]

I can generate the wav with G-Lim, but failed with waveglow and tacotron2.
For waveglow:
Traceback (most recent call last):
File "synthesis.py", line 69, in
"results", words + "_" + str(num) + "_waveglow.wav"))
File "/scratch/workspace/bokangz/project/FastSpeech-master/waveglow/inference.py", line 36, in inference
audio = waveglow.infer(mel, sigma=sigma)
File "/scratch/workspace/bokangz/project/FastSpeech-master/glow.py", line 280, in infer
output = self.WN[k]((audio_0, spect))
File "/proj/xcdhdstaff1/bokangz/anaconda3/envs/fastSpeech/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in call
result = self.forward(*input, **kwargs)
File "/scratch/workspace/bokangz/project/FastSpeech-master/glow.py", line 166, in forward
torch.IntTensor([self.n_channels]))
RuntimeError: /pytorch/torch/csrc/jit/fuser/cuda/fused_kernel.cpp:202: a PTX JIT compilation failed

for tacotron2
Traceback (most recent call last):
File "synthesis.py", line 71, in
tacotron2 = utils.get_Tacotron2()
File "/scratch/workspace/bokangz/project/FastSpeech-master/utils.py", line 70, in get_Tacotron2
model.load_state_dict(torch.load(checkpoint_path)['state_dict'])
KeyError: 'state_dict'

[xcmyz]

You need download Tacotron2's model from https://github.com/NVIDIA/tacotron2

[BuaaAlban]

You need download Tacotron2's model from https://github.com/NVIDIA/tacotron2
But this problem still exist
File "/scratch/workspace/bokangz/project/FastSpeech-master/glow.py", line 166, in forward
torch.IntTensor([self.n_channels]))
RuntimeError: /pytorch/torch/csrc/jit/fuser/cuda/fused_kernel.cpp:202: a PTX JIT compilation failed

#---------------------------
sovled by comment line 33 in glow.py #@torch.jit.trace