No module named 'glow'

[jglozma]

I am trying to run example with TRTIS described on https://github.com/NVIDIA/DeepLearningExamples/tree/master/PyTorch/SpeechSynthesis/Tacotron2/notebooks/trtis

When running step that was supposed to produce ONNX model I have got following error

python export_waveglow_trt.py --waveglow waveglow_256channels_ljs_v3.pt --wn-channels 256 --amp-run
_:::NVLOGv0.2.2 None 1575584732.593067884 (/workspace/tacotron2/dllogger/logger.py:251) args: {"waveglow": "waveglow_256channels_ljs_v3.pt", "output": "waveglow.onnx", "amp_run": true, "sigma_infer": 0.6}
Traceback (most recent call last):
File "export_waveglow_trt.py", line 158, in
main()
File "export_waveglow_trt.py", line 155, in main
export_onnx(parser, args)
File "export_waveglow_trt.py", line 117, in export_onnx
waveglow = load_and_setup_model('WaveGlow', parser, args.waveglow, args.amp_run)
File "/workspace/tacotron2/inference.py", line 122, in load_and_setup_model
state_dict = torch.load(checkpoint)['state_dict']
File "/opt/conda/lib/python3.6/site-packages/torch/serialization.py", line 422, in load
return _load(f, map_location, pickle_module, **pickle_load_args)
File "/opt/conda/lib/python3.6/site-packages/torch/serialization.py", line 609, in load
result = unpickler.load()
ModuleNotFoundError: No module named 'glow'

[jglozma]

I found glow.py in https://github.com/NVIDIA/waveglow. Next hurdle to overcome is

root@b5fb39a822ce:/workspace/tacotron2# python export_waveglow_trt.py --waveglow waveglow_256channels.pt --wn-channels 256 --amp-run
:::NVLOGv0.2.2 None 1575856662.273496628 (/workspace/tacotron2/dllogger/logger.py:251) args: {"waveglow": "waveglow_256channels.pt", "output": "waveglow.onnx", "amp_run": true, "sigma_infer": 0.6}
/opt/conda/lib/python3.6/site-packages/torch/serialization.py:489: SourceChangeWarning: source code of class 'glow.WaveGlow' has changed. you can retrieve the original source code by accessing the object's source attribute or set torch.nn.Module.dump_patches = True and use the patch tool to revert the changes.
warnings.warn(msg, SourceChangeWarning)
/opt/conda/lib/python3.6/site-packages/torch/serialization.py:489: SourceChangeWarning: source code of class 'torch.nn.modules.conv.ConvTranspose1d' has changed. you can retrieve the original source code by accessing the object's source attribute or set torch.nn.Module.dump_patches = True and use the patch tool to revert the changes.
warnings.warn(msg, SourceChangeWarning)
/opt/conda/lib/python3.6/site-packages/torch/serialization.py:489: SourceChangeWarning: source code of class 'glow.WN' has changed. you can retrieve the original source code by accessing the object's source attribute or set torch.nn.Module.dump_patches = True and use the patch tool to revert the changes.
warnings.warn(msg, SourceChangeWarning)
/opt/conda/lib/python3.6/site-packages/torch/serialization.py:489: SourceChangeWarning: source code of class 'torch.nn.modules.conv.Conv1d' has changed. you can retrieve the original source code by accessing the object's source attribute or set torch.nn.Module.dump_patches = True and use the patch tool to revert the changes.
warnings.warn(msg, SourceChangeWarning)
/opt/conda/lib/python3.6/site-packages/torch/serialization.py:489: SourceChangeWarning: source code of class 'glow.Invertible1x1Conv' has changed. you can retrieve the original source code by accessing the object's source attribute or set torch.nn.Module.dump_patches = True and use the patch tool to revert the changes.
warnings.warn(msg, SourceChangeWarning)
Traceback (most recent call last):
File "export_waveglow_trt.py", line 158, in
main()
File "export_waveglow_trt.py", line 155, in main
export_onnx(parser, args)
File "export_waveglow_trt.py", line 117, in export_onnx
waveglow = load_and_setup_model('WaveGlow', parser, args.waveglow, args.amp_run)
File "/workspace/tacotron2/inference.py", line 122, in load_and_setup_model
state_dict = torch.load(checkpoint)['state_dict']
KeyError: 'state_dict'

[ghost]

Hi @jglozma how did you train your checkpoint? Does it come from the https://github.com/NVIDIA/waveglow repository?

[jglozma]

Hi,

I have not done any training myself yet. I downloaded two checkpoints from NGC
https://ngc.nvidia.com/catalog/models/nvidia:waveglow_ljs_256channels/files

• the latest waveglow_256channels_ljs_v3.pt
• few month ago waveglow_256channels.pt

Both checkpoints produce the same error above

[ghost]

please download the checkpoint from https://ngc.nvidia.com/models/nvidia:waveglow256pyt_fp16

[jglozma]

It worked. Thanks for your help!

[kb-rahul]

@grzegorzkarchnv : I am unable to access the model from the link. Has anything changed recently?

[jglozma]

it is there. By default, it opens at "Overview". Click on "File Browser" and you will see it

[kb-rahul]

Thanks

[MuruganR96]

please download the checkpoint from https://ngc.nvidia.com/models/nvidia:waveglow256pyt_fp16

@grzegorzkarchnv sir, i have tried it, but unable to download. facing issues. can you please look into this issue.
#754 (comment)
#763 (comment)