Torch hub command not working

[raphpapercup]

Hello, Thanks a lot for your implementation. I am slightly confused because I am trying to run your torch hub command but it does not work.

Also if I look in /home/user/.cache/torch/checkpoints/, I cannot find the checkpoint even though I do get the logging: Downloading: "https://github.com/seungwonpark/melgan/releases/download/v0.1-alpha/nvidia_tacotron2_LJ11_epoch3200.pt" to /home/user/.cache/torch/checkpoints/nvidia_tacotron2_LJ11_epoch3200.pt

[seungwonpark]

Hi, can you please briefly describe about your environment?

In my case, it worked for CentOS but not for Windows 10.

[te0006]

Not OP but exactly the same error here. Environment:

• Ubuntu 16.04.06 LTS 'xenial'
• Python 3.6.9 (Anaconda)
• Pytorch 1.1.0
• CUDA 9.0_0, CuDNN 7.6.0

HTH - would be nice to try your work!

[MorganCZY]

@raphpapercup @te0006 Have you solve this error? I encountered the same error.

[phtephanx]

Bug is still persisting.

[delip]

I tried a different approach of manually downloading the pretrained model and then try loading it via PyTorch, and I hit a different bug for that.

[jacobhallberg]

This worked for me. I downloaded the checkpoint from the releases page.

from melgan.model.generator import Generator
from melgan.utils.hparams import HParam

checkpoint = torch.load("./nvidia_tacotron2_LJ11_epoch6400.pt") 
hp = HParam("./melgan/config/default.yaml")

melgan = Generator(hp.audio.n_mel_channels).cuda()
melgan.load_state_dict(checkpoint["model_g"])
melgan.eval(inference=False)

[janvainer]

I am having the same issue. not able to download the pretrained checkpoint.

hash_prefix = HASH_REGEX.search(filename).group(1) AttributeError: 'NoneType' object has no attribute 'group'

This worked for me. I downloaded the checkpoint from the releases page.

from melgan.model.generator import Generator
from melgan.utils.hparams import HParam

checkpoint = torch.load("./nvidia_tacotron2_LJ11_epoch6400.pt") 
hp = HParam("./melgan/config/default.yaml")

melgan = Generator(hp.audio.n_mel_channels).cuda()
melgan.load_state_dict(checkpoint["model_g"])
melgan.eval(inference=False)

This also worked for me. Go to Releases and download the checkpoint from the last release.

[deepconsc]

I tried a different approach of manually downloading the pretrained model and then try loading it via PyTorch, and I hit a different bug for that.

@delip you should reach checkpoint's generator module by loading the state dict via:

model.load_state_dict(torch.load('nvidia-tacotron2_LJ11_epoch6400.pt')['model_g'])

instead of

model.load_state_dict(torch.load('nvidia-tacotron2_LJ11_epoch6400.pt'))

[yihan0512]

having the same issue with torch.hub.load_state_dict_from_url()

[washingtonsk8]

I had the same problem but, in another application. In my case, I am the model "provider". Computing the SHA-256 Hash of the file and adding it to the checkpoint filename worked for me.

1. On Linux, I ran: sha256sum <checkpoint_model_filename.ext>
2. I copied the first eight digits of the hash and added them to the filename: checkpoint_model_filename-<sha256>.ext
3. Then, I changed the filename in the release.

So, in this case, the model's filename should be https://github.com/seungwonpark/melgan/releases/download/v0.1-alpha/nvidia_tacotron2_LJ11_epoch3200-<sha256>.pt

Source: https://pytorch.org/docs/stable/hub.html#torch.hub.load_state_dict_from_url