Cannot create model archive in waveglow example

Your issue may already be reported! Please search on the issue tracker before creating one.

Context

While creating model archive file for waveglow example the create_mar.sh script fails with 404 error.

+ wget https://api.ngc.nvidia.com/v2/models/nvidia/tacotron2pyt_fp32/versions/1/files/nvidia_tacotron2pyt_fp32_20190306.pth
--2020-10-29 15:08:01--  https://api.ngc.nvidia.com/v2/models/nvidia/tacotron2pyt_fp32/versions/1/files/nvidia_tacotron2pyt_fp32_20190306.pth
Resolving api.ngc.nvidia.com (api.ngc.nvidia.com)... 54.213.120.107, 54.186.38.188
Connecting to api.ngc.nvidia.com (api.ngc.nvidia.com)|54.213.120.107|:443... connected.
HTTP request sent, awaiting response... 404 
2020-10-29 15:08:02 ERROR 404: (no description)

• torchserve version:
• torch version:1.6.0
• torchvision version [if any]:
• torchtext version [if any]:
• torchaudio version [if any]:
• java version:
• Operating System and version: Ubuntu 16.04 64bit

Your Environment

• Installed using source? [yes/no]:no
• Are you planning to deploy it using docker container? [yes/no]:yes
• Is it a CPU or GPU environment?:GPU
• Using a default/custom handler? [If possible upload/share custom handler/model]:
• What kind of model is it e.g. vision, text, audio?:audio
• Are you planning to use local models from model-store or public url being used e.g. from S3 bucket etc.? [If public url then provide link.]:
• Provide config.properties, logs [ts.log] and parameters used for model registration/update APIs:
• Link to your project [if any]:

Expected Behavior

Current Behavior

Possible Solution

Fix the broken links

Steps to Reproduce

1. Navigate to waveglow example in torchserve
2. Run create_mar.sh script …

Failure Logs [if any]

[alvarobartt]

The issue seems to be related with a model update NVIDIA did, which somehow removed/deleted the previous model, so the specified version of that model couldn’t be found at NVIDIA Catalog - Models.

On the other hand, both Tacotron2 and Waveglow models are available, but the version of those models is a more updated one than the one defined in the create_mar.sh script. Now the availability of those models (with the same configuration as the one used on the previous ones) is as it follows:

• Tacotron2 checkpoint can be found at https://ngc.nvidia.com/catalog/models/nvidia:tacotron2_pyt_ckpt_fp32 and the model (.pth) can be found at https://ngc.nvidia.com/catalog/models/nvidia:tacotron2_pyt_ckpt_fp32/files, as it figures as a separate file on the File Browser tab.
• Waveglow checkpoint can be found at https://ngc.nvidia.com/catalog/models/nvidia:waveglow_ckpt_fp32 and the model (.pth) can be found at https://ngc.nvidia.com/catalog/models/nvidia:waveglow_ckpt_fp32/files, as it figures as a separate file on the File Browser tab.

So that the create_mar.sh script has been updated with the new model URLs, anyway I will create the PR now so as to check whether the tests pass or not. In case that the tests work I’ll also resolve the other issues you send me as a reference!

[chauhang]

@alvarobartt Please also add the related requirements.txt file for the waveglow example so that the pip install dependencies get picked up by TorchServe for running the models.