ModuleNotFoundError: No module named 'nvgpu' in sagemaker batch transform

Question

I am trying to do batch transform inference with a `ml.g4dn.xlarge` instance using a GPU. However, when I run the inference, I get
`ModuleNotFoundError: No module named 'nvgpu'`
I tried to add the nvgpu library to the requirements.txt file of the model when training, but i have not found a version that works. 
I also read here : https://github.com/pytorch/serve/issues/1813#issuecomment-1231025086 that the issue could be the framework_version, but I have tried using 1.9.0 like it suggest but the issue remains. 
Any idea what could be the issue or what version of nvgpu to add to requirements?

Answer

Hi Alex,

I just tried to replicate the same using the following example,

https://github.com/aws/amazon-sagemaker-examples/blob/main/sagemaker-python-sdk/pytorch_batch_inference/sagemaker_batch_inference_torchserve.ipynb

I did not face this issue. I have tried the framework versions “1.9.0” and "1.13.1” with the instance type “ml.g4dn.xlarge”. Can you try using the more recent framework versions starting with 1.13.1?

If you are still facing the issue, please share the code example that you are following at your end.

Answer

I am also encountering this issue with py_version="py38" and framework_version="1.12" with ml.p2.xlarge.  Any solution would be appreciated.

ModuleNotFoundError: No module named 'nvgpu' in sagemaker batch transform

Relevant content