Nuance Ai Service Library updated to python 3.9

[nate-mozinski-microsoft]

The Nuance AI service library has been updated to Python 3.9. We are looking to test these updates with the Monai integration. 

Currently, the platforms/nuance Dockerfile uses pytorch:21.07-py3 and the requirements.txt file uses monai-deploy-app-sdk==0.4.0, monai[all]==0.9.0.
In order to support python 3.9, we updated to pytorch:23.12-py3.
So that we were using the latest versions of the libraries. We also updated to

monai-deploy-app-sdk==0.6.0
monai[all]==1.3.0.  

With these updates we received several errors in the logs:

ImportError: libcudart.so.11.0: cannot open shared object file: No such file or directory
Multiple versions of MONAI may have been installed?

After these errors the container exits.

We wanted to confirm that these versions should indeed be compatible with each other and if there is anything else we need to do to support the latest versions of these libraries. 

[nate-mozinski-microsoft]

To provide more context and additional information. I am working withing the monai-deploy-app-sdk/platforms/nuance_pin space.
My most recent updates include updating the Dockerfile to indicate the latest version of pytorch 24.01 and the latest lungct example code lung_nodule_ct_detection_v0.5.9(previously lung_nodule_ct_detection_v0.2.0).
I also updated the requirements.txt file from what is currently in the repo to:

monai-deploy-app-sdk==0.6.0
monai[all]
pydicom==2.3.0
highdicom==0.19.0
torch==2.0.1

All of the above are to try to bring me to the latest versions of the Monia tooling. This ends up being, pytorch 24.01, MD SDK 0.6.0, and Monai 1.3.0. The added torch==2.0.1 comes from #461 and resolves the libcudart.so.11.0 issue above.

With this combination I am getting through the build phase but am seeing error at runtime:

AttributeError: module 'monai.deploy.core' has no attribute 'input'

[nate-mozinski-microsoft]

Additional output from the error:

File "/app/app_wrapper.py", line 86, in initialize_class
monai_module = import_module(monai_app_class_module)
File "/usr/lib/python3.10/importlib/init.py", line 126, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "", line 1050, in _gcd_import
File "", line 1027, in _find_and_load
File "", line 1006, in _find_and_load_unlocked
File "", line 688, in _load_unlocked
File "", line 883, in exec_module
File "", line 241, in _call_with_frames_removed
File "/app/app/lung_nodule.py", line 16, in
from app.inference import LungNoduleInferenceOperator
File "/app/app/inference.py", line 79, in
@md.input("image", Image, IOType.IN_MEMORY)
AttributeError: module 'monai.deploy.core' has no attribute 'input'

This shows that the error stems from line 79 of platforms/nuance_pin/app/inference.py where monai.deploy.core attempts to call:

input("image", Image, IOType.IN_MEMORY)

Has this input functionality been deprecated or modified in one of the newer versions of the MONAI Deploy App SDK?