ID 597018 - Working with .cat files larger than 4GB causes errors

Problem summary
Working with .cat files larger than 4GB causes errors
 
Depending on platform, than can happen either in the CatFileCreator node (Windows) or in the Inference node (Linux, macOS)
 
Customer reported version
nuke.15.1v3
 
Customer reported platform
rocky9
 
Steps to reproduce
 
1) Download the attached model_creator.py
 
2) In Terminal, run the following command, this may take a while:
 

python <path/to>/model_creator.py 1100 -o <path/to/output>.pt

*You can vary the number size to change how big the .pt file created is, 1100 should create a 5.2GB~ file
 
3) Launch NukeX
 
4) Create a CatFileCreator node
 
5) Load in the .pt file into the Torscript File knob, set the Cat File knob to an output path, and see the Channels knobs to rgba.red, rgba.green, rgba.blue, and give the Model ID knob a name:
 

 
6) Press Create .cat file and Inference
 
Expected behaviour
For the cat file to be created and the Inference node to be useable
 
Actual behaviour
On Windows, Nuke crashes when you press Create .cat file and Inference
On Linux and macOS, the file is created, but errors saying: Unable to load model: PytorchStreamReader failed reading zip archive: not a ZIP archive in Nuke 16 or Unable to open cat file in older versions
 
Workaround
None
 
Reproduced by support
This bug has been reproduced in:
NukeX 16.0v1 - Windows 10, Rocky 9, macOS Sonoma
NukeX 13.1v1 - Windows 10, Rocky 9
 
Earliest version tested
NukeX 13.1v1 - This feature did not exist before this version