ID 579879 - The Inference node does not properly serialize its knobs from .cat files

Problem summary:
The Inference node does not properly serialize its knobs from .cat files
 
Customer reported version:
Nuke 14.0v6
 
Customer reported platform:
Windows 10
 
Steps to reproduce:
1) Have a .cat file that contains at least one custom knob, such as the semantic segmentation model available on the Cattery: https://community.foundry.com/cattery/37764/semantic-segmentation
2) Launch NukeX and create an Inference node in the Node Graph.
3) In the Inference1 node's Properties, select the .cat file and set the custom knob(s) to a non-default value. In this example, the Object knob is switched from "wall" to "road":

4) Copy the Inference 1 node and paste it into the Node Graph. Do not open the Properties for the Inference2 node.
5) Copy the new Inference2 node and paste it into the Node Graph.
6) Open the Properties for all three Inference nodes, and observe how the knob values have been reset to default and were not saved for Inference3:

7) Save, close, and re-open this Nuke script.
8) Open the Properties for all three Inference nodes, and observe how the knob values have been reset to default and were not saved for Inference2:

 
Expected behavior:
The Inference node's knobs should always be serialized and saved.
 
Actual behavior:
Inference nodes with custom knobs from a .cat model file are not saved in all scenarios and can be reset to the default values.
 
Workaround:
No known workaround at this time.
 
Reproduced by Support in:
NukeX 15.1v1 - Windows 10, macOS 13 Ventura
NukeX 15.0v1 - Windows 10
NukeX 14.0v1 - Windows 10
NukeX 13.1v1 - Windows 10
 
Earliest version tested:
NukeX 13.1v1 - This feature (the Inference node's ability to load .cat files created from PyTorch models) did not exist before this version