model memory leak profiling

[ghost]

How do you guys profile the memory usage of the deployed model?

I am deploying a model using triton serving, but turns out the host memory increases as the number of requests increases during the stress test. I am trying to figure out what part of my model occurs the leak and seeing this document.
I tried with nsys and capture the nsys report, but I couldn't find the memory heap stacks in the report...

Thank you in advance.

[ghost]

Found out this issue and it seems that the contributors are used to check the possible leaks with this tool. I will go with this tool.

[sboudouk]

Did you succeed into profiling & identifying your model leak ?