don’t just run a tool. know its limitations. memory leak checkers regularly emit false positives on real-world code.
those aren’t real memory leaks. it may be false positives or unavoidable due to the complexity of all involved libraries (threading, OpenCL, …). at the very least, those look like constant initial waste, not growing.
if you can present a situation that leaks memory at an appreciable rate and can make it to a gigabyte or more of growth on top of the initial process memory, then you’ve got a case.
any MRE for a “leaks!” claim better contain an infinite loop. if it doesn’t and process memory still grows boundlessly, that would be exceedingly rare.