how should i debug a memory realloc issue when in a big neural network inference case

[minisparrow]

in simple operator, i compiled a pytorch operator and get a compiled module and then load it via runtime(vulkan backend), it can get the correct result. But when run a neural network, it reports problem when realloc. (realloc(): invalide next size). how could we debug this issue and find what cause this problem? can we debug it one layer by layer to see which layer cause this memory problem? thanks all:)