GGML Fragmentation Issue

[zhouwg]

This issue could be found in various inference engines based on ggml library, such as bark.cpp, stable-diffusion.cpp, chatglm......

What's the "Fragmentation Issue of a complicated software": https://www.techlila.com/android-fragmentation/

A reasonable/workaroud method is integrate these derived inference engines into llama.cpp/examples/ and ensure them works fine as expected with updated ggml library.