llama.cpp

9eb496b4 - also support tinygemma3 test model