llama.cpp

0bbc87b1 - vulkan: for small AMD GPUs, reduce submission threshold based on CU count (#25240)