llama.cpp

b2f460bd - vulkan: skip zero size tensors in backend copies (#20233)