llama.cpp

ca4a8370 - vulkan: reject ops when a tensor is too large to allocate (#18646)