llama.cpp

vulkan: reject ops when a tensor is too large to allocate

#18646

Merged