llama.cpp

vulkan: add tensor-parallel allreduce function

#25051

Merged