llama.cpp

vulkan: Use ggml_vk_tensor_subbuffer in mul_mat_vec(id) paths

#17244

Merged