llama.cpp

Fix build shared ggml when CUDA is enabled

#1929

Merged