llama.cpp

30af6e2b - ggml: auto apply iGPU flag CUDA/HIP if integrated device (#23007)