llama.cpp

4329d1ac - CUDA: use min compute capability of GPUs actually used (#2506)