llama.cpp

550cf726 - CUDA: fix GET_ROWS for large tensors (#15882)