llama.cpp

4696d567 - CUDA: fix crash on large batch size for quant. MoE (#13537)