llama.cpp

5c86c9ed - CUDA: fix crash on large batch size for MoE models (#13384)