whisper.cpp

f8c75dc4 - CUDA: fix crash on large batch size for MoE models (llama/13384)