whisper.cpp

b5dec374 - CUDA: mul_mat_vec_q for batch sizes > 1 (llama/5351)