whisper.cpp

1b590bbb - vulkan: use GL_NV_cooperative_matrix_decode_vector for faster matmul (llama/23541)