llama.cpp

1701d4c5 - opencl: mark `mul_mat` `f32f32` as supporting non-contiguous tensors (#13790)