3 ms·is this approach viable on training? wat about kernels that require different grids?by perfobotto 1y agois this approach viable on training? wat about kernels that require different grids?