r2 student full readouts

every trained r2/r2s student · every token position · all 17 layers · click a cell for the full readout · α-scaled unit injection of the base-model residual