A practitioner finds that their model trains well with learning rate 0.01 on a batch size of 32. They increase the batch size to 256. How should the learning rate be adjusted according to the linear scaling rule?