I am trying to apply Collaborative Filtering (Lesson 5) on Book-Crossing Dataset (http://www2.informatik.uni-freiburg.de/~cziegler/BX/). But it seems the loss is fluctuating between 11 and 12 and doesn’t seems to improve despite trying different learning rates. (the lr_find method suggest a learning rate of about 1e-3). Any suggestions that can be applied to improve the model?