Properly calculate cosine similarities for low memory on large datasets?

Question

I am following this tutorial here to just learn a bit about content recommenders: https://www.datacamp.com/community/tutorials/recommender-systems-python

but i ran into a