How do I introduce a blocking function within sklearn's TF-IDF Vectorization method?

Question

To elaborate, I have a data table of nearly 15 million entries. My chief goal is to perform Approximate Nearest Neighbor matching with hopefully high precision. I am RAM locked,