Comment by carbocation

My kingdom for renaming this paper to something like "Tensor Product Attention is a Memory-Efficient Approach for Long-Sequence Language Modeling"