igtrnt
igtrnt
@jcmgray Here is a bigger example: ``` inputs = ["ABC", "D", "EF", "CE", "G", "H", "BH", "D", "BH"] output = '' sizes = {'A':9, 'B':11, 'C':7, 'D':24,'E':8, 'F': 2, 'G':5,...
I don't see how cache can be fixed. The purpose of cache is to avoid recomputing resulting `k12` tensor. But since `k12` depends not only on `k1` and `k2`, but...
Yes, deduplication is done, but that's when `k1==k2` - those are contracted first. But I am talking about something else. Here is an example: `'ij,jz,jk,z->`', `sizes = {'i' : 10,...