Labs · Size and tokens

BPE against WordPiece

Two merge rules on one editable word table — the most frequent pair against the highest-scoring one — and how each then splits a word it has never seen.

score(a,b)=f(ab)f(a) f(b)\text{score}(a,b) = \frac{f(ab)}{f(a)\,f(b)}10 words · 20 merges

low, lower, lowest, new, newer, newest, wide, wider, widest, slow: the -er / -est endings become tokens.

Merge by merge, then a word neither has seen

Reads /api/labs/subword — loading

Computing…

BPE against WordPiece — Labs — TransformerLab