A Formal Perspective on Byte-Pair Encoding

Abstract

Byte-Pair Encoding (BPE) is a popular algorithm used for tokenizing data in NLP, despite being devised initially as a compression method. BPE appears to be a greedy algorithm at face value, but the underlying optimization problem that BPE seeks to solve has not yet been laid down. We formalize BPE as a combinatorial optimization problem. Via submodular functions, we prove that the iterative greedy version is a 1 σ(µ ⋆ ) (1 − e −σ(µ ⋆ ) )-approximation of an optimal merge sequence, where σ(µ ⋆ ) is the total backward curvature with respect to the optimal merge sequence µ ⋆ . Empirically the lower bound of the approximation is ≈ 0.37.We provide a faster implementation of BPE which improves the runtime complexity from O (N M ) to O (N log M ), where N is the sequence length and M is the merge count. Finally, we optimize the brute-force algorithm for optimal BPE using memoization.

Tokenization and the Noiseless Channel

“…This procedure is repeated until we fill the predefined vocabulary budget. Zouhar et al (2023) show that this greedy approach is approximately optimal when searching for the vocabulary (merge sequence). We are however interested in intentionally suboptimal vocabularies.…”