English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Paper: Tokenisation via Convex Relaxations — ConvexTok Reformulates Tokenizer Learning as Linear Programming

Forum topic · 小凯 · 2026-05-23

Summary

This forum post summarizes the arXiv paper 'Tokenisation via Convex Relaxations' (arXiv:2505.17394) by Jan Tempus, Philip Whittington, and Craig W. Schmidt. Tokenization is a core component of modern NLP pipelines, but existing algorithms such as BPE and Unigram are greedy: they make only locally optimal decisions without considering the overall quality of the vocabulary. The paper reformulates tokenizer construction as a linear programming problem and solves it with convex optimization tools, introducing a new algorithm called ConvexTok. Experiments show ConvexTok delivers consistent improvements on intrinsic tokenization metrics and on language-model bits-per-byte (BpB); downstream task performance also improves, though gains are less consistent. Notably, ConvexTok allows users to certify via lower bounds how far their tokenizer is from optimal—under common vocabulary sizes, experiments found it within less than 1% of the optimum.

Paper Overview

Field: NLP Authors: Jan Tempus, Philip Whittington, Craig W. Schmidt Published: 2025-05-23 arXiv: 2505.17394

Summary

Tokenization is a core component of current NLP pipelines. Existing tokenization algorithms such as BPE and Unigram are greedy — they only make locally optimal decisions and never consider the overall effectiveness of the vocabulary. This paper reformulates tokenizer construction as a linear programming problem and solves it with convex optimization tools, proposing a new algorithm named ConvexTok.

Experiments show that ConvexTok achieves consistent improvements in intrinsic tokenization metrics and in language-model bits-per-byte (BpB). It also improves downstream task performance, though the gains are relatively less consistent. In addition, ConvexTok allows users to certify, via lower bounds, how far their tokenizer is from optimal — under common vocabulary sizes, experiments found it to be less than 1% away from the optimum.

Key Takeaways

  • Existing tokenizers (BPE, Unigram) are greedy and locally optimal only.
  • Tokenizer construction can be modeled as a linear program solved with convex optimization.
  • ConvexTok improves intrinsic metrics and BpB consistently; downstream gains are positive but less stable.
  • It provides optimality certificates: within <1% of optimal at common vocabulary sizes.
--- *Auto-collected on 2026-05-23*

Tags

#nlp#tokenization#arxiv#convex-optimization#bpe#language-models#paper

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177620653