arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2409.04599cs.CL
收藏

BPE 变得挑剔:在分词器训练期间高效优化词表

BPE Gets Picky: Efficient Vocabulary Refinement During Tokenizer Training

Center for Artificial Intelligence, Technical University of Applied Sciences Würzburg-Schweinfurt(维尔茨堡-施韦因富特应用技术大学人工智能中心);PleIAs;Institute of Computer Science, University of Tartu(塔尔图大学计算机科学研究所);University of California, San Diego(加利福尼亚大学圣迭戈分校)

Pavel Chizhov, Catherine Arnett, Elizaveta Korotkova, Ivan P. Yamshchikov

AI总结:

本文提出 Picky BPE 算法,在分词器训练过程中动态精炼词表,以消除训练不足的 token、提升词表效率并保持压缩能力,同时不降低甚至改善下游任务性能。

摘要:

语言模型可从高效分词中大幅获益。然而,它们仍主要使用经典的 BPE 算法,这是一种简单且可靠的方法。已有研究表明,这会导致训练不足的 token 和次优压缩等问题,进而可能影响下游性能。我们提出 Picky BPE,这是一种改进的 BPE 算法,可在分词器训练期间执行词表优化。我们的方法提高了词表效率,消除了训练不足的 token,并且不会损害文本压缩。实验表明,该方法不会降低下游性能,并且在若干情况下还能提升性能。

英文摘要:

Language models can largely benefit from efficient tokenization. However, they still mostly utilize the classical BPE algorithm, a simple and reliable method. This has been shown to cause such issues as under-trained tokens and sub-optimal compression that may affect the downstream performance. We introduce Picky BPE, a modified BPE algorithm that carries out vocabulary refinement during tokenizer training. Our method improves vocabulary efficiency, eliminates under-trained tokens, and does not compromise text compression. Our experiments show that our method does not reduce the downstream performance, and in several cases improves it.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑