Verified80% confidenceFactTime unknown
BPE（Byte Pair Encoding）是大语言模型使用的Token化算法，通过统计训练语料中字符对的共现频率迭代合并最常见的字符组合
6
Sources
80%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Caveman：让Claude像原始人说话，省65% Token的开源神器
githubJuliusBrussee
Related Entities
Related Claims
VerifiedBPE算法起源于数据压缩领域，通过迭代合并高频字符对来构建词汇表79% similarUnverifiedMainstream large models generally adopt the BPE (Byte Pair Encoding) algorithm for tokenization78% similarUnverifiedBPE（字节对编码）算法最初是一种数据压缩算法，后被引入NLP领域并应用于GPT、Claude等主流大语言模型的分词器设计78% similarVerifiedBPE的核心思路是从字符级别出发，反复合并出现频率最高的相邻字符对，直到词汇表达到预设大小77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/21862API
curl https://kongchang.com/api/v1/knowledge/claims/21862MCP
get_claim(id=21862)