Primary sourcearXiv: OpenAI
View sourceEvent ·Official first-hand
在线分词实现最优延迟:对 tiktoken 首 token 延迟最高快 34.8 倍
arXiv 论文(2610.19139)形式化了在线分词问题,提出延迟指标衡量输出滞后于最优解的程度。作者给出近线性时间算法,延迟在常数因子内最优,并证明除非 P=NP,多项式时间算法无法将延迟再改进超过常数倍。实现版本兼容 OpenAI 的 tiktoken 与 HuggingFace 的 tokenizers,以首 token 时间计延迟较 tiktoken 快 7.9 到 34.8 倍。吞吐量较 tiktoken 提升 2.4 到 7.0 倍,较 tokenizers 提升 15.8 到 38.1 倍,内存占用近似恒定、不随输入规模增长。该实现同样超过近期的 gigatoken 分词器。
Current read
分词器也能流式跑了:新算法对比 tiktoken,首 token 延迟最高快 34.8 倍,内存还不随输入长度涨。
9 sources34° AI heatLast updated 2026/10/13 20:08:40
Evidence
9 sourcesRelated source 1AI Will
View sourceRelated source 2AIGCLINK
View sourceRelated source 3IT之家
View sourceRelated source 4Simon Willison’s Weblog
View sourceRelated source 5The Rundown AI
View sourceRelated source 6Aravind Srinivas
View sourceRelated source 7kimmonismus
View sourceRelated source 8elvis
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.