06RESEARCH PAPERPreprint
Mercury Token Economy and Internal Routing Cost
發表日期
摘要
ABSTRACTZenodo 原文A Mercury methods paper defining token economy as the joint measurement of answer quality, output token count, tokenizer burden, internal activation mass, route dispersion, per-token internal load, and latency decomposition. The paper asks why the same prompt can cost different numbers of tokens and different response times across language models. Copyright (c) 2026 Chen, Ho Yiing (norika). Licensed under Creative Commons Attribution 4.0 International (CC BY 4.0).
全文 PDF
FULL TEXT1 個檔- 下載 PDFpaper.pdf・50 KB・本站
PDF 與 Zenodo 上的檔案相同（大小與 md5 已核對）。
論文資料
DETAILSZenodo 紀錄 20757827- 作者
- Chen, Ho Yiing
- 發表日期
- 資源類型
- Preprint
- DOI
- 10.5281/zenodo.20757827
- Zenodo
- zenodo.org/records/20757827
- 授權
- Creative Commons Attribution 4.0 International
作者 Chen, Ho Yiing 是織田紀香的本名（陳禾穎）英文拼寫。標題、摘要、關鍵字與授權照 Zenodo 原文。
關鍵字
KEYWORDS20 個- Mercury
- token economy
- internal routing cost
- LLM latency
- tokenization
- output token count
- tokenizer burden
- route dispersion
- internal activation mass
- per-token internal load
- prompt cost
- model efficiency
- language model interpretability
- mechanistic interpretability
- local LLMs
- response latency
- prefill cost
- decode cost
- LLM evaluation
- model routing