[functionstackx]

Expand the glossary from 194 to 204 terms in both English and Simplified Chinese. The additions cover open-loop load, request arrival and completion rates, queue depth and waiting time, KV working sets, eviction, request preemption, draft acceptance rate, and inter-chunk latency.

Correct acceptance length so it no longer treats a fraction as a token count or assumes all engines exclude bonus tokens. Clarify AgentX client concurrency and the distinction between token-normalized latency and streamed chunk gaps. Existing entry URLs remain unchanged.

The skill reference records the new terminology and the API coverage exclusion: this is static editorial content, with no new measured fields, selectors, calculations, or API contracts.

Validation details

Local browser integration checks rendered all 20 new localized term pages at 375px with no horizontal overflow or page errors. Verified acronym search, category and alphabet filters, empty-state recovery, related-term navigation, and the English/Chinese language toggle. Desktop and mobile screenshots were inspected. The local Cypress smoke suite passed, including the new bilingual glossary journeys.

All six component shards and sixteen Chrome integration shards passed in the final-head E2E run. The glossary entries also appear in localized sitemap URLs and expose DefinedTerm JSON-LD.

Technical references

• vLLM acceptance metrics distinguishes accepted draft fractions from bonus-inclusive mean acceptance length.
• AIPerf metrics reference documents token-normalized latency, chunk timing, and request throughput.
• vLLM preemption documents KV-pressure preemption and recomputation.
• Each new entry links to relevant existing InferenceX articles and related terms.

AI model disclosure

GPT-6 Astra prepared the content, translations, regression tests, and verification. No delegated models contributed.

Chinese editorial review

The English and Chinese entries are paired by canonical slug. Numeric units, distinctions, and scope qualifiers were checked together. Human Chinese-maintainer editorial review has not been performed.

中文

术语表由 194 项扩充至 204 项，英文和简体中文同步更新。新增内容涵盖开环负载、请求到达与完成速率、队列深度与排队时间、KV 工作集、缓存驱逐、请求抢占、草稿接受率及流式 chunk 间延迟。

修正接受长度的定义，明确接受率与 token 数的区别，并说明部分引擎会计入目标模型额外生成的 token。同时补充 AgentX 客户端并发量和流式计时的口径说明，保留所有现有词条 URL。

同步更新 API skill 的术语说明和覆盖范围记录。本次仅修改静态说明内容，不新增实测字段、筛选条件、计算逻辑或 API 契约。中文维护者尚未进行人工编辑审核。

_{Created with Perplexity}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
inferencemax-app	Ready	Preview	Nov 14, 2026 6:50am UTC