arXiv cs.CL· Juan S. Santillana·· 12 小时前AI 评分18
耦合负载下的条件校准契约：切换式推理的分支感知审计
Conditional Calibration Contracts under Coupled Load: A Branch-Aware Audit of Switched Inference
AI 导读
研究将突发流量下的条件校准契约作为待验证规范，用 CA-MHD 在离散事件模拟器中结合负载耦合蒸馏、闭环回放与成本感知 Schmitt 路由器，覆盖 10 个训练种子 × 4 个内容-负载耦合级别共 40 个单元。
正文 · 原文
Abstract:Bursty serving changes both queue states and the population selected for each model. We study conditional calibration contracts as a specification to be validated, rather than a certified property of an existing router. CA-MHD combines load-coupled distillation, closed-loop replay and a cost-informed Schmitt router in a discrete-event simulator with a hardware service table. Across 40 cells (ten training seeds and four content-load coupling levels), we distinguish the original teacher, frozen wide heads on an adapted trunk, and the narrow student. A common teacher-trajectory audit has median absolute student-teacher deficit 0.06 percentage points, but does not certify the student-selected deployment support. Historical results use a kernel correctness gap, not the published SmoothECE; we separately prove selected-branch composition for a residual functional at a common fixed scale. The deployed wide branch has lower ordinal accuracy than the original teacher in all 40 cells. Same-checkpoint student-only serving has lower historical ordinal gap in 27/40 cells; routing also exposes violation tails up to 17.5 percent. Student p99 latency remains above the 1.5 ms target. Retrospective temporal recalibration improves the ordinal gap in 39/40 cells, but causal delayed-label availability is unverified and no incremental per-mode benefit is detected after multiplicity adjustment. An in-sample CUSUM detects 30/40 checkpoint changes at five historical false alarms, with no prospective false-alarm guarantee. These findings support a branch-aware audit and a scoped negative result, not demonstrated routing superiority. Independent budgets, selected-support validation and net utility under a satisfied SLA remain open.
| Comments: | 12 pages including references and appendices, ACM sigconf. Historical 40-cell campaign (10 training seeds x 4 coupling levels), with separate three-seed sweeps/ablations. Includes proofs, prospective FIFO mechanism checks, and limitations; prospective contract validation remains open |
| Subjects: | Machine Learning (cs.LG); Computation and Language (cs.CL); Distributed, Parallel, and Cluster Computing (cs.DC) |
| Cite as: | arXiv:2611.02121 [cs.LG] |
| (or arXiv:2611.02121v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2611.02121 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Juan Salas [view email]
[v1]
Sun, 1 Nov 2026 18:56:13 UTC (460 KB)
来源：arXiv cs.CL · arxiv.org