← 研踪 PaperTrailarXiv:2606.30852 · v3
When Does Learning to Stop Help? Configuring Risk-Controlled Early Exit for Reasoning-Model Services
首次提交 2026-06-30 03:33
版本更新 2026-11-18 14:12
雷达观测 2026-11-19 19:08
摘要
Early exit lowers the cost of serving a reasoning model by stopping its reasoning once a provisional answer looks final. An operator must choose between a learned stopping rule and one static token budget, and savings against full reasoning do not settle the choice. We present a procedure that decides on calibration data from the workload. It certifies both families with Learn-then-Test to one bound on the fraction of all questions that full reasoning answers correctly but early exit does not, prices provisional answers as the serving stack does, and deploys the cheapest certified candidate. Across four models and five benchmarks neither family is better throughout. At a 5% risk target the learned rule saves 10.4-17.0 percentage points more tokens than the static budget on free-form mathematics with Qwen3, with accuracy 2.4-2.9 points below full reasoning, and less on MMLU-Pro on average. Which family costs less also changes with the risk target, in 8 of 15 settings, and with the price of a provisional answer. Across three prices, committing to one family in advance gives up as much as 27.4 points on some workload, whereas selecting on calibration data falls at most 11.6 points short of the better family, on a task with only 99 calibration questions. In a vLLM engine with Qwen3-8B, mean response time fell in proportion to generated tokens. Policies served on a task other than their calibration task exceeded the risk target in 18 of 44 task pairs; recalibration on the new task restored risk control.
作者备注:49 pages, 11 figures. Code: https://github.com/dongzhe1/llm-think-stop
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v3 · 2026-11-18 14:12 · 最新观测版本
When Does Learning to Stop Help? Configuring Risk-Controlled Early Exit for Reasoning-Model Services
Early exit lowers the cost of serving a reasoning model by stopping its reasoning once a provisional answer looks final. An operator must choose between a learned stopping rule and one static token budget, and savings against full reasoning do not settle the choice. We present a procedure that decides on calibration data from the workload. It certifies both families with Learn-then-Test to one bound on the fraction of all questions that full reasoning answers correctly but early exit does not, prices provisional answers as the serving stack does, and deploys the cheapest certified candidate. Across four models and five benchmarks neither family is better throughout. At a 5% risk target the learned rule saves 10.4-17.0 percentage points more tokens than the static budget on free-form mathematics with Qwen3, with accuracy 2.4-2.9 points below full reasoning, and less on MMLU-Pro on average. Which family costs less also changes with the risk target, in 8 of 15 settings, and with the price of a provisional answer. Across three prices, committing to one family in advance gives up as much as 27.4 points on some workload, whereas selecting on calibration data falls at most 11.6 points short of the better family, on a task with only 99 calibration questions. In a vLLM engine with Qwen3-8B, mean response time fell in proportion to generated tokens. Policies served on a task other than their calibration task exceeded the risk target in 18 of 44 task pairs; recalibration on the new task restored risk control.
查看此版本解读 →查看此版本原文 ↗