固定推理预算下的答案交付研究：停止策略与终结提示的组合实验
Answer Delivery Under a Shared Token Budget: An Audited Study of Stopping and Finalization
论文测了在固定 token 预算下怎么让模型真的把答案交出来，Qwen3-4B 和 DeepSeek 蒸馏版各涨了 8 和 7 个点，做推理部署的可以看看。
一项 arXiv 研究针对固定推理预算下的答案交付问题，测试了 first-valid-box 停止与 charged finalization cue 组合的策略。在每模型 100 道独立审计的 GSM8K 测试题上，该策略使 Qwen3-4B-Thinking-2507 的交付端点准确率相比 first-box 停止提升 8.0 个百分点，DeepSeek-R1-Distill-Llama-8B 提升 7.0 个百分点。预锁定重复实验给出 7.0 和 10.0 个百分点的提升，两个种子下均未观察到相对 first-box 停止的明显损害。物理时隙节省的点估计低于 1%，实验覆盖两个检查点、一个任务和两个固定种子条件。
Answer Delivery Under a Shared Token Budget: An Audited Study of Stopping and Finalization
An inference budget is useful only when the system delivers an answer within it. We study a fixed-budget policy combining first-valid-box stopping with a charged finalization cue. On 100 fresh, independently audited GSM8K test questions per model, it improves delivered endpoint accuracy over first-box stopping by 8.0 percentage points for Qwen3-4B-Thinking-2507 and 7.0 for DeepSeek-R1-Distill-Llama-8B. A prelocked repeat on the same questions gives 7.0 and 10.0 points; neither seed shows a gross harm relative to first-box stopping. The point estimates of incremental physical-slot savings are below one percent. A finalization-only ablation separates cue recovery from stopping, and a fixed scalar reader tests extraction as an alternative explanation. The cue is prior art; the contribution is a component and readout analysis with charged injections and verified paths. Results cover two checkpoints, one task, one prompt family, and two fixed seed conditions.