arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2605.07686cs.LG
收藏

耦合税:共享令牌预算如何削弱固定输出限制下的思维链

The Coupling Tax: How Shared Token Budgets Undermine Visible Chain-of-Thought Under Fixed Output Limits

National Taiwan University(国立台湾大学)

Wenhua Nie, Junlin Liu, Jianan Wu, Zijie Meng, Yilong Fan, Zhang Zijian, Haoran Zheng, Jyh-Shing Roger Jang

AI总结:

研究发现共享令牌预算会导致思维链推理效率下降,通过不同任务验证了耦合税现象,并提出分割预算生成作为缓解方法,提升数学任务准确率。

摘要:

思维链推理常被视为通过延长思考过程提升语言模型准确性的方式。本文识别出一种反向效应,即耦合税:当推理轨迹和最终答案共享同一个输出令牌预算时,长轨迹会挤占本应支持的答案。在GSM8K、MATH-500和五个BIG-Bench Hard任务中,非思考模式在预算上限2048 tokens内均匹配或优于思考模式,而更难的任务使交叉点向更大预算转移。我们推导出截断浪费分解式,Acc_think(b)=α_c F_L(b)+α_t(1-F_L(b)),该式从链长和准确性统计预测交叉点,并解释了Qwen家族内的反比例缩放。DeepSeek-R1-Distill-Llama-8B的复制显示,不同思考接口下同样存在此模式。作为缓解措施,分割预算生成将推理和答案预算分开;在完整MATH-500上,IRIS达到74.0%的准确率,强化提取变体达到78.8%,固定非Oracle SC+IRIS门达到83.6%。结果表明,测试时推理应被视为预算分配问题,而不仅仅是是否可用更长的轨迹。

英文摘要:

Chain-of-thought reasoning is often treated as a monotone way to improve language-model accuracy by letting a model think longer. We identify a countervailing effect, the coupling tax: when reasoning traces and final answers share one output-token budget, long traces can crowd out the answer they are meant to support. Across GSM8K, MATH-500, and five BIG-Bench Hard tasks with Qwen3 models at three scales, non-thinking mode matches or outperforms thinking mode on GSM8K and MATH-500 at every budget up to 2048 tokens, while harder tasks shift the crossover to larger budgets. We derive a truncation-waste decomposition, $\mathrm{Acc}_{\mathrm{think}}(b)=α_c F_L(b)+α_t(1-F_L(b))$, that predicts this crossover from chain-length and accuracy statistics and explains inverse scaling within the Qwen family. A DeepSeek-R1-Distill-Llama-8B replication shows the same pattern under a different thinking interface. As a mitigation, split-budget generation decouples reasoning and answer budgets; on full MATH-500, IRIS reaches 74.0% accuracy, a strengthened extraction variant reaches 78.8%, and a fixed non-oracle SC+IRIS gate reaches 83.6%. The results show that test-time reasoning should be evaluated as a budget-allocation problem, not only as a question of whether longer traces are available.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑