arXiv cs.CL· Samar Ansari·· 12 小时前AI 评分31
指令微调如何让语言模型学会并强化数字选择启发式
3, 7, 14, 17, 42: Instruction Tuning Installs and Sharpens Number-Choice Heuristics in Language Models
AI 导读
研究对 18 个开源模型系列在 N 最大到 100 的五个区间、29 种解码配置下精确计算模型对每个整数赋予的概率，其中 15 组有可用的基座与指令微调配对检查点。
正文 · 原文
Abstract:We measure, without sampling error, the probability a language model assigns to each integer from 1 to N when asked for a random number. For a fixed prompt the model defines one probability vector, which we compute by summing over each integer's scored spellings rather than estimate by sampling, removing the noise that limits earlier studies [1, 4]. We evaluate 18 open-weight model lines at five ranges up to N = 100, under 29 decoding configurations, on two inference stacks; 15 of them have matched base and instruction-tuned checkpoints that both give usable answers. Instruction tuning makes the distribution less uniform in all 15 pairs. It installs one heuristic, round-number avoidance, and sharpens another, endpoint avoidance. Base checkpoints do not avoid round numbers once the endpoints are removed from those sets; tuned checkpoints avoid them, multiples of 10 almost completely. Base checkpoints return the endpoints 1 and N at 0.30 to 0.60 of the interior rate; tuned checkpoints almost never return them. A preference for odd numbers is consistent across tuned checkpoints only at N = 100, where base checkpoints also show it weakly. Endpoint avoidance and the odd preference fit the account Kubovy and Psotka [6] proposed for human number choice; round-number avoidance is not part of that account, and the midpoint avoidance it predicts is essentially absent, with one slight exception at N = 100. Tuning does not install a new favourite integer shared across models: above N = 10 no modal integer is shared by more than 42 percent of tuned conditions, and P (17) rises in 10 of 15 pairs and falls in 5. The concentration is present in the model's own distribution: at temperature 1.0, nucleus truncation at top-p 0.95 adds a median of 0.035 to the total variation distance.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.36178 [cs.CL] |
| (or arXiv:2610.36178v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.36178 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Mohammad Samar Ansari Ph.D. [view email]
[v1]
Mon, 26 Oct 2026 19:12:46 UTC (132 KB)
来源：arXiv cs.CL · arxiv.org