PECA：面向小模型临床推理的可微长度控制简洁化微调方法
Parameter-Efficient Conciseness Adaptation: Differentiable Length Control for Concise Clinical Reasoning in Smaller Language Models
AI 导读
论文提出 PECA（Parameter-Efficient Conciseness Adaptation），在 QLoRA 监督微调目标中加入可微简洁项，通过在生成超出 token 预算时提高序列结束概率来控制回答长度，无需偏好数据、辅助模型或专属教师模型。
正文 · 原文
Abstract:Smaller language models (SLMs) make it feasible to deploy clinical natural language processing where large proprietary models are too costly, and parameter-efficient fine-tuning allows them to be specialised on commodity hardware. One obstacle persists after training: on medical tasks these models answer at far greater length than the clinical question requires. In clinical decision support and point-of-care use, where time is short, such verbosity is hard to act on even when the underlying reasoning is sound. Existing length-control methods depend on preference optimisation, reward models, or proprietary teachers that undercut the accessibility making SLMs attractive. We introduce Parameter-Efficient Conciseness Adaptation (PECA), which adds a single differentiable conciseness term to the QLoRA supervised fine-tuning objective. Because the token count is discrete and non-differentiable, PECA instead raises the end-of-sequence probability once generation exceeds a token budget, a smooth signal optimised during ordinary training, with no preference data, no auxiliary model, and a single commodity GPU. On a 1.7-billion-parameter model fine-tuned on the medical-o1-reasoning-SFT corpus, PECA reduces mean response length by 72.8% while improving multiple-choice diagnostic accuracy from 40.8% to 44.0%. A sweep over the conciseness strength traces an interpretable accuracy-to-length frontier for balancing brevity and diagnostic completeness.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.17128 [cs.CL] |
| (or arXiv:2611.17128v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.17128 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Tanzeel Abbas Mr [view email]
[v1]
Thu, 12 Nov 2026 17:21:12 UTC (1,556 KB)
来源：arXiv cs.AI · arxiv.org