让大模型在微调后更懂自己多有把握，避免盲目自信。
Functional-level Uncertainty Quantification for Calibrated Fine-tuning on LLMs
- 用专家混合框架，从功能层面动态校准模型置信度。
- 在多个任务上使预测误差率降低超25%，保持高准确率。
- 适合需要可靠置信度的场景，如医疗问答和自动评测。
大型语言模型（LLMs）的准确不确定性量化对可靠置信度估计至关重要，但基于参数高效微调（PEFT）的模型在数据有限时常出现过度自信问题。现有方法多为事后估算，未能提升适配器对特定输入输出关系的专属性。本文提出功能级不确定性量化校准微调（UQ4CT），通过提示相关的LoRA专家混合构造函数空间，以校准损失在训练中同步对齐功能级置信度与预测正确性。在四个多项选择基准和两个开放式生成型问答任务上，UQ4CT将期望校准误差（ECE）降低超过25%，同时保持高精度。在分布外情况下，仍维持优异校准性能和竞争力的准确性，显著提升微调后模型的可靠性与泛化能力。
原文摘要 · Abstract (English)
Accurate uncertainty quantification in large language models (LLMs) is essential for reliable confidence estimation, yet fine-tuned LLMs often become overconfident under limited adaptation data. Existing uncertainty methods for PEFT-based LLMs are largely post hoc, estimating uncertainty after fine-tuning rather than improving how adapters specialize to task-specific input-output relationships. We propose Functional-Level Uncertainty Quantification for Calibrated Fine-Tuning (UQ4CT), which calibrates uncertainty over the functional space induced by prompt-dependent mixtures of LoRA experts. UQ4CT implements this perspective through a mixture-of-experts fine-tuning framework, where a calibration loss aligns functional-level confidence with predictive correctness during training. Across four multiple-choice benchmarks and two open-ended generative QA tasks, UQ4CT reduces Expected Calibration Error (ECE) by over $25\%$ while preserving high accuracy. Under distribution shift, UQ4CT maintains superior calibration and competitive accuracy, demonstrating improved reliability and generalization for fine-tuned LLMs.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。