标注分歧高的样本在LoRA微调中会越训越差，且可预测。
Annotation Entropy Predicts Per-Example Learning Dynamics in LoRA Fine-Tuning
- 用标注熵衡量样本争议度，关联每例损失曲线下面积。
- 高争议样本的损失随训练上升，六模型均验证此现象。
- 适合关注微调稳定性或数据质量的研究者参考。
我们发现，LoRA微调在有争议的样本上会出现反向学习：标注者意见分歧大的样本在训练过程中损失持续上升，这种现象在全量微调中几乎不存在，且在六个测试模型（四个编码器、两个解码器）中均一致出现。该发现基于对ChaosNLI中每个样本100个标注的标注熵，与SNLI和MNLI上每例损失曲线下面积（AULC）的相关性分析。25种条件下相关性均为正（斯皮尔曼ρ=0.06–0.43），解码器模型在相同LoRA秩下的相关性强于编码器。该效应经部分相关性控制后仍存在，并在不同随机种子和数据集间可复现。初步的噪声注入实验与此结果一致。
原文摘要 · Abstract (English)
We find that LoRA fine-tuning exhibits un-learning on contested examples: items with high annotator disagreement show increasing loss during training, a qualitatively distinct pattern largely absent under full fine-tuning and consistent across all six models tested (four encoder, two decoder-only). This discovery emerges from correlating annotation entropy, computed from ChaosNLI's 100 labels per example, with per-example area under the loss curve (AULC) on SNLI and MNLI. The correlation is positive in all 25 conditions tested (Spearman $ρ= 0.06$-$0.43$), with decoder-only models showing stronger correlations than encoders at matched LoRA rank. The effect survives partial-correlation controls and replicates across seeds and datasets. A preliminary noise-injection experiment is consistent with these findings.
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。