Transformers过早停止思考，小LoRA可修复
用早期层rank-8 LoRA延长冻结Transformer的链式计算。
为什么重要
提示小规模任务LoRA可释放冻结模型中被低估的链式计算能力。
阅读原文arxiv.org
AI 解读（根据原文生成）
论文研究预训练Transformer在上下文中沿引用链推理时使用深度不足的问题。在早期一层加入任务训练的rank-8 LoRA、冻结全部模型权重，可显著延长计算。文中报告Qwen3-8B在24行链上精确率从15.5%提升到99%，Ouro-1.4B经循环可达到更长链。代码和交互式演示已提供。
要点（AI 据原文整理，数字均已与原文核对）
- Thirteen base models reliably follow only 1.4-3.6 lines, and extra pretrained loops add little.
- A task-trained rank-8 LoRA at one early layer extends this computation with all model weights frozen.
- Qwen3-8B improves from 15.5% to 99% exact accuracy on 24-line chains; a longer-trained LoRA reaches 50 lines.
- Ouro-1.4B reaches 60 lines after four loops and at least 160 after eight.
- A frozen-model measurement locates the last useful intervention layer within tolerance in three of four held-out models.
- Task-specific LoRAs also improve MuSiQue.