用语义损失防止大模型在因果推理中崩溃，让模型真正理解逻辑。
On Semantic Loss Fine-Tuning Approach for Preventing Model Collapse in Causal Reasoning
- 引入基于图的逻辑约束和动态权重调度的语义损失函数
- 在传递性与d-分离任务上分别达70.4%和68.6%准确率，提升42.7%
- 适合需要稳定因果推理能力的研究者与开发者
在因果推理任务上对Transformer模型进行标准微调会导致灾难性模型崩溃，模型仅学习到始终预测“是”或“否”的简单模式。我们发现，在未使用语义损失的情况下，对Gemma 270M在传递性和d-分离任务上微调时，崩溃率达100%，模型虽获得73.9%的虚假高准确率，但未真正掌握因果推理。为此，我们提出一种包含图逻辑约束和动态lambda调度的语义损失函数，使模型在传递性任务上达到70.4%、在d-分离任务上达到68.6%的准确率，且预测稳定依赖上下文，相较崩溃基线提升42.7%。对抗测试显示，在1,000个结构化推理样本上，语义模型准确率达67%-70%，而崩溃模型仅43%-71%。通过在五种模型变体上超过20万样本的全面基准测试，验证了语义损失对稳定因果推理不可或缺。
原文摘要 · Abstract (English)
Standard fine-tuning of transformer models on causal reasoning tasks leads to catastrophic model collapse, where models learn trivial solutions such as always predicting "Yes" or "No" regardless of input structure. We demonstrate that fine-tuning Gemma 270M on transitivity and d-separation tasks without semantic loss results in 100% collapse rate, with models achieving misleadingly high accuracy (73.9%) while learning no causal reasoning. We propose a semantic loss function with graph-based logical constraints and dynamic lambda scheduling that prevents this collapse. Our approach achieves 70.4% accuracy on transitivity tasks and 68.6% on d-separation tasks with stable, context-dependent predictions, representing a 42.7% improvement over collapsed baselines. Adversarial evaluation on 1,000 structural reasoning samples shows semantic models achieve 67-70% accuracy while collapsed models fail catastrophically at 43-71%. We validate our findings through comprehensive benchmarking on 200,000+ evaluation samples across five model variants, demonstrating that semantic loss is essential and not optional, for stable causal reasoning in transformers.
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。