用小模型的错误模式指导大模型推理，提升效率与准确性
CritICL: Inference-Time Weak-to-Strong Generalization from Small Language Model Failure Modes
- 利用弱模型失败模式生成批判性提示，动态或静态引导推理
- 在多个基准上超越标准上下文学习，接近甚至超过测试时扩展方法
- 仅需少量生成，显著降低计算成本，适合资源受限场景
近期的推理时扩展技术显著提升了大语言模型（LLM）的推理能力，但通常依赖重复生成或外部验证。为解决这一问题，我们提出 CritICL——一种新型推理时框架，在保持高效率的同时改进推理表现。关键洞察是：同一系列模型中，不同规模的模型失败模式具有结构性规律。不将失败视为噪声，而是将其作为指导信号。具体而言，我们利用弱模型产生的失败模式，通过批判性上下文示例融入推理过程。提出两种变体：CritICL-dynamic 动态预测输入相关的失败模式并检索对应批判；CritICL-static 使用全局失败模式特征提供稳定引导。实验表明，CritICL 在多个基准上持续优于标准上下文学习，并达到或超越测试时扩展方法的表现，同时生成次数和令牌开销显著更低。代码已公开于：https://github.com/umwyf/CRITICL
原文摘要 · Abstract (English)
Recent advances in inference-time scaling have significantly improved the reasoning performance of large language models (LLMs). However, these methods typically rely on repeated generation or external verification. To address this limitation, we introduce CritICL, a novel inference-time framework that improves reasoning while maintaining high efficiency. Our key insight is that LLM failure modes exhibit structured patterns across model scales within the same family. Instead of treating failures as undesirable outputs, CritICL leverages them as a source of guidance. Specifically, we utilize failure modes derived from weaker models and incorporate them into inference through critique-based in-context examples. We propose two variants: CritICL-dynamic, which adaptively predicts input-specific failure modes and retrieves critiques, and CritICL-static, which uses a global failure mode profile to provide stable guidance. Experimental results show that CritICL consistently outperforms standard in-context learning and achieves performance competitive with or superior to test-time scaling methods, while requiring significantly fewer generations and lower token cost. Code available at: https://github.com/umwyf/CRITICL
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。