CritICL：從小型語言模型失敗模式中實現推理時弱到強泛化

摘要

近期在推論時擴展（inference-time scaling）方面的進展已顯著提升了大型語言模型（LLMs）的推理性能。然而，這些方法通常依賴於重複生成或外部驗證。為了解決此限制，我們提出 CritICL，一個新穎的推論時框架，能在保持高效率的同時改善推理能力。我們的關鍵洞察是，同一家族內不同規模的 LLM 會表現出結構化的失敗模式。CritICL 不將失敗視為不理想的輸出，而是將其利用為指導來源。具體而言，我們利用從較弱模型推導出的失敗模式，並透過基於評論的上下文範例（critique-based in-context examples）將其納入推論。我們提出兩個變體：CritICL-dynamic，它自適應地預測特定於輸入的失敗模式並檢索評論；以及 CritICL-static，它使用全域失敗模式概況提供穩定的指導。實驗結果顯示，CritICL 持續優於標準的上下文學習，並達到與測試時擴展方法競爭或更優的性能，同時所需的生成次數和 token 成本顯著更低。程式碼位於：https://github.com/umwyf/CRITICL

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