Activation Control for Efficiently Eliciting Long Chain-of-thought Ability of Language Models
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866909621116993536 |
|---|---|
| author | Zhao, Zekai Liu, Qi Zhou, Kun Liu, Zihan Shao, Yifei Hu, Zhiting Huang, Biwei |
| author_facet | Zhao, Zekai Liu, Qi Zhou, Kun Liu, Zihan Shao, Yifei Hu, Zhiting Huang, Biwei |
| contents | Despite the remarkable reasoning performance, eliciting the long chain-of-thought (CoT) ability in large language models (LLMs) typically requires costly reinforcement learning or supervised fine-tuning on high-quality distilled data. We investigate the internal mechanisms behind this capability and show that a small set of high-impact activations in the last few layers largely governs long-form reasoning attributes, such as output length and self-reflection. By simply amplifying these activations and inserting "wait" tokens, we can invoke the long CoT ability without any training, resulting in significantly increased self-reflection rates and accuracy. Moreover, we find that the activation dynamics follow predictable trajectories, with a sharp rise after special tokens and a subsequent exponential decay. Building on these insights, we introduce a general training-free activation control technique. It leverages a few contrastive examples to identify key activations, and employs simple analytic functions to modulate their values at inference time to elicit long CoTs. Extensive experiments confirm the effectiveness of our method in efficiently eliciting long CoT reasoning in LLMs and improving their performance. Additionally, we propose a parameter-efficient fine-tuning method that trains only a last-layer activation amplification module and a few LoRA layers, outperforming full LoRA fine-tuning on reasoning benchmarks with significantly fewer parameters. Our code and data are publicly released. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_17697 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Activation Control for Efficiently Eliciting Long Chain-of-thought Ability of Language Models Zhao, Zekai Liu, Qi Zhou, Kun Liu, Zihan Shao, Yifei Hu, Zhiting Huang, Biwei Computation and Language Machine Learning Despite the remarkable reasoning performance, eliciting the long chain-of-thought (CoT) ability in large language models (LLMs) typically requires costly reinforcement learning or supervised fine-tuning on high-quality distilled data. We investigate the internal mechanisms behind this capability and show that a small set of high-impact activations in the last few layers largely governs long-form reasoning attributes, such as output length and self-reflection. By simply amplifying these activations and inserting "wait" tokens, we can invoke the long CoT ability without any training, resulting in significantly increased self-reflection rates and accuracy. Moreover, we find that the activation dynamics follow predictable trajectories, with a sharp rise after special tokens and a subsequent exponential decay. Building on these insights, we introduce a general training-free activation control technique. It leverages a few contrastive examples to identify key activations, and employs simple analytic functions to modulate their values at inference time to elicit long CoTs. Extensive experiments confirm the effectiveness of our method in efficiently eliciting long CoT reasoning in LLMs and improving their performance. Additionally, we propose a parameter-efficient fine-tuning method that trains only a last-layer activation amplification module and a few LoRA layers, outperforming full LoRA fine-tuning on reasoning benchmarks with significantly fewer parameters. Our code and data are publicly released. |
| title | Activation Control for Efficiently Eliciting Long Chain-of-thought Ability of Language Models |
| topic | Computation and Language Machine Learning |
| url | https://arxiv.org/abs/2505.17697 |