CtrlCoT: Dual-Granularity Chain-of-Thought Compression for Controllable Reasoning
Zhenxuan Fan, Jie Cao, Yang Dai, Zheqi Lv, Wenqiao Zhang, Zhongle Xie, Peng LU, Beng Chin Ooi

TL;DR
CtrlCoT introduces a dual-granularity compression framework for chain-of-thought reasoning that reduces token usage by 30.7% while improving reasoning accuracy, balancing semantic abstraction and token-level pruning.
Contribution
It proposes a novel dual-granularity approach combining hierarchical reasoning abstraction and logic-preserving pruning with distribution alignment for efficient CoT compression.
Findings
Achieves 30.7% token reduction on MATH-500 dataset.
Outperforms baselines with 7.6 percentage points higher accuracy.
Demonstrates more reliable reasoning with compressed traces.
Abstract
Chain-of-thought (CoT) prompting improves LLM reasoning but incurs high latency and memory cost due to verbose traces, motivating CoT compression with preserved correctness. Existing methods either shorten CoTs at the semantic level, which is often conservative, or prune tokens aggressively, which can miss task-critical cues and degrade accuracy. Moreover, combining the two is non-trivial due to sequential dependency, task-agnostic pruning, and distribution mismatch. We propose \textbf{CtrlCoT}, a dual-granularity CoT compression framework that harmonizes semantic abstraction and token-level pruning through three components: Hierarchical Reasoning Abstraction produces CoTs at multiple semantic granularities; Logic-Preserving Distillation trains a logic-aware pruner to retain indispensable reasoning cues (e.g., numbers and operators) across pruning ratios; and Distribution-Alignment…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsTopic Modeling · Natural Language Processing Techniques · Logic, Reasoning, and Knowledge
