REVIEW 23 cited by
CoT-Valve: Length-Compressible Chain-of-Thought Tuning
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Chain-of-Thought significantly enhances a model's reasoning capability, but it also comes with a considerable increase in inference costs due to long chains. With the observation that the reasoning path can be easily compressed under easy tasks but struggle on hard tasks, we explore the feasibility of elastically controlling the length of reasoning paths with only one model, thereby reducing the inference overhead of reasoning models dynamically based on task difficulty. We introduce a new tuning and inference strategy named CoT-Valve, designed to allow models to generate reasoning chains of varying lengths. To achieve this, we propose to identify a direction in the parameter space that, when manipulated, can effectively control the length of generated CoT. Moreover, we show that this property is valuable for compressing the reasoning chain. We construct datasets with chains from long to short for the same questions and explore two enhanced strategies for CoT-Valve: (1) a precise length-compressible CoT tuning method, and (2) a progressive chain length compression approach. Our experiments show that CoT-Valve successfully enables controllability and compressibility of the chain and shows better performance than the prompt-based control. We applied this method to QwQ-32B-Preview, reducing reasoning chains on GSM8K from 741 to 225 tokens with a minor performance drop (95.07% to 94.92%) and on AIME from 6827 to 4629 tokens, with only one additional incorrect answer.
Forward citations
Cited by 23 Pith papers
-
rePIRL: Learn PRM with Inverse RL for LLM Reasoning
rePIRL learns token-level process rewards for LLM reasoning via a guided-cost-learning-style IRL objective, and shows these rewards improve reasoning policies on math/coding benchmarks.
-
Reconsidering Overthinking: Penalizing Internal and External Redundancy in CoT Reasoning
A dual-penalty RL method that compresses chain-of-thought traces by separately penalizing internal semantic stagnation and external post-answer continuation reduces reasoning length by about 40% while preserving accur...
-
Do Thinking Tokens Help or Trap? Towards More Efficient Large Reasoning Model
Suppressing 'thinking tokens' in a 1.5B reasoning model preserves accuracy while cutting tokens, and the proposed DuP-PO RL method improves both accuracy and efficiency over GRPO.
-
AALC: Large Language Model Efficient Reasoning via Adaptive Accuracy-Length Control
AALC, an accuracy-aware length reward, reduces reasoning-model response length by over 50% on math benchmarks while preserving or improving accuracy.
-
Exploring and Exploiting the Inherent Efficiency within Large Reasoning Models for Self-Guided Efficiency Enhancement
Large reasoning models already contain the ability to reason concisely; steering their internal activations or rewarding brevity can cut reasoning length by up to half without hurting accuracy.
-
How Far Are We from Optimal Reasoning Efficiency?
The authors define a reasoning efficiency frontier and a gap metric (REG), then train models with REO-RL to shrink the gap by at least 50% with only small accuracy losses.
-
PixelThink: Towards Efficient Chain-of-Pixel Reasoning
A soft token-budget penalty driven by task difficulty and model confidence roughly halves the reasoning tokens used by segmentation MLLMs while slightly improving mask accuracy on ReasonSeg-derived benchmarks.
-
Learning Composable Chains-of-Thought
Training atomic reasoning models on prefix and suffix tagged CoT data, then merging or multitask-combining them, improves compositional generalization on unseen skill combinations.
-
Is Your Model Thinking or Just Stagnating? PUMA: Diagnosing Reasoning Pathology via Phase-Momentum Alignment
PUMA detects when a reasoning model's entropy drop aligns with hidden-state momentum, truncates at that point, and reports improved accuracy-efficiency on 1.5B-32B reasoning models.
-
LISA: Linear-Indexed Sparse Attention for Efficient Long-Context Reasoning
LISA replaces standard attention with linear attention plus indexer-selected sparse attention, claiming O(nM) inference cost and accuracy gains, but the complexity analysis ignores the indexer's full-context scoring.
-
Implicit Reasoning in Large Language Models: A Comprehensive Survey
A survey organizing implicit (silent) reasoning in LLMs into three execution paradigms, plus evidence, benchmarks, and challenges.
-
ThinkDial: An Open Recipe for Controlling Reasoning Effort in Large Language Models
An open training recipe that teaches a 32B LLM to reason in three adjustable modes, cutting token use by half or three quarters with small accuracy losses.
-
Think in Blocks: Adaptive Reasoning from Direct Response to Deep Reasoning
A training framework that makes LLMs predict an integer reasoning budget as a number of blocks and adapt reasoning depth to task difficulty.
-
CoRE: Enhancing Metacognition with Label-free Self-evaluation in LRMs
A training-free and label-free detector of cyclic hidden-state patterns triggers early exit during chain-of-thought reasoning, reducing token length while mostly preserving or improving accuracy.
-
Reasoning on a Budget: A Survey of Adaptive and Controllable Test-Time Compute in LLMs
The survey's L1/L2 taxonomy and benchmark show that current reasoning models waste compute on easy problems and underthink hard ones, motivating more adaptive inference.
-
Efficient Reasoning Through Suppression of Self-Affirmation Reflections in Large Reasoning Models
Suppressing low-probability 'wait' tokens that trigger self-affirmation reflections shortens reasoning-model output by 8-18% with roughly unchanged accuracy.
-
Fast on the Easy, Deep on the Hard: Efficient Reasoning via Powered Length Penalty
Reward shaping with a powered length penalty makes LLMs answer easy questions with far fewer tokens while preserving or slightly improving accuracy on hard math benchmarks.
-
RAST: Reasoning Activation in LLMs via Small-model Transfer
Injecting the logit gap between a small RL-trained model and its base into a larger base model at decoding time improves reasoning accuracy on math and code benchmarks, sometimes matching or exceeding RL training of t...
-
PATS: Process-Level Adaptive Thinking Mode Switching
PATS adapts the number of beam-search candidates per reasoning step using process reward model scores, improving accuracy-efficiency tradeoffs on math benchmarks.
-
From Long to Short: LLMs Excel at Trimming Own Reasoning Chains
EDIT searches across step-count prompts to find the shortest reasoning path the model consistently answers correctly.
-
Wait, We Don't Need to "Wait"! Removing Thinking Tokens Improves Reasoning Efficiency
Suppressing "Wait"-like reflection tokens at decode time reduces reasoning token counts by 27-51% across five R1-style model families, with mixed accuracy effects.
-
Towards Concise and Adaptive Thinking in Large Reasoning Models: A Survey
A comprehensive review that categorizes methods for shortening and adaptively triggering chain-of-thought reasoning in large language models.
- Optimizing Length Compression in Large Reasoning Models
Discussion (0). Sign in to comment.