mlsanityData to 14 Oct 2026

Stepwise Perplexity-Guided Refinement for Efficient Chain-of-Thought Reasoning in Large Language Models

arXiv 2502.13260 ↗First posted 18 Feb 2025cs.CL cs.AI cs.LGcited by 51

On the map: LLM reasoning → Prompting and chain-of-thought reasoning

Abstract (arXiv)

Chain-of-Thought (CoT) reasoning, which breaks down complex tasks into intermediate reasoning steps, has significantly enhanced the performance of large language models (LLMs) on challenging tasks. However, the detailed reasoning process in CoT often incurs long generation times and high computational costs, partly due to the inclusion of unnecessary steps. To address this, we propose a method to identify critical reasoning steps using perplexity as a measure of their importance: a step is deemed critical if its removal causes a significant increase in perplexity. Our method enables models to focus solely on generating these critical steps. This can be achieved through two approaches: refining demonstration examples in few-shot CoT or fine-tuning the model using selected examples that include only critical steps. Comprehensive experiments validate the effectiveness of our method, which achieves a better balance between the reasoning accuracy and efficiency of CoT.

Key facts. Cited by 51 papers in the map (data through 14 Oct 2026). Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning. Builds on 30 of its 38 references that are in the map. First posted on arXiv 18 Feb 2025.

Builds on

30 of 38 references.

Closest ancestors · 4

Same direction, newest first

  1. 2023Graph of Thoughts: Solving Elaborate Problems with Large Language Models18 Aug
  2. 2023Better Zero-Shot Reasoning with Self-Adaptive Prompting23 May
  3. 2022Large Language Models are Zero-Shot Reasoners24 May
  4. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Models21 Mar

Other references · 26

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2024The Llama 3 Herd of Modelscited by 14,715
  3. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  4. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  5. 2019Fine-Tuning Language Models from Human Preferencescited by 2,150
  6. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,986
  7. 2023LIMA: Less Is More for Alignmentcited by 1,100
  8. 2024Large Language Monkeys: Scaling Inference Compute with Repeated Samplingcited by 991
  9. 2024MathVerse: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?cited by 586
  10. 2024ORPO: Monolithic Preference Optimization without Reference Modelcited by 562
  11. 2023Detecting Language Model Attacks with Perplexitycited by 431
  12. 2019Analysing Mathematical Reasoning Abilities of Neural Modelscited by 335
14 more
  1. 2024MUSE: Machine Unlearning Six-Way Evaluation for Language Modelscited by 261
  2. 2023How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Compositioncited by 223
  3. 2023Fine-Tuning or Retrieval? Comparing Knowledge Injection in LLMscited by 178
  4. 2023Deductive Verification of Chain-of-Thought Reasoningcited by 124
  5. 2024Forest-of-Thought: Scaling Test-Time Compute for Enhancing LLM Reasoningcited by 110
  6. 2024Perplexed by Perplexity: Perplexity-Based Data Pruning With Small Reference Modelscited by 84
  7. 2024Archon: An Architecture Search Framework for Inference-Time Techniquescited by 60
  8. 2023Improving Large Language Model Fine-tuning for Solving Math Problemscited by 49
  9. 2024Binary Classifier Optimization for Large Language Model Alignmentcited by 45
  10. 2024Rethinking Data Selection for Supervised Fine-Tuningcited by 35
  11. 2024Few shot chain-of-thought driven reasoning to prompt LLMs for open ended medical question answeringcited by 31
  12. 2020Misinformation Has High Perplexitycited by 15
  13. 2024Perplexed: Understanding When Large Language Models are Confusedcited by 9
  14. 2024Make LLMs better zero-shot reasoners: Structure-orientated autonomous reasoningcited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 32; 2026: 19.

51 papers.

Sort:

By area:

  1. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Modelscited by 464
  2. 2025Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Modelscited by 384
  3. 2025Dynamic Early Exit in Reasoning Modelscited by 219
  4. 2025Reasoning Models Can Be Effective Without Thinkingcited by 187
  5. 2025Efficient Reasoning Models: A Surveycited by 95
  6. 2025S-GRPO: Early Exit via Reinforcement Learning in Reasoning Modelscited by 87
  7. 2025Efficient Inference for Large Reasoning Models: A Surveycited by 61
  8. 2025R-KV: Redundancy-aware KV Cache Compression for Reasoning Modelscited by 54
  9. 2025AdaCtrl: Towards Adaptive and Controllable Reasoning via Difficulty-Aware Budgetingcited by 50
  10. 2025Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Modelscited by 37
  11. 2025Harnessing the Reasoning Economy: A Survey of Efficient Reasoning for Large Language Modelscited by 36
  12. 2025ConCISE: Confidence-guided Compression in Step-by-step Efficient Reasoningcited by 33
38 more
  1. 2025Pruning the Unsurprising: Efficient LLM Reasoning via First-Token Surprisalcited by 32
  2. 2025FaithCoT-Bench: Benchmarking Instance-Level Faithfulness of Chain-of-Thought Reasoningcited by 26
  3. 2026Does Your Reasoning Model Implicitly Know When to Stop Thinking?cited by 19
  4. 2025Causal Sufficiency and Necessity Improves Chain-of-Thought Reasoningcited by 17
  5. 2026GlimpRouter: Efficient Collaborative Inference by Glimpsing One Token of Thoughtscited by 17
  6. 2025Not All Thoughts are Generated Equal: Efficient LLM Reasoning via Multi-Turn Reinforcement Learningcited by 16
  7. 2025DRPO: Efficient Reasoning via Decoupled Reward Policy Optimizationcited by 16
  8. 2025Stop When Enough: Adaptive Early-Stopping for Chain-of-Thought Reasoningcited by 16
  9. 2025Generative AI Act II: Test Time Scaling Drives Cognition Engineeringcited by 15
  10. 2025LIMOPro: Reasoning Refinement for Efficient and Effective Test-time Scalingcited by 15
  11. 2025Adaptive GoGI-Skip: Coupling Goal-Gradient Importance with Dynamic Uncertainty for Efficient Reasoningcited by 12
  12. 2025AutoL2S: Auto Long-Short Reasoning for Efficient Large Language Modelscited by 12
  13. 2025Exploring the System 1 Thinking Capability of Large Reasoning Modelscited by 10
  14. 2025Understanding and Steering the Cognitive Behaviors of Reasoning Models at Test-Timecited by 10
  15. 2025The Overthinker's DIET: Cutting Token Calories with DIfficulty-AwarE Trainingcited by 9
  16. 2025Probing the Difficulty Perception Mechanism of Large Language Modelscited by 8
  17. 2025BadThink: Triggered Overthinking Attacks on Chain-of-Thought Reasoning in Large Language Modelscited by 8
  18. 2026Neural Chain-of-Thought Search: Searching the Optimal Reasoning Path to Enhance Large Language Modelscited by 8
  19. 2025CoRE: Enhancing Metacognition with Label-free Self-evaluation in LRMscited by 5
  20. 2026Do LLMs Encode Functional Importance of Reasoning Tokens?cited by 5
  21. 2025GateLens: A Reasoning-Enhanced LLM Agent for Automotive Software Release Analyticscited by 4
  22. 2026Compress the Easy, Explore the Hard: Difficulty-Aware Entropy Regularization for Efficient LLM Reasoningcited by 4
  23. 2026Detecting Unfaithful Chain-of-Thought via Circuit-Guided Internal-External Discrepancycited by 3
  24. 2025Benchmarking Chinese Medical LLMs: A Medbench-based Analysis of Performance Gaps and Hierarchical Optimization Strategiescited by 2
  25. 2025From Long to Lean: Performance-aware and Adaptive Chain-of-Thought Compression via Multi-round Refinementcited by 2
  26. 2026ConMax: Confidence-Maximizing Compression for Efficient Chain-of-Thought Reasoningcited by 2
  27. 2025Can Reasoning Path still be Effective as Input? Bridging Post-Reasoning to Chain-of-Thought Compressioncited by 1
  28. 2026Segment-Level Attribution for Selective Learning of Long Reasoning Tracescited by 1
  29. 2026Short Chains, Deep Thoughts: Balancing Reasoning Efficiency and Intra-Segment Capability via Split-Merge Optimizationcited by 1
  30. 2026rePIRL: Learn PRM with Inverse RL for LLM Reasoningcited by 1
  31. 2026Reforming the Mechanism: Editing Reasoning Patterns in LLMs with Circuit Reshapingcited by 1
  32. 2026Reasoning as Compression: Unifying Budget Forcing via the Conditional Information Bottleneckcited by 1
  33. 2026Step-GRPO: Internalizing Dynamic Early Exit for Efficient Reasoningcited by 1
  34. 2026CRISP: Compressing Redundancy in Chain-of-Thought via Intrinsic Saliency Pruningcited by 1
  35. 2026ChainPrune: Evaluating and Reducing Redundancy in Long Chain-of-Thought Reasoningcited by 1
  36. 2026Bridging Efficiency and Transparency: Explainable CoT Compression in Multimodal Large Reasoning Modelscited by 0
  37. 2026WebClipper: Efficient Evolution of Web Agents with Graph-based Trajectory Pruningcited by 0
  38. 2026Thinking Economically: A Hierarchical Framework for Adaptive-Complexity Reasoning in LLMscited by 0

50 most cited of 51 All 51 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024Fine-Tuning on Diverse Reasoning Chains Drives Within-Inference CoT Refinement in LLMs
  2. 2026Fragile Thoughts: How Large Language Models Handle Chain-of-Thought Perturbations
  3. 2025Reasoning Beyond Language: A Comprehensive Survey on Latent Chain-of-Thought Reasoning
  4. 2025Early Stopping Chain-of-thoughts in Large Language Models
  5. 2025Fractured Chain-of-Thought Reasoning
  6. 2024A Theoretical Understanding of Chain-of-Thought: Coherent Reasoning and Error-Aware Demonstration
  7. 2024A Hopfieldian View-based Interpretation for Chain-of-Thought Reasoning
  8. 2026OS-Pruner: Pruning Chains-of-Thought of Reasoning Models via Optimal Stopping