Skip to content
Projekte

yueliu1999/Awesome-Efficient-Inference-for-LRMs

Auf GitHub ansehen

yueliu1999/Awesome-Efficient-Inference-for-LRMs has 244 GitHub stars and 20 forks. Star changes are not measured yet. Last checked 2026-11-15 20:05 UTC. Last push 2026-10-26.

So werden diese Zahlen gemessen

Gemessener Sternzuwachs

in 24h
Aktualisierung verzögert
in 7 days
Aktualisierung verzögert
in 15 days
Aktualisierung verzögert
in 1 month
Aktualisierung verzögert

Nettoänderungen in überlappenden Zeiträumen, keine Tagessummen. Fehlende Messwerte bleiben leer. Sterne zeigen Interesse, nicht Qualität.

Über uns

[IEEE T-PAMI] Awesome-Efficient-Inference-for-LRMs is a collection of state-of-the-art, novel, exciting, token-efficient methods for Large Reasoning Models (LRMs). It contains papers, codes, datasets, evaluations, and analyses. https://arxiv.org/pdf/2503.23077

Lernen

README.md

Original

Awesome-Efficient-Inference-for-LRMs

Awesome-Efficient-Inference-for-LRMs is a collection of state-of-the-art, novel, exciting, token-efficient methods for Large Reasoning Models (LRMs). It contains papers, codes, datasets, evaluations, and analyses. Any additional things regarding efficient inference for LRMs, PRs, and issues are welcome, and we are glad to add you to the contributor list here. Any problems, please contact [email protected]. Our survey paper is online: Efficient Inference for Large Reasoning Models: A Survey. If you find this repository useful to your research or work, it is really appreciated if to star this repository and cite our papers here. :sparkles:

overview

Update

  • (2026/06/07) Added 40+ new papers (2025.05–2026.06) across Explicit Compact CoT, Implicit Latent CoT, Limitations & Challenges, Further Improvement, and Survey.
  • (2026/05/31) Our survey paper has been accepted by the IEEE T-PAMI 2026.
  • (2025/03/29) Our survey paper is online: Efficient Inference for Large Reasoning Models: A Survey.

Reference

If you find this repository helpful for your research, we would greatly appreciate it if you could cite our papers. :sparkles:

@article{liu2025efficient,
  title={Efficient Inference for Large Reasoning Models: A Survey},
  author={Liu, Yue and Wu, Jiaying and He, Yufei and Gao, Hongcheng and Chen, Hongyu and Bi, Baolong and Zhang, Jiaheng and Huang, Zhiqi and Hooi, Bryan},
  journal={arXiv preprint arXiv:2503.23077},
  year={2025}
}

Bookmarks

Papers

Survey

Time Title Venue Paper Code
2026.05 Efficient Inference for Large Reasoning Models: A Survey IEEE T-PAMI'26 link link
2025.08 Don't Overthink It: A Survey of Efficient R1-style Large Reasoning Models arXiv'25 link -
2025.07 Towards Concise and Adaptive Thinking in Large Reasoning Models: A Survey arXiv'25 link -
2025.07 Reasoning on a Budget: A Survey of Adaptive and Controllable Test-Time Compute in LLMs arXiv'25 link -
2025.04 Reasoning Beyond Language: A Comprehensive Survey on Latent Chain-of-Thought Reasoning arXiv'25 link link
2025.04 Efficient Reasoning Models: A Survey arXiv'25 link link
2025.03 Efficient Inference for Large Reasoning Models: A Survey arXiv'25 link link
2025.03 Harnessing the Reasoning Economy: A Survey of Efficient Reasoning for Large Language Models arXiv'25 link link
2025.03 A Survey of Efficient Reasoning for Large Reasoning Models: Language, Multimodality, and Beyond arXiv'25 link link
2025.03 Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models arXiv'25 link link

Explicit Compact CoT

Time Title Venue Paper Code
2026.10 DoubtLess: Training-Free Contrastive Decoding for Efficient Reasoning arXiv'26 link link
2026.05 Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models arXiv'26 link link
2026.03 Efficient Reasoning with Balanced Thinking (ReBalance) arXiv'26 link -
2026.03 Draft-Thinking: Learning Efficient Reasoning in Long Chain-of-Thought LLMs arXiv'26 link -
2026.02 Towards Efficient Large Language Reasoning Models via Extreme-Ratio Chain-of-Thought Compression arXiv'26 link -
2026.02 Self-Verification Dilemma: Experience-Driven Suppression of Overused Checking in LLM Reasoning arXiv'26 link -
2026.02 Short Chains, Deep Thoughts: Balancing Reasoning Efficiency and Intra-Segment Capability via Split-Merge Optimization arXiv'26 link -
2026.01 Mitigating Overthinking in Large Reasoning Models via Difficulty-aware Reinforcement Learning arXiv'26 link -
2025.10 DLER: Doing Length pEnalty Right - Incentivizing More Intelligence per Token via Reinforcement Learning arXiv'25 link -
2025.10 Learning to Reason Efficiently with Discounted Reinforcement Learning arXiv'25 link -
2025.10 Beyond Token Length: Step Pruner for Efficient and Accurate Reasoning in Large Language Models arXiv'25 link -
2025.09 Your Models Have Thought Enough: Training Large Reasoning Models to Stop Overthinking arXiv'25 link -
2025.08 Making Slow Thinking Faster: Compressing LLM Chain-of-Thought via Step Entropy arXiv'25 link link
2025.08 BudgetThinker: Empowering Budget-aware LLM Reasoning with Control Tokens arXiv'25 link -
2025.08 Efficient Reasoning for Large Reasoning Language Models via Certainty-Guided Reflection Suppression arXiv'25 link -
2025.07 Think Clearly: Improving Reasoning via Redundant Token Pruning arXiv'25 link -
2025.06 Structured Reasoning for LLMs: A Unified Framework for Efficiency and Explainability ICLR'26 link link
2025.06 AALC: Large Language Model Efficient Reasoning via Adaptive Accuracy-Length Control arXiv'25 link -
2025.06 Steering LLM Thinking with Budget Guidance arXiv'25 link link
2025.06 Exploring and Exploiting the Inherent Efficiency within Large Reasoning Models for Self-Guided Efficiency Enhancement arXiv'25 link -
2025.05 REA-RL: Reflection-Aware Online Reinforcement Learning for Efficient Reasoning arXiv'25 link link
2025.05 Can Pruning Improve Reasoning? Revisiting Long-CoT Compression with Capability in Mind for Better Reasoning arXiv'25 link -
2025.05 Long-Short Chain-of-Thought Mixture Supervised Fine-Tuning Eliciting Efficient Reasoning in Large Language Models arXiv'25 link link
2025.05 ConCISE: Confidence-guided Compression in Step-by-step Efficient Reasoning arXiv'25 link -
2025.05 Scalable Chain of Thoughts via Elastic Reasoning arXiv'25 link -
2025.05 S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models arXiv'25 link -
2025.05 Making Small Language Models Efficient Reasoners: Intervention, Supervision, Reinforcement arXiv'25 link -
2025.05 Accelerating Chain-of-Thought Reasoning: When Goal-Gradient Importance Meets Dynamic Skipping arXiv'25 link -
2025.05 SelfBudgeter: Adaptive Token Allocation for Efficient LLM Reasoning arXiv'25 link -
2025.05 Not All Thoughts are Generated Equal: Efficient LLM Reasoning via Multi-Turn Reinforcement Learning arXiv'25 link link
2025.05 Fractured Chain-of-Thought Reasoning arXiv'25 link -
2025.05 Efficient RL Training for Reasoning Models via Length-Aware Optimization arXiv'25 link -
2025.05 DRP: Distilled Reasoning Pruning with Skill-aware Step Decomposition for Efficient Large Reasoning Models arXiv'25 link -
2025.05 FlashThink: An Early Exit Method For Efficient Reasoning arXiv'25 link -
2025.05 Optimizing Anytime Reasoning via Budget Relative Policy Optimization arXiv'25 link link
2025.05 VeriThinker: Learning to Verify Makes Reasoning Model Efficient arXiv'25 link link
2025.05 Reasoning Path Compression: Compressing Generation Trajectories for Efficient LLM Reasoning arXiv'25 link link
2025.05 ThinkLess: A Training-Free Inference-Efficient Method for Reducing Reasoning Redundancy arXiv'25 link -
2025.05 Learn to Reason Efficiently with Adaptive Length-based Reward Shaping arXiv'25 link link
2025.05 R1-Compress: Long Chain-of-Thought Compression via Chunk Compression and Search arXiv'25 link link
2025.05 Incentivizing Dual Process Thinking for Efficient Large Language Model Reasoning arXiv'25 link -
2025.05 ConciseRL: Conciseness-Guided Reinforcement Learning for Efficient Reasoning Models arXiv'25 link link
2025.05 TrimR: Verifier-based Training-Free Thinking Compression for Efficient Test-Time Scaling arXiv'25 link -
2025.05 Not All Tokens Are What You Need In Thinking arXiv'25 link link
2025.05 LIMOPro: Reasoning Refinement for Efficient and Effective Test-time Scaling arXiv'25 link link
2025.05 Walk Before You Run! Concise LLM Reasoning via Reinforcement Learning arXiv'25 link link
2025.05 CoThink: Token-Efficient Reasoning via Instruct Models Guiding Reasoning Models arXiv'25 link -
2025.05 Don't Think Longer, Think Wisely: Optimizing Thinking Dynamics for Large Reasoning Models arXiv'25 link -
2025.05 A*-Thought: Efficient Reasoning via Bidirectional Compression for Low-Resource Settings arXiv'25 link link
2025.05 Efficient Reasoning via Chain of Unconscious Thought arXiv'25 link link
2025.04 Syzygy of Thoughts: Improving LLM CoT with the Minimal Free Resolution arXiv'25 link link
2025.03 Sketch-of-thought: Efficient llm reasoning with adaptive cognitive-inspired sketching (SoT) arXiv'25 link link
2025.03 SOLAR: Scalable Optimization of Large-scale Architecture for Reasoning arXiv'25 link -
2025.03 InftyThink: Breaking the Length Limits of Long-Context Reasoning in Large Language Models arXiv'25 link -
2025.03 L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning arXiv'25 link link
2025.03 Optimizing Test-Time Compute via Meta Reinforcement Fine-Tuning arXiv'25 link link
2025.02 Chain of Draft: Thinking Faster by Writing Less arXiv'25 link link
2025.02 Meta-Reasoner: Dynamic Guidance for Optimized Inference-time Reasoning in Large Language Models arXiv'25 link -
2025.02 TokenSkip: Controllable Chain-of-Thought Compression in LLMs arXiv'25 link link
2025.02 LightThinker: Thinking Step-by-Step Compression arXiv'25 link link
2025.02 CoT-Valve: Length-Compressible Chain-of-Thought Tuning arXiv'25 link -
2025.02 Self-Training Elicits Concise Reasoning in Large Language Models arXiv'25 link link
2025.02 DAST: Context-Aware Compression in LLMs via Dynamic Allocation of Soft Tokens arXiv'25 link -
2025.02 Training Language Models to Reason Efficiently arXiv'25 link link
2025.02 Anthropic. Claude 3.7 sonnet and claude code Anthropic'25 link -
2025.02 Stepwise Perplexity-Guided Refinement for Efficient Chain-of-Thought Reasoning in Large Language Models arXiv'25 link -
2025.01 Kimi k1.5: Scaling Reinforcement Learning with LLMs arXiv'25 link -
2025.01 O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning arXiv'25 link link
2025.01 Think Smarter not Harder: Adaptive Reasoning with Inference Aware Optimization arXiv'25 link -
2024.12 C3oT: Generating Shorter Chain-of-Thought without Compromising Effectiveness arXiv'24 link -
2024.12 Token-Budget-Aware LLM Reasoning arXiv'24 link link
2024.12 Verbosity-Aware Rationale Reduction: Effective Reduction of Redundant Rationale via Principled Criteria arXiv'24 link -
2024.11 Can Language Models Learn to Skip Steps? arXiv'24 link link
2024.07 Concise Thoughts: Impact of Output Length on LLM Reasoning and Cost arXiv'24 link -
2024.07 Distilling System 2 into System 1 arXiv'24 link -
takeaway1

Implicit Latent CoT

Time Title Venue Paper Code
2026.05 Selective Latent Thinking: Adaptive Compression of LLM Reasoning Chains arXiv'26 link link
2026.05 LatentRAG: Latent Reasoning and Retrieval for Efficient Agentic RAG arXiv'26 link -
2026.04 Thinking Without Words: Efficient Latent Reasoning with Abstract Chain-of-Thought arXiv'26 link -
2026.02 LoopFormer: Elastic-Depth Looped Transformers for Latent Reasoning via Shortcut Modulation arXiv'26 link -
2025.10 KaVa: Latent Reasoning via Compressed KV-Cache Distillation arXiv'25 link -
2025.05 Soft Thinking: Unlocking the Reasoning Potential of LLMs in Continuous Concept Space arXiv'25 link link
2025.05 Think Silently, Think Fast: Dynamic Latent Compression of LLM Reasoning Chains arXiv'25 link link
2025.02 CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillation arXiv'25 link -
2025.02 Token Assorted: Mixing Latent and Text Tokens for Improved Language Model Reasoning arXiv'25 link -
2025.02 SoftCoT: Soft Chain-of-Thought for Efficient Reasoning with LLMs arXiv'25 link -
2025.01 Efficient Reasoning with Hidden Thinking arXiv'25 link link
2024.12 Training Large Language Models to Reason in a Continuous Latent Space arXiv'24 link -
2024.12 Compressed Chain of Thought: Efficient Reasoning Through Dense Representations arXiv'24 link -
2024.05 From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step arXiv'24 link link
2023.11 Implicit Chain of Thought Reasoning via Knowledge Distillation arXiv'23 link link
takeaway2

Limitations and Challenges

User-centric Controllable Reasoning

Time Title Venue Paper Code
2026.02 Conformal Thinking: Risk Control for Reasoning on a Compute Budget arXiv'26 link -
2025.10 e1: Learning Adaptive Control of Reasoning Effort arXiv'25 link -
2025.05 When to Continue Thinking: Adaptive Thinking Mode Switching for Efficient Reasoning (ASRR) arXiv'25 link -
2025.02 OpenAI o3-mini System Card OpenAI'25 link -
2025.02 Anthropic. Claude 3.7 sonnet and claude code Anthropic'25 link -

Interpretability of Reasoning

Time Title Venue Paper Code
2026.06 Unlocking the Black Box of Latent Reasoning: An Interpretability-Guided Approach to Intervention arXiv'26 link -
2026.04 Are Latent Reasoning Models Easily Interpretable? arXiv'26 link -
2026.04 LLM Reasoning Is Latent, Not the Chain of Thought arXiv'26 link -
2024.05 FiDeLiS: Faithful Reasoning in Large Language Model for Knowledge Graph Question Answering arXiv'24 link -
2024.02 Challenges and barriers of using large language models (LLM) such as ChatGPT for diagnostic medicine with a focus on digital pathology - a recent scoping review Diagnostic pathology'24 link -
2024.02 (A)I Am Not a Lawyer, But...: Engaging Legal Experts towards Responsible LLM Policies for Legal Advice arXiv'24 link -
2023.12 Retrieval-Augmented Generation for Large Language Models: A Survey arXiv'23 link link
2023.10 Contribution and performance of ChatGPT and other Large Language Models (LLM) for scientific and research advancements: a double-edged sword International Research Journal of Modernization in Engineering Technology and Science'23 link -
2023.08 Reasoning in Large Language Models Through Symbolic Math Word Problems arXiv'23 link -
2021.12 Chapter 1. Neural-Symbolic Learning and Reasoning: A Survey and Interpretation1 Neuro-Symbolic Artificial Intelligence'21 link -

Reasoning Safety

Time Title Venue Paper Code
2025.05 The First Impression Problem: Internal Bias Triggers Overthinking in Reasoning Models arXiv'25 link -
2025.03 Optimizing Test-Time Compute via Meta Reinforcement Fine-Tuning arXiv'25 link link
2025.03 Detecting misbehavior in frontier reasoning models OpenAI'25 link -
2025.02 Evaluating the Paperclip Maximizer: Are RL-Based Language Models More Likely to Pursue Instrumental Goals? arXiv'25 link link
2025.01 O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning arXiv'25 link link
2025.01 GuardReasoner: Towards Reasoning-based LLM Safeguards arXiv'25 link link
2025.01 Kimi k1.5: Scaling Reinforcement Learning with LLMs arXiv'25 link -
2024.12 C3oT: Generating Shorter Chain-of-Thought without Compromising Effectiveness arXiv'24 link -
2024.10 FlipAttack: Jailbreak LLMs via Flipping arXiv'24 link link
2023.10 Privacy in Large Language Models: Attacks, Defenses and Future Directions arXiv'23 link -

Broader Application

Time Title Venue Paper Code
2025.03 SOLAR: Scalable Optimization of Large-scale Architecture for Reasoning arXiv'25 link -
2025.03 Large language models (LLM) in computational social science: prospects, current state, and challenges Social Network Analysis and Mining'25 link -
2025.03 Gemini robotics brings ai into the physical world Google'25 link -
2025.03 Nvidia isaac gr00t n1: An open foundation model for humanoid robots. Nvidia'25 link -
2025.02 Anthropic. Claude 3.7 sonnet and claude code Anthropic'25 link -
2025.02 TokenSkip: Controllable Chain-of-Thought Compression in LLMs arXiv'25 link link
2025.02 RoboBrain: A Unified Brain Model for Robotic Manipulation from Abstract to Concrete arXiv'25 link -
2025.02 From Personas to Talks: Revisiting the Impact of Personas on LLM-Synthesized Emotional Support Conversations arXiv'25 link -
2025.02 Introducing deep research OpenAI'25 link -
2025.02 Introducing gpt-4.5 OpenAI'25 link -
2025.01 Kimi k1.5: Scaling Reinforcement Learning with LLMs arXiv'25 link -
2024.12 Compressed Chain of Thought: Efficient Reasoning Through Dense Representations arXiv'24 link -
2024.08 Large Language Model Agent in Financial Trading: A Survey arXiv'24 link -
2024.04 Automated Social Science: Language Models as Scientist and Subjects arXiv'24 link link
2023.11 LLM4Drive: A Survey of Large Language Models for Autonomous Driving arXiv'24 link link
takeaway3

Further Improvement

New Architecture

Time Title Venue Paper Code
2026.06 Genesis 2: Cascade MoE for Ultra-Efficient CPU Inference GitHub'26 - link
2025.10 ThinKV: Thought-Adaptive KV Cache Compression for Efficient Reasoning Models ICLR'26 link -
2025.05 R-KV: Redundancy-aware KV Cache Compression for Reasoning Models arXiv'25 link -
2025.02 Large Language Diffusion Models arXiv'25 link link
2025.02 UniGraph2: Learning a Unified Embedding Space to Bind Multimodal Graphs arXiv'25 link link
2024.03 Graph of Thoughts: Solving Elaborate Problems with Large Language Models AAAI Conference on Artificial Intelligence'24 link link
2024.02 UniGraph: Learning a Unified Cross-Domain Foundation Model for Text-Attributed Graphs arXiv'24 link link
2023.12 Mamba: Linear-Time Sequence Modeling with Selective State Spaces arXiv'23 link link
2023.10 Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models arXiv'23 link link
2023.05 RWKV: Reinventing RNNs for the Transformer Era arXiv'23 link link
2017.10 Mastering the game of Go without human knowledge nature'17 link -

Model Merge

Time Title Venue Paper Code
2026.04 Multi-objective Evolutionary Merging Enables Efficient Reasoning Models arXiv'26 link -
2025.09 The Thinking Spectrum: An Empirical Study of Tunable Reasoning in LLMs through Model Merging arXiv'25 link -
2025.06 Accelerated Test-Time Scaling with Model-Free Speculative Sampling arXiv'25 link -
2025.03 Unlocking Efficient Long-to-Short LLM Reasoning with Model Merging arXiv'25 link link
2025.03 SOLAR: Scalable Optimization of Large-scale Architecture for Reasoning arXiv'25 link -
2025.03 Optimizing Test-Time Compute via Meta Reinforcement Fine-Tuning arXiv'25 link link
2025.02 Meta-Reasoner: Dynamic Guidance for Optimized Inference-time Reasoning in Large Language Models arXiv'25 link -
2025.01 Kimi k1.5: Scaling Reinforcement Learning with LLMs arXiv'25 link -
2025.01 DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning arXiv'25 link -
2024.12 Token-Budget-Aware LLM Reasoning arXiv'24 link link
2024.08 Model Merging in LLMs, MLLMs, and Beyond: Methods, Theories, Applications and Opportunities arXiv'24 link link
2024.07 The Llama 3 Herd of Models arXiv'24 link link

Agent Router

Time Title Venue Paper Code
2026.05 Breaking the Reward Barrier: Accelerating Tree-of-Thought Reasoning via Speculative Exploration (SPEX) arXiv'26 link -
2026.04 Step-GRPO: Internalizing Dynamic Early Exit for Efficient Reasoning arXiv'26 link -
2026.04 Early Stopping for Large Reasoning Models via Confidence Dynamics arXiv'26 link -
2026.03 Ares: Adaptive Reasoning Effort Selection for Efficient LLM Agents arXiv'26 link -
2025.09 FastTTS: Accelerating Test-Time Scaling for Edge LLM Reasoning ASPLOS'26 link -
2025.06 SPECS: Faster Test-Time Scaling through Speculative Drafts arXiv'25 link -
2025.05 R2R: Efficiently Navigating Divergent Reasoning Paths with Small-Large Model Token Routing arXiv'25 link link
2025.04 SpecReason: Fast and Accurate Inference-Time Compute via Speculative Reasoning arXiv'25 link link
2025.02 Confident or Seek Stronger: Exploring Uncertainty-Based On-device LLM Routing From Benchmarking to Generalization arXiv'25 link -
2025.01 RouteLLM: Learning to Route LLMs from Preference Data The Thirteenth International Conference on Learning Representations'25 link -
2024.10 Learning to Route LLMs with Confidence Tokens arXiv'24 link -
2024.09 RouterDC: Query-Based Router by Dual Contrastive Learning for Assembling Large Language Models The Thirty-Ninth Annual Conference on Neural Information Processing Systems link
takeaway4

Contributors

yueliu1999 jiayingwu19 yf-he bhooi Rohan-GRH

(back to top)