mlsanityData to 7 Oct 2026

Fuel Gauge: Estimating Chain-of-Thought Length Ahead of Time in Large Multimodal Models

arXiv 2603.10335 ↗First posted 11 Mar 2026cs.CVcited by 0

On the map: LLM reasoning → Multimodal reasoning

also in 1 more
Abstract (arXiv)

Reasoning Large Multi-modality Models (LMMs) have become the de facto choice for many applications. However, these models rely on a Chain-of-Thought (CoT) process that is lengthy and unpredictable at runtime, often resulting in inefficient use of computational resources (due to memory fragmentation) and sub-optimal accuracy (due to under- and over-thinking). We observe empirically that the CoT process follows a very simple form, whose behavior is independent of the specific generated samples. This suggests that the CoT length can be estimated ahead of time based on a hidden parameter representing the amount of "fuel" available to support the reasoning process. Based on this insight, we propose Fuel Gauge, the first method which extracts this hidden signal and predicts CoT length ahead of time. We demonstrate the utility on the Fuel Gauge on two downstream tasks: predictive KV cache allocation, which addresses memory fragmentation in LMM serving systems, and CoT length modulation, which mitigates under-thinking and over-thinking. Extensive experiments on LMMs across text-only, image-text, and video-text question answering benchmarks demonstrate the effectiveness, generalizability, and practical value of our Fuel Gauge. For example, on the GPQA-Diamond benchmark, our Fuel Gauge achieves less than half the CoT length prediction error compared to the baseline; this translates into a 13.37x reduction in the memory allocation frequency.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Multimodal reasoning, one of 17 directions in LLM reasoning. Builds on 24 of its 49 references that are in the map. First posted on arXiv 11 Mar 2026.

Builds on

24 of 49 references.

References · 24

  1. 2017Proximal Policy Optimization Algorithmscited by 14,441
  2. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  3. 2025Qwen3 Technical Reportcited by 8,195
  4. 2020Measuring Massive Multitask Language Understandingcited by 7,768
  5. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  6. 2016SGDR: Stochastic Gradient Descent with Warm Restartscited by 4,993
  7. 2024GPT-4o System Cardcited by 4,467
  8. 2017Decoupled Weight Decay Regularizationcited by 1,722
  9. 2025gpt-oss-120b & gpt-oss-20b Model Cardcited by 1,258
  10. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMscited by 564
  11. 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 488
  12. 2025GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Modelscited by 446
12 more
  1. 2023A Long Way to Go: Investigating Length Correlations in RLHFcited by 285
  2. 2022Extracting Latent Steering Vectors from Pretrained Language Modelscited by 217
  3. 2025When More is Less: Understanding Chain-of-Thought Length in LLMscited by 182
  4. 2025Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMscited by 162
  5. 2025SuperGPQA: Scaling LLM Evaluation across 285 Graduate Disciplinescited by 67
  6. 2025Reinforcement Learning from Human Feedbackcited by 53
  7. 2025Intern-S1: A Scientific Multimodal Foundation Modelcited by 49
  8. 2025LLM Unlearning via Neural Activation Redirectioncited by 38
  9. 2025OptimalThinkingBench: Evaluating Over and Underthinking in LLMscited by 35
  10. 2024On Effects of Steering Latent Representation for Large Language Model Unlearningcited by 27
  11. 2025Revisiting Overthinking in Long Chain-of-Thought from the Perspective of Self-Doubtcited by 17
  12. 2024vTensor: Flexible Virtual Tensor Management for Efficient LLM Servingcited by 13

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025MME-CoT: Benchmarking Chain-of-Thought in Large Multimodal Models for Reasoning Quality, Robustness, and Efficiency
  2. 2024KAM-CoT: Knowledge Augmented Multimodal Chain-of-Thoughts Reasoning
  3. 2023Multimodal Chain-of-Thought Reasoning in Language Models
  4. 2026Attention-guided Fine-tuning of Multimodal Large Language Models Improves Chain-of-Thought Reasoning
  5. 2026Bridging Efficiency and Transparency: Explainable CoT Compression in Multimodal Large Reasoning Models
  6. 2026Crystal-KV: Efficient KV Cache Management for Chain-of-Thought LLMs via Answer-First Principle
  7. 2026Zipping the Thought: When and How Compressed Reasoning Data Works in LLM Post-Training
  8. 2023Enhancing Human-like Multi-Modal Reasoning: A New Challenging Dataset and Comprehensive Framework