Do LLMs Encode Functional Importance of Reasoning Tokens ? (2026.acl-long)

Copied to clipboard

Challenge: Existing compact reasoning approaches generate long reasoning chains, but they lack a mechanism to encode token-level functional importance for answer generation.
Approach: They propose a procedure that iteratively removes reasoning tokens from models and prunes them to yield length-controlled reasoning chains.
Outcome: The proposed procedure outperforms a frontier model at reasoning lengths and shows that attention scores predict greedy pruning ranks.

Similar Papers

Think Clearly: Improving Reasoning via Redundant Token Pruning (2025.findings-emnlp)

Copied to clipboard

Challenge: Recent large language models show promising capabilities in long-form reasoning . however, they tend to include substantial redundancy in reasoning paths .
Approach: They propose a structure-aware pruning method that prioritizes removing redundant tokens . they remove redundant token and then resume the reasoning generation .
Outcome: The proposed method shows strong performance on reasoning-intensive benchmarks without training.
Your Reasoning Model Knows What Counts: Self-Guided Chain-of-Thought Pruning for Efficient Reasoning (2026.acl-long)

Copied to clipboard

Challenge: Existing approaches to Chain-of-Thought reasoning are often degraded because they disregard the model’s intrinsic reasoning dependency.
Approach: They propose a self-guided pruning framework that leverages the model’s intrinsic likelihood landscape to identify segments that are extraneous to its specific reasoning pattern.
Outcome: The proposed framework reduces output length while maintaining or improving accuracy on multiple benchmarks.
Not All Tokens Matter: Towards Efficient LLM Reasoning via Token Significance in Reinforcement Learning (2026.acl-long)

Copied to clipboard

Challenge: Large language models (LLMs) often produce unnecessarily long explanations that reduce efficiency.
Approach: They propose a length-aware reward that selectively penalizes insignificance tokens . they also propose 'dynamic length control' that encourages more detailed reasoning .
Outcome: The proposed method reduces response length while maintaining correctness, the authors show . it selectively penalizes insignificance tokens while maintaining accuracy .
Pru-CoT: Towards Efficient Reasoning Distillation via Pruning Chain-of-Thought (2026.findings-acl)

Copied to clipboard

Challenge: Existing heuristics fail to capture global causal logic due to rigid rules and limited search spaces.
Approach: They propose a framework that extracts the essential logical structure from reasoning chains.
Outcome: Experiments show that Pru-CoT models generate more compact reasoning paths compared to models trained on verbose data.
Beyond Token Length: Step Pruner for Efficient and Accurate Reasoning in Large Language Models (2026.findings-acl)

Copied to clipboard

Challenge: Existing reinforcement learning methods for large reasoning models suffer from excessive verbosity, known as "overthinking." Existing models penalize generated tokens to promote conciseness, but these methods encounter two challenges: they may develop hacking behavior in later stages of training by discarding reasoning steps.
Approach: They propose a framework that steers large reasoning models toward more efficient reasoning . they prioritize correctness while imposing penalties for redundant steps .
Outcome: The proposed framework reduces token usage by 69.7% on AIME24.
Forget What Matters, Keep the Rest: Selective Unlearning of Informative Tokens (2026.acl-long)

Copied to clipboard

Challenge: Recent studies have explored token-wise loss regularizers that prioritize informative tokens, but rely on ground-truth confidence or external linguistic parsers, which limits their ability to capture contextual information or the model’s overall predictive state.
Approach: They propose an Entropy-guided Token Weighting (ETW) token-level unlearning regularizer that uses entropy of the predictive distribution as a proxy for token informativeness.
Outcome: The proposed token-level unlearning regularizer can achieve more effective unlearning while better preserving model utility than existing token-based approaches.
Verbosity-Aware Rationale Reduction: Sentence-Level Rationale Reduction for Efficient and Effective Reasoning (2025.findings-acl)

Copied to clipboard

Challenge: Recent advances in Large Language Models (LLMs) demonstrate remarkable reasoning capabilities comparable to human cognitive abilities.
Approach: They propose a sentence-level rationale reduction framework that leverages likelihood-based criteria to selectively remove redundant reasoning sentences while preserving reasoning capabilities.
Outcome: The proposed framework improves performance by 7.71% while reducing token generation by 19.87% compared to models trained with complete reasoning paths.
Evolving Sparsity: Leveraging Token Importance Dynamics for Efficient LLM Decoding with Sparse Attention (2026.acl-long)

Copied to clipboard

Challenge: Efficient long-context inference remains a major challenge for large language models (LLMs), as the cost of attention computation during auto-regressive decoding grows linearly with the context length.
Approach: They propose to model token importance as a dynamic process that evolves over decoding steps and propagates through model layers.
Outcome: The proposed method outperforms baseline sparse attention methods and achieves speedups of up to 5.36 for attention latency and 2.33 for end-to-end decoding.
Attention Score is not All You Need for Token Importance Indicator in KV Cache Reduction: Value Also Matters (2024.emnlp-main)

Copied to clipboard

Challenge: Recent studies have focused on scaling the context size of large language models (LLMs) however, the enormous inference costs of LLMs limit their applications.
Approach: They propose a method which uses attention scores and the l 1 norm to evaluate token importance.
Outcome: Extensive experiments on LLaMA2-7B-chat and Vicuna-v1.5-7B show that the proposed method outperforms attention-score-only baselines in over 12 tasks.
Graph-Based Chain-of-Thought Pruning for Reducing Redundant Reflections in Reasoning LLMs (2026.findings-acl)

Copied to clipboard

Challenge: Extending CoT through RL can induce undesirable thinking patterns such as overthinking . prior work has focused on inefficient reflection, which manifests in two problematic patterns: Indiscriminate Reflection and Repetitive Reflectione .
Approach: They propose a graph-based approach to optimize CoT by pruning each linear CoT into a directed acyclic graph with explicit dependency edges.
Outcome: The proposed approach reduces the average reasoning tokens by 42% while maintaining or improving accuracy.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations