Skip to content
← Back to explorer

Accordion-Thinking: Self-Regulated Step Summaries for Efficient and Readable LLM Reasoning

Zhicheng Yang, Zhijiang Guo, Yinya Huang, Yongxin Wang, Wenlei Shi, Yiwei Wang, Xiaodan Liang, Jing Tang · Feb 3, 2026 · Citations: 0

Data freshness

Extraction: Recent

Check recency before relying on this page for active eval decisions. Use stale pages as context and verify against current hub results.

Metadata refreshed

Apr 9, 2026, 2:14 PM

Recent

Extraction refreshed

Apr 9, 2026, 2:14 PM

Recent

Extraction source

Persisted extraction

Confidence unavailable

Abstract

Scaling test-time compute via long Chain-of-Thought unlocks remarkable gains in reasoning capabilities, yet it faces practical limits due to the linear growth of KV cache and quadratic attention complexity. In this paper, we introduce Accordion-Thinking, an end-to-end framework where LLMs learn to self-regulate the granularity of the reasoning steps through dynamic summarization. This mechanism enables a Fold inference mode, where the model periodically summarizes its thought process and discards former thoughts to reduce dependency on historical tokens. We apply reinforcement learning to incentivize this capability further, uncovering a critical insight: the accuracy gap between the highly efficient Fold mode and the exhaustive Unfold mode progressively narrows and eventually vanishes over the course of training. This phenomenon demonstrates that the model learns to encode essential reasoning information into compact summaries, achieving effective compression of the reasoning context. Our Accordion-Thinking demonstrates that with learned self-compression, LLMs can tackle complex reasoning tasks with minimal dependency token overhead without compromising solution quality, and it achieves a three times throughput while maintaining accuracy on a 48GB GPU memory configuration, while the structured step summaries provide a human-readable account of the reasoning process.

Low-signal caution for protocol decisions

Use this page for context, then validate protocol choices against stronger HFEPX references before implementation decisions.

  • Structured extraction is still processing; current fields are metadata-first.

HFEPX Relevance Assessment

Signal extraction is still processing. This page currently shows metadata-first guidance until structured protocol fields are ready.

Best use

Background context only

Use if you need

A provisional background reference while structured extraction finishes.

Main weakness

Structured extraction is still processing; current fields are metadata-first.

Trust level

Provisional

Eval-Fit Score

Unavailable

Eval-fit score is unavailable until extraction completes.

Human Feedback Signal

Not explicit in abstract metadata

Evaluation Signal

Weak / implicit signal

HFEPX Fit

Provisional (processing)

Extraction confidence: Provisional

Field Provenance & Confidence

Each key protocol field shows extraction state, confidence band, and data source so you can decide whether to trust it directly or validate from full text.

Human Feedback Types

provisional

None explicit

Confidence: Provisional Source: Persisted extraction inferred

No explicit feedback protocol extracted.

Evidence snippet: Scaling test-time compute via long Chain-of-Thought unlocks remarkable gains in reasoning capabilities, yet it faces practical limits due to the linear growth of KV cache and quadratic attention complexity.

Evaluation Modes

provisional

Automatic metrics

Confidence: Provisional Source: Persisted extraction inferred

Includes extracted eval setup.

Evidence snippet: Scaling test-time compute via long Chain-of-Thought unlocks remarkable gains in reasoning capabilities, yet it faces practical limits due to the linear growth of KV cache and quadratic attention complexity.

Quality Controls

provisional

Not reported

Confidence: Provisional Source: Persisted extraction inferred

No explicit QC controls found.

Evidence snippet: Scaling test-time compute via long Chain-of-Thought unlocks remarkable gains in reasoning capabilities, yet it faces practical limits due to the linear growth of KV cache and quadratic attention complexity.

Benchmarks / Datasets

provisional

Not extracted

Confidence: Provisional Source: Persisted extraction inferred

No benchmark anchors detected.

Evidence snippet: Scaling test-time compute via long Chain-of-Thought unlocks remarkable gains in reasoning capabilities, yet it faces practical limits due to the linear growth of KV cache and quadratic attention complexity.

Reported Metrics

provisional

Accuracy

Confidence: Provisional Source: Persisted extraction inferred

Useful for evaluation criteria comparison.

Evidence snippet: We apply reinforcement learning to incentivize this capability further, uncovering a critical insight: the accuracy gap between the highly efficient Fold mode and the exhaustive Unfold mode progressively narrows and eventually vanishes over the course of training.

Rater Population

provisional

Unknown

Confidence: Provisional Source: Persisted extraction inferred

Rater source not explicitly reported.

Evidence snippet: Scaling test-time compute via long Chain-of-Thought unlocks remarkable gains in reasoning capabilities, yet it faces practical limits due to the linear growth of KV cache and quadratic attention complexity.

Human Data Lens

Structured extraction is still processing. Below are provisional signals inferred from abstract text only.

  • Potential human-data signal: No explicit human-data keywords detected.
  • Potential benchmark anchors: No benchmark names detected in abstract.
  • Abstract highlights: 3 key sentence(s) extracted below.

Evaluation Lens

Evaluation fields are currently inferred heuristically from abstract text.

  • Potential evaluation modes: Automatic metrics
  • Potential metric signals: Accuracy
  • Confidence: Provisional (metadata-only fallback).

Research Brief

Deterministic synthesis

Scaling test-time compute via long Chain-of-Thought unlocks remarkable gains in reasoning capabilities, yet it faces practical limits due to the linear growth of KV cache and quadratic attention complexity.

Generated Apr 9, 2026, 2:14 PM · Grounded in abstract + metadata only

Key Takeaways

  • Scaling test-time compute via long Chain-of-Thought unlocks remarkable gains in reasoning capabilities, yet it faces practical limits due to the linear growth of KV cache and quadratic attention complexity.
  • In this paper, we introduce Accordion-Thinking, an end-to-end framework where LLMs learn to self-regulate the granularity of the reasoning steps through dynamic summarization.
  • This mechanism enables a Fold inference mode, where the model periodically summarizes its thought process and discards former thoughts to reduce dependency on historical tokens.

Researcher Actions

  • Compare this paper against nearby papers in the same arXiv category before using it for protocol decisions.
  • Validate inferred eval signals (Automatic metrics) against the full paper.
  • Use related-paper links to find stronger protocol-specific references.

Caveats

  • Generated from abstract + metadata only; no PDF parsing.
  • Signals below are heuristic and may miss details reported outside the abstract.

Recommended Queries

Related Papers

Papers are ranked by protocol overlap, extraction signal alignment, and semantic proximity.

No related papers found for this item yet.

Need human evaluators for your AI research? Scale annotation with expert AI Trainers.