KaVa: Latent Reasoning via Compressed KV-Cache Distillation

• URL: http://arxiv.org/abs/2510.02312v1
• Date: Thu, 02 Oct 2025 17:59:51 GMT
• Title: KaVa: Latent Reasoning via Compressed KV-Cache Distillation
• Abstract summary: KaVa is a framework that distills knowledge directly from a compressed KV-cache of a teacher into a latent-reasoning student via self-distillation.<n>We show that the abstract, unstructured knowledge within compressed KV-cache, which lacks direct token correspondence, can serve as a rich supervisory signal for a latent reasoning student.
• Score: 8.058130367727005
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Language Models (LLMs) excel at multi-step reasoning problems with explicit chain-of-thought (CoT), but verbose traces incur significant computational costs and memory overhead, and often carry redundant, stylistic artifacts. Latent reasoning has emerged as an efficient alternative that internalizes the thought process, but it suffers from a critical lack of supervision, limiting its effectiveness on complex, natural-language reasoning traces. In this work, we propose KaVa, the first framework that bridges this gap by distilling knowledge directly from a compressed KV-cache of the teacher into a latent-reasoning student via self-distillation, leveraging the representational flexibility of continuous latent tokens to align stepwise KV trajectories. We show that the abstract, unstructured knowledge within compressed KV-cache, which lacks direct token correspondence, can serve as a rich supervisory signal for a latent reasoning student. Empirically, the approach consistently outperforms strong latent baselines, exhibits markedly smaller degradation from equation-only to natural-language traces, and scales to larger backbones while preserving efficiency. These results establish compressed KV-cache distillation as a scalable supervision signal for latent reasoning, combining the accuracy of CoT-trained teachers with the efficiency and deployability of latent inference.

Related papers

• WhisperRec: Latent Reasoning for Efficient Foundation Recommendation Models [51.88075626243057]
  We propose WhisperRec, an efficient latent reasoning framework for foundation recommendation models (FRMs)<n>WhisperRec compresses teacher-generated Chain-of-Thought (CoT) into learnable latent reasoning tokens.<n>We show that WhisperRec consistently outperforms explicit-CoT methods and conventional baselines.
  arXiv  Detail & Related papers  (2026-07-29T08:48:35Z)
• Answer-then-Edit: Reasoning Skeleton Editing for Anti-Distillation with Preserved Utility [13.916452483499185]
  Anti-distillation (AD) has been proposed to generate defensive outputs that hinder distillation effectiveness.<n>We propose textbfunderlineSkeleton-textbfunderlineGuided textbfunderlineReasoning textbfunderlineEditing (SGRE)<n>In the answer stage, the teacher model first generates clean reasoning traces, preserving the original reasoning accuracy.<n>In the editing stage, we introduce a three-stage strategy consisting of reasoning skeleton extraction, skeleton graph coarsening, and skeleton verbal
  arXiv  Detail & Related papers  (2026-05-12T04:36:20Z)
• CRISP: Compressing Redundancy in Chain-of-Thought via Intrinsic Saliency Pruning [25.840307083323456]
  Long Chain-of-Thought (CoT) reasoning is pivotal for the success of recent reasoning models but suffers from high computational overhead and latency.<n>This paper presents textbfCompressing textbfRedundancy in Chain-of-Thought via textbfIntrinsic textbfSaliency textbfPruning (textbfCRISP), a framework that compresses CoT by exploiting the model's intrinsic saliency.
  arXiv  Detail & Related papers  (2026-04-19T07:26:55Z)
• Constraint-Rectified Training for Efficient Chain-of-Thought [60.52883907721588]
  Chain-of-Thought (CoT) has significantly enhanced the reasoning capabilities of Large Language Models (LLMs)<n>While longer reasoning traces can improve answer quality and unlock abilities such as self-correction, they also incur high inference costs and often introduce redundant steps, known as overthinking.<n>Recent research seeks to develop efficient reasoning strategies that balance reasoning length and accuracy.
  arXiv  Detail & Related papers  (2026-02-13T02:13:45Z)
• Accordion-Thinking: Self-Regulated Step Summaries for Efficient and Readable LLM Reasoning [62.680551162054975]
  We introduce an end-to-end framework where LLMs learn to self-regulate the granularity of the reasoning steps through dynamic summarization.<n>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.<n>Our Accordion-Thinker demonstrates that with learned self-compression, LLMs can tackle complex reasoning tasks with minimal dependency token overhead.
  arXiv  Detail & Related papers  (2026-02-03T08:34:20Z)
• ForesightKV: Optimizing KV Cache Eviction for Reasoning Models by Learning Long-Term Contribution [84.41751286055909]
  We develop a training-based KV cache eviction framework that learns to predict which KV pairs to evict during longtext generations.<n>We formulate cache eviction as a Markov Decision Process and apply the GRPO algorithm to mitigate the significant language modeling loss increase on low-entropy tokens.
  arXiv  Detail & Related papers  (2026-02-03T07:16:51Z)
• ReGuLaR: Variational Latent Reasoning Guided by Rendered Chain-of-Thought [49.203970812338916]
  Explicit reasoning chains introduce substantial computational redundancy.<n>Recent latent reasoning methods attempt to mitigate this by compressing reasoning processes into latent space.<n>We propose Rendered CoT-Guided variational Latent Reasoning (ReGuLaR)
  arXiv  Detail & Related papers  (2026-01-30T17:08:06Z)
• Latent-Space Contrastive Reinforcement Learning for Stable and Efficient LLM Reasoning [16.244366307890832]
  We propose textbfDeepLatent Reasoning (DLR), a latent-space bidirectional contrastive reinforcement learning framework.<n>This framework shifts the trial-and-error cost from expensive token-level full sequence generation to the continuous latent manifold.<n> Experiments demonstrate that DLR achieves more stable training convergence, supports longer-horizon reasoning chains, and facilitates the sustainable accumulation of reasoning capabilities.
  arXiv  Detail & Related papers  (2026-01-24T03:18:22Z)
• SkipKV: Selective Skipping of KV Generation and Storage for Efficient Inference with Large Reasoning Models [25.509962883211]
  Large reasoning models (LRMs) often cost significant key-value (KV) cache overhead, due to their linear growth with the chain-of-thought (CoT) reasoning process.<n>We present textbfSkipKV, a KV compression method for selective textiteviction and textitgeneration operating at a coarse-grained sentence-level sequence removal.
  arXiv  Detail & Related papers  (2025-12-08T19:32:06Z)
• Provable Benefit of Curriculum in Transformer Tree-Reasoning Post-Training [76.12556589212666]
  We show that curriculum post-training avoids the exponential complexity bottleneck.<n>Under outcome-only reward signals, reinforcement learning finetuning achieves high accuracy with sample complexity.<n>We establish guarantees for test-time scaling, where curriculum-aware querying reduces both reward oracle calls and sampling cost from exponential to order.
  arXiv  Detail & Related papers  (2025-11-10T18:29:54Z)
• Fast Thinking for Large Language Models [67.7238685892317]
  We introduce Latent Codebooks for Fast Thinking, a framework that uses concise CoT sketches only during training to learn a codebook of discrete strategy priors.<n>At inference, the model conditions on a handful of continuous thinking switches distilled from the codebook in a single pass, enabling strategy-level guidance without producing explicit reasoning tokens.
  arXiv  Detail & Related papers  (2025-09-28T04:19:48Z)
• SIM-CoT: Supervised Implicit Chain-of-Thought [108.30049193668083]
  Implicit Chain-of-Thought (CoT) methods offer a token-efficient alternative to explicit CoT reasoning in Large Language Models.<n>We identify a core latent instability issue when scaling the computational budget of implicit CoT.<n>We propose SIM-CoT, a plug-and-play training module that introduces step-level supervision to stabilize and enrich the latent reasoning space.
  arXiv  Detail & Related papers  (2025-09-24T17:01:32Z)
• ConciseHint: Boosting Efficient Reasoning via Continuous Concise Hints during Generation [74.37307916314407]
  We propose a framework dubbed ConciseHint, which continuously encourages the reasoning model to speak concisely.<n>Experiments on the state-of-the-art LRMs, including DeepSeek-R1 and Qwen-3 series, demonstrate that our method can effectively produce concise reasoning.
  arXiv  Detail & Related papers  (2025-06-23T16:20:44Z)
• EPiC: Towards Lossless Speedup for Reasoning Training through Edge-Preserving CoT Condensation [37.6583581020347]
  We study the problem of CoT condensation for resource-efficient reasoning training.<n>We propose an Edge-Preserving Condensation method, EPiC, which selectively retains only the initial and final segments of each CoT trace.
  arXiv  Detail & Related papers  (2025-06-04T17:49:10Z)
• SCOUT: Teaching Pre-trained Language Models to Enhance Reasoning via Flow Chain-of-Thought [37.53215651690168]
  Chain of Thought (CoT) prompting improves the reasoning performance of large language models (LLMs) by encouraging step by step thinking.<n>While promising, CoT-based approaches often require costly pretraining and lack a principled framework for how reasoning should evolve.<n>We propose SCOUT, a lightweight fine tuning framework that enables Flow CoT style reasoning without the need for pretraining.
  arXiv  Detail & Related papers  (2025-05-30T03:43:24Z)
• ToDi: Token-wise Distillation via Fine-Grained Divergence Control [9.958797874295355]
  Token-wise Distillation (ToDi) is a novel method that adaptively combines Forward KL and Reverse KL per token using a sigmoid-based weighting function.<n>ToDi consistently outperforms recent distillation baselines using uniform or less granular strategies.
  arXiv  Detail & Related papers  (2025-05-22T06:51:16Z)
• ConCISE: Confidence-guided Compression in Step-by-step Efficient Reasoning [64.93140713419561]
  Large Reasoning Models (LRMs) perform strongly in complex reasoning tasks via Chain-of-Thought (CoT) prompting, but often suffer from verbose outputs.<n>Existing fine-tuning-based compression methods either operate post-hoc pruning, risking disruption to reasoning coherence, or rely on sampling-based selection.<n>We introduce ConCISE, a framework designed to generate concise reasoning chains, integrating Confidence Injection to boost reasoning confidence, and Early Stopping to terminate reasoning when confidence is sufficient.
  arXiv  Detail & Related papers  (2025-05-08T01:40:40Z)
• Efficient Reasoning with Hidden Thinking [48.96945580741641]
  Chain-of-Thought (CoT) reasoning has become a powerful framework for improving complex problem-solving capabilities.<n>We propose $textbfHeima$ (as hidden llama), an efficient reasoning framework that leverages reasoning CoTs at hidden latent space.<n>Heima model achieves higher generation efficiency while maintaining or even better zero-shot task accuracy.
  arXiv  Detail & Related papers  (2025-01-31T15:10:29Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.