CacheSpec is an inference optimization framework that transforms Program-of-Thoughts (PoT) style programs into reusable cache objects for large language models. By employing a small model for semantic variable extraction on cache hits and speculative drafting during target-LLM generation, CacheSpec reduces inference latency and improves cache reuse. Experiments on shopping, web, formula, and code QA datasets demonstrate up to 3.1× speedup in latency and 2.8× throughput gains over traditional PoT methods, while maintaining or improving task quality.
arXiv:2607. 20507v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used for program-aided reasoning, agentic decision making, and structured task execution, but these applications often incur high inference cost.
arXiv:2609.07966v1 Announce Type: new Abstract: Key--value (KV) cache compression is an effective way to reduce the memory overhead of large language model (LLM) inference, particularly for long-cont...
arXiv:2509.14093v3 Announce Type: replace-cross Abstract: Chain-of-Thought (CoT) prompting can substantially improve the reasoning ability of large language models (LLMs), but it often comes with hig...
arXiv:2608. 04771v1 Announce Type: new Abstract: Large Reasoning Models (LRMs) excel on complex tasks through long chain-of-thought (CoT) reasoning, but their lengthy intermediate steps cause severe overthinking that inflates inference cost.
Large Reasoning Models (LRMs) excel on complex tasks through long chain-of-thought (CoT) reasoning, but their lengthy intermediate steps cause severe overthinking that inflates inference cost. KV-cache compression is a common solution, yet existing reasoning-oriented methods apply a uniform policy across the trajectory and judge compression only by what it removes from the cache.
arXiv:2608.20359v1 Announce Type: new Abstract: Large language models (LLMs) are deployed for increasingly complex tasks involving planning and multi-step decision making, but high-quality performanc...
StepKV introduces a step-aware approach to compressing the key-value cache used during large language model inference, treating reasoning steps as primary units of retention rather than individual tokens. By linking cache entries to the steps that generated them and estimating each step’s utility from trajectory signals, StepKV assigns a combined token‑ and step‑level score to guide pruning. Experiments on multi‑hop question answering and long‑horizon web reasoning show that StepKV maintains accuracy even under tight cache budgets, outperforming token‑level baselines that suffer sharp performance drops.
BeaconKV is a training‑free key‑value cache compression technique for Large Reasoning Models that uses beacon queries—compact representatives of query clusters—to predict which KV pairs will be revisited during long‑horizon reasoning. By focusing on Thought Revisiting Tokens that re‑attend distant context, BeaconKV reduces memory usage up to 5.8× and improves throughput by over 4.3× while largely preserving cache accuracy across multiple open‑source LRMs and reasoning benchmarks.
The paper surveys memory‑centric compression techniques for large language models, framing long‑context handling as a memory‑management problem. It distinguishes compute‑active memory from compute‑inactive memory, and models operations such as writing, offloading, recall, and forgetting as state transitions within an information lifecycle. The authors propose a unified cost model and argue that future research should focus on dynamic, lifecycle‑aware memory management rather than static compression.
arXiv:2608. 19662v1 Announce Type: new Abstract: Agentic language models repeatedly encode tool and skill schemas that recur across requests in different combinations and orders, preventing standard prefix caching from reusing their key--value (KV) states.
The paper investigates how explicit reasoning in language models can lead to a "reasoning‑trace overflow," where the model spends most of its 32k‑token output budget on reasoning rather than producing code. Across seven models and six code benchmarks, overflow is rare on simple tasks but common on harder benchmarks, often due to repetitive reasoning loops. The study shows that disabling reasoning can sometimes be competitive, while prompt steering and budgeted transitions can mitigate overflow and improve token efficiency.
Long chain-of-thought (CoT) trajectories in large language model (LLM) reasoning cause severe inference bottlenecks due to rapid key-value (KV) cache growth. Current decoding-time compression methods mitigate this issue via token eviction, but typically assume a uniform budget distribution across all layers and heads.