inference-efficiency

Tag

Cards List
#inference-efficiency

RestoreKV: Recovering Full-Cache Behavior Under Aggressive Query-Agnostic KV Cache Eviction

Hugging Face Daily Papers ↗ · 2026-08-02 Cached

RestoreKV introduces a learned restoration mechanism that complements query-agnostic KV cache eviction, recovering full-cache behavior under aggressive budgets by generating a compact context-conditioned restore cache via a single LoRA-adapted pass, improving performance on four long-context benchmarks.

0 favorites 0 likes
#inference-efficiency

Masked Distillation: Internalizing the Chain-of-Thought in Language Models

arXiv cs.AI ↗ · 2026-07-28 Cached

Masked distillation is a knowledge-distillation framework that trains a student LLM to predict only solution tokens while a reasoning teacher provides feedback, aiming to internalize chain-of-thought computation into model parameters. The method shows task-dependent success, working on GSM8K but requiring small scaffolds for harder tasks like Countdown.

0 favorites 0 likes
#inference-efficiency

Adversarial Prompts for Acceptance Collapse in Speculative Decoding

arXiv cs.CL ↗ · 2026-07-27 Cached

Introduces ADSD, a prompt-suffix attack that causes acceptance collapse in speculative decoding by forcing the draft model to propose tokens the target model is unlikely to accept, increasing inference time while preserving task quality.

0 favorites 0 likes
#inference-efficiency

MiniCache: Reusable Program Caching with Small Model Interfaces for Efficient LLM Inference

arXiv cs.AI ↗ · 2026-07-24 Cached

MiniCache is a program caching framework that reuses computation across similar requests by parameterizing Program-of-Thought programs, using small models for semantic variable extraction and speculative drafting to improve LLM inference efficiency.

0 favorites 0 likes
#inference-efficiency

Show HN: Echo – Fable-level results at 1/3 the cost using open-weight models

Hacker News Top ↗ · 2026-07-23 Cached

Echo is a system that achieves performance comparable to the Fable model at one-third the cost by efficiently allocating inference across open-weight models. It provides free credits and requires no credit card.

0 favorites 0 likes
#inference-efficiency

SelKV: Selective KV Cache Merging with Per-Token Merge-or-Drop and Attention Compensation

arXiv cs.AI ↗ · 2026-07-21 Cached

SelKV is a training-free framework for KV cache compression that uses a soft cosine gate for selective merging and an attention-ratio compensation mechanism to correct softmax imbalance, achieving near-lossless generation at 25% cache size and 3.3x decoding speedup on LongBench.

0 favorites 0 likes
#inference-efficiency

Adaptive Multi-Step Lookahead Decoding for Diffusion Language Models

arXiv cs.CL ↗ · 2026-07-20 Cached

Proposes AdaLook, an adaptive multi-step lookahead decoding framework for masked diffusion language models that dynamically determines rollout depth and branch expansion based on candidate-score variance, achieving better accuracy-decoding steps trade-off compared to existing one-step lookahead decoding methods.

0 favorites 0 likes
#inference-efficiency

Less Experts, Faster Decoding: Cost-Aware Speculative Decoding for Mixture-of-Experts

arXiv cs.CL ↗ · 2026-07-15 Cached

This paper proposes EcoSpec, a cost-aware speculative decoding framework for Mixture-of-Experts models that accounts for expert activation costs during draft selection. It achieves up to 1.62× speedup on large-scale MoE models like DeepSeek-V3.1, Qwen3-235B-A22B, and GPT-OSS-120B by reducing expert footprint without modifying the target model's verification rule.

0 favorites 0 likes
#inference-efficiency

Doomed from the Start: Early Abort of LLM Agent Episodes via a Recall-Controlled Probe Cascade

arXiv cs.AI ↗ · 2026-07-08 Cached

This paper proposes a recall-controlled abort cascade that uses lightweight probes on LLM agent internal representations to detect and abort doomed episodes early, saving up to 47% inference compute while maintaining high recall of successful episodes.

0 favorites 0 likes
#inference-efficiency

Understanding Evaluation Illusion in Diffusion Large Language Models

arXiv cs.CL ↗ · 2026-06-30 Cached

This paper identifies evaluation inconsistencies in diffusion LLM decoding methods, showing that prompt template choice significantly impacts rankings, and proposes guidelines for reliable evaluation.

0 favorites 0 likes
#inference-efficiency

BlockPilot: Instance-Adaptive Policy Learning for Diffusion-based Speculative Decoding

Hugging Face Daily Papers ↗ · 2026-06-30 Cached

BlockPilot proposes an instance-adaptive policy that predicts the optimal block size for diffusion-based speculative decoding, achieving significant speedup with minimal overhead.

0 favorites 0 likes
#inference-efficiency

@FinanceYF5: Paper:

X AI KOLs Following ↗ · 2026-06-25 Cached

This paper introduces LatentMAS, a training-free framework for multi-agent systems that enables large language model agents to collaborate directly in continuous latent space via shared latent working memory, achieving up to 14.6% higher accuracy and 4x faster inference while reducing token usage by over 70%.

0 favorites 0 likes
#inference-efficiency

Learning to Refine Hidden States for Reliable LLM Reasoning

arXiv cs.LG ↗ · 2026-06-17 Cached

Proposes ReLAR, a reinforcement-guided latent refinement framework that iteratively updates hidden representations in LLMs before decoding, improving reasoning reliability and efficiency compared to chain-of-thought methods.

0 favorites 0 likes
#inference-efficiency

Models Take Notes at Prefill: KV Cache Can Be Editable and Composable

arXiv cs.LG ↗ · 2026-06-17 Cached

This paper proposes that the KV cache in transformers acts as a notebook of memoized conclusions, enabling surgical editing and composition without full recomputation. The method achieves significant latency reductions while preserving decision equivalence across model scales.

0 favorites 0 likes
#inference-efficiency

PolyKV: Heterogeneous Retention and Allocation for KV Cache Compression

arXiv cs.LG ↗ · 2026-06-16 Cached

PolyKV is a layer-wise KV cache compression framework that assigns heterogeneous eviction policies and non-uniform budgets per layer, significantly improving over uniform baselines on LongBench with LLaMA-3.1-8B and Qwen3-8B.

0 favorites 0 likes
#inference-efficiency

@liliana_hotsko: How do you give a code LLM knowledge of an entire repository without paying for it at every single query? We introduce …

X AI KOLs Timeline ↗ · 2026-06-11 Cached

Introduces Code2LoRA, a hypernetwork that converts an entire code repository into a LoRA adapter for code LLMs, eliminating inference-time token overhead for repository-level knowledge.

0 favorites 0 likes
#inference-efficiency

TimeRouter: Efficient and Adaptive Routing of Time-Series Foundation Models

arXiv cs.LG ↗ · 2026-06-11 Cached

TimeRouter introduces an efficient routing framework for time-series foundation models that uses lightweight discriminative routing and selective gating to adaptively select the best expert model without LLM overhead, achieving state-of-the-art on the GIFT-EVAL leaderboard.

0 favorites 0 likes
#inference-efficiency

@MaximeRivest: current llm architecture is stupid (if not stupid its, at least, wasteful). take these 3 prompts of 4 context chunks: […

X AI KOLs Following ↗ · 2026-06-09 Cached

A tweet criticizes current LLM architecture for wasteful recomputation due to order-dependent context, and proposes encoding context units separately to enable order-invariant, efficient caching and generation.

0 favorites 0 likes
#inference-efficiency

When to Think Deeply: Inhibitory Deliberation for LLM Reasoning

arXiv cs.CL ↗ · 2026-06-08 Cached

IDPR is a framework for response-conditioned inhibitory deliberation that first generates a fast intuitive answer, then uses an inhibition controller to decide whether to invoke slow reasoning, achieving efficiency gains while maintaining accuracy.

0 favorites 0 likes
#inference-efficiency

LazyAttention: Efficient Retrieval-Augmented Generation with Deferred Positional Encoding

arXiv cs.CL ↗ · 2026-06-04 Cached

LazyAttention introduces a novel attention mechanism that defers positional encoding to enable zero-copy, position-agnostic KV cache reuse across multiple requests. The approach reduces time-to-first-token by 1.37× and increases throughput by 1.40× compared to Block-Attention in RAG settings with skewed document distributions.

0 favorites 0 likes
← Previous
Next →
← Back to home

Submit Feedback