While draft-model-free speculative decoding offers a promising path to efficient LLM inference, it is frequently constrained by stale draft candidates and the high computational cost of the verification. To address these challenges, we propose ECHO, a hierarchical dual-loop framework that exploits the functional asymmetry between LLM layers. Leveraging the high discriminative efficiency of early layers and the authoritative distribution of final layers, ECHO bifurcates inference into a high-frequency inner loop and a low-frequency outer loop. Within the inner loop, early-layer bonus logits drive rapid, multi-step draft-tree exploration at a minimal cost. Simultaneously, the outer loop performs authoritative full-model verification through a state-reuse mechanism. Crucially, the outer loop also utilizes final-layer bonus logits to correct existing paths and supplement the tree with high-confidence candidates for subsequent cycles. Experimental results across diverse benchmarks demonstrate that ECHO significantly boosts mean accepted tokens and achieves a 2.4× to 2.9× speedup, outperforming existing state-of-the-art baselines with negligible engineering overhead and no extra deployment parameters, albeit with a one-shot fine-tuning dependency for optimal acceleration. The code is available at https://github.com/whucs21Mzy/ECHO.

On-policy distillation (OPD) for long-horizon language agents fails even when the teacher remains sequence-informative: in a failing Qwen3 configuration, the teacher still separates good from bad rollouts (sequence AUROC ≈ 0.75), yet the local token-gradient alignment 𝛼_diag decays from 0.71 to 0.09 and standard OPD recovers only 1.3% of the teacher–student gap. We restrict our claims to two coupled mechanisms, jointly termed CHOP (Counterfactual Hierarchical On-Policy distillation): (i) a per-state reliability gate routing between per-token reverse Kullback–Leibler matching and per-segment sliced-Wasserstein matching, and (ii) a counterfactual total-variation gate activating privileged information (PI) only when the PI changes the teacher’s distribution. The scope is explicit: CHOP targets low-reliability, long-horizon traces where token-local support matching plateaus; in short-horizon or high-reliability regimes Top-K local-support matching (LSM) is already sufficient and CHOP’s segment branch is rarely activated. CHOP recovers 31.7% of the anisotropic-Qwen3 gap (vs. 1.3%/14.5%/24.7% for OPD/Top-K LSM/Li et al. 2026); on SWE-bench Verified under a fixed OpenHands harness, CHOP-deployable prevents the OPD-induced regression (6.5%\!→\!13.2%, matched to off-the-shelf 11.8%, not a significant gain over off-the-shelf) and oracle PI reveals further headroom (14.2%, 95% CI [11.4,17.3]); on AIME 2024 at a 60K-token stress cap (a diagnostic horizon, not a deployment target), CHOP holds 44.9 vs. 42.8 for Top-K LSM (p=0.032), while standard OPD collapses to 28.4. Compute-matched, leakage, and PI-removed-teacher controls attribute the gain to supervision-object routing rather than extra compute or PI contamination.

As a crucial method in prompt engineering, In-Context Learning (ICL) enhances the generalization and knowledge utilization capabilities of Large Language Models (LLMs) (Dong et al., 2024). However, the lengthy retrieved contexts and limited token throughput in autoregressive models significantly constrain reasoning speed. To address this challenge, we propose N-Gram Trie Speculative Decoding, a novel approach that leverages the overlap between context and model output. This method constructs an n-gram trie from the context to generate drafts, accelerating token generation for LLMs. We evaluate our approach on summarization, Retrieval-Augmented Generation (RAG), and context-based Question Answering (QA) tasks. Experimental results on Vicuna-7B, Llama2-7B-Chat, and Llama3-8B-Instruct demonstrate substantial speed improvements without compromising accuracy. Compared with various strong baselines, our method achieves the highest mean speedup, showcasing its effectiveness and efficiency.

Large Language Models (LLMs) have demonstrated remarkable capabilities across diverse natural language processing tasks. However, their extensive memory requirements, particularly due to KV cache growth during long-text understanding and generation, present significant challenges for deployment in resource-constrained environments. Quantization has emerged as a promising solution to reduce memory consumption while preserving historical information. We propose XQuant, a training-free and plug-and-play framework that achieves ultra-low equivalent bit-width KV cache quantization. XQuant introduces two key innovations: a computationally negligible data-free calibration method and cross-layer KV cache compression, enabling quantization to sub-1.4 bits. Extensive experiments on TruthfulQA and LongBench demonstrate that XQuant outperforms state-of-the-art methods (e.g., KIVI-2bit and AsymKV-1.5bit) by achieving lower bit-width while maintaining superior performance, establishing a better trade-off between memory efficiency and model accuracy. The source code is available at https://github.com/brinenick511/XQuant.

Task-agnostic prompt compression leverages the redundancy in natural language to reduce computational overhead and enhance information density within prompts, especially in long-context scenarios. Existing methods predominantly rely on information entropy as the metric to compress lexical units, aiming to achieve minimal information loss. However, these approaches overlook two critical aspects: (i) the importance of attention-critical tokens at the algorithmic level, and (ii) shifts in information entropy during the compression process. Motivated by these challenges, we propose a dynamic attention-aware approach for task-agnostic prompt compression (DAC). This approach effectively integrates entropy and attention information, dynamically sensing entropy shifts during compression to achieve fine-grained prompt compression. Extensive experiments across various domains, including LongBench, GSM8K, and BBH, show that DAC consistently yields robust and substantial improvements across a diverse range of tasks and LLMs, offering compelling evidence of its efficacy.

Large language models (LLMs) based on Transformer Decoders have become the preferred choice for conversational generative AI. Despite the overall superiority of the Decoder architecture, the gradually increasing Key-Value (KV) cache during inference has emerged as a primary efficiency bottleneck, both in aspects of memory consumption and data transfer bandwidth limitations. To address these challenges, we propose a paradigm called KV-Latent. By down-sampling the Key-Value vector dimensions into a latent space, we can significantly reduce the KV Cache footprint and improve inference speed, only with a small amount of extra training, less than 1% of pre-training takes. Besides, we enhanced the stability of Rotary Positional Embedding applied on lower-dimensional vectors by modifying its frequency sampling mechanism, avoiding noise introduced by higher frequencies while retaining position attenuation. Our experiments, including both models with Grouped Query Attention and those without, have yielded satisfactory results. Finally, we conducted comparative experiments to study the impact of separately reducing Key and Value components on model’s performance. Our approach allows for the construction of more efficient language model systems, and opens the new possibility on KV Cache saving and efficient LLMs.

Large Language Models (LLMs) have achieved impressive accomplishments in recent years. However, the increasing memory consumption of KV cache has possessed a significant challenge to the inference system. Eviction methods have revealed the inherent redundancy within the KV cache, demonstrating its potential for reduction, particularly in deeper layers. However, KV cache reduction for shallower layers has been found to be insufficient. Based on our observation that, the KV cache exhibits a high degree of similarity. Based on this observation, we proposed a novel KV cache reduction method, SpindleKV, which balances both shallow and deep layers. For deep layers, we employ an attention weight based eviction method, while for shallow layers, we apply a codebook based replacement approach which is learnt by similarity and merging policy. Moreover, SpindleKV addressed the Grouped-Query Attention (GQA) dilemma faced by other attention based eviction methods. Experiments on two common benchmarks with three different LLMs shown that SpindleKV obtained better KV cache reduction effect compared to baseline methods, while preserving similar or even better model performance.