arXiv:2512.23765v2 Announce Type: replace-cross Abstract: Speculative decoding (SD) accelerates large language model (LLM) inference by using a lightweight draft model to propose tokens and a stronge...
The paper introduces SwitchSD, an adaptive framework that treats copying as a latent control signal in large language model decoding. By training lightweight probes on internal representations, SwitchSD accurately detects genuine copy intent (AUC > 0.99) and dynamically switches between neural drafting and context-based copying. Experiments on Llama and Qwen models show up to 15 % throughput gains over state‑of‑the‑art baselines such as EAGLE3.
DLite is a history‑free speculative decoding framework designed for agentic AI workloads. It bypasses the need for a historical KV cache by reusing contextual hidden states from the target model and performing local speculative predictions, supplemented by a lightweight sequential head to maintain causal dependencies. Experiments demonstrate that DLite improves acceptance length by 2–6× and outperforms the state‑of‑the‑art DSpark by up to 70%.
The paper introduces SwitchSD, an adaptive framework that controls speculative decoding by distinguishing genuine copy intent from accidental repetitions using lightweight probes on a model’s internal representations. SwitchSD dynamically switches between neural drafting and context-based copying, achieving up to 15% throughput gains over existing baselines such as EAGLE3. The approach turns copying from a noisy heuristic into a principled, model-aware decoding regime.
Large Language Models suffer from high inference latency, especially when generating long chains of thought. Existing speculative decoding methods draft and verify tokens in parallel but ignore semantic equivalence, causing inefficient rejections. The proposed SemanticSpec framework verifies entire semantic sequences by probing internal hidden states, achieving up to 2.7× speedup on DeepSeekR1-32B and 2.1× on QwQ-32B while outperforming token‑level and sequence‑level baselines in both efficiency and effectiveness.
The paper introduces DescaPE, a decoding framework that uses internal model signals to reduce hallucinations in large language models. By identifying a factual‑salient layer span and training a lightweight probe to approximate its signal, DescaPE penalizes high‑risk continuations and rewards factually grounded ones during inference. Experiments on five factuality benchmarks across three LLMs show that DescaPE improves factuality with only a 1.10× latency overhead.
arXiv:2407. 21082v3 Announce Type: replace-cross Abstract: This paper presents a modular approach to accelerate inference in large language models (LLMs) by adding early exit heads at intermediate transformer layers.
The paper introduces DCOCO (DFlash Contextual Corrector), a lightweight mechanism that restores intra-block dependencies in speculative decoding by evaluating multiple candidate tokens for contextual compatibility across adjacent draft positions. DCOCO maintains parallel drafting while enabling contextually informed token selection at low additional cost. Across multiple verifier backbones and benchmarks, DCOCO achieves the highest average throughput, delivering speedups of up to 6.6× over vanilla decoding and outperforming competing methods by 6–23% under concurrent serving loads.
arXiv:2609.36590v1 Announce Type: cross Abstract: Self-speculative decoding accelerates large language model (LLM) inference by drafting tokens from the target model itself, but faces a sharp tradeof...
Speculative Decoding (SD) accelerates large language model inference by allowing a lightweight draft model to propose tokens that are subsequently verified in parallel by a larger target model. Recent approaches introduce lossy verification schemes to further improve efficiency by relaxing strict distributional matching.
arXiv:2607. 12696v1 Announce Type: cross Abstract: Sparse Mixture-of-Experts (MoE) models have become an important approach for scaling Large Language Models (LLMs), but their inference efficiency depends strongly on expert activation patterns.
The paper introduces Variance‑Calibrated Modulation (VCM), a training‑free pre‑decoding technique that reshapes language model probability distributions before truncation. VCM uses two dynamic mechanisms: a Contextual Searchlight via PMI to suppress stopwords and highlight context‑relevant tokens, and an Adaptive Self‑Debiasing that applies scale‑invariant penalization based on real‑time logit standard deviation. Experiments on open‑ended generation, factual QA, and mathematical reasoning show that VCM consistently reduces the likelihood trap, improving diversity, coherence, and reasoning accuracy with minimal computational cost.