PreSTO introduces Predictive SubTree Prefetching to accelerate PowerMH, a power sampling method for large language models. By prefetching future proposals in a tree of accept/reject decisions and using a predictive traversal rule, PreSTO maintains the same Metropolis–Hastings transition kernel while reducing model calls. Experiments on seven datasets and five base LLMs show median speedups of 1.71×, 1.91×, and 2.33× for PowerMH, EntropyCut, and MultiTryMH, respectively, with fewer and shorter model calls.
arXiv:2608.30427v1 Announce Type: cross Abstract: Speculative decoding speeds up generation with an efficient draft model (drafter) that proposes tokens for a target model to verify in one pass, pres...
arXiv:2606. 09926v1 Announce Type: cross Abstract: Sampling from the sequence-level power distribution $p^\alpha$ elicits RL-level reasoning from base language models without any parameter updates, but the standard Metropolis--Hastings (MH), a Markov Chain Monte Carlo (MCMC) sampler, is both expensive and slow-mixing.
arXiv:2609.22098v1 Announce Type: new Abstract: Speculative decoding accelerates language-model inference by letting a cheap drafter propose tokens that the target model verifies in parallel. Recent...
TreeSpark is a new training and inference framework that extends the semi‑autoregressive DSpark architecture to improve speculative decoding. It builds parent‑conditioned candidate trees in a single backbone pass and then replays their expansion, pruning, and packing during training, using survival negative log‑likelihood and a capped coverage hinge to weight retained edges and promote target‑important alternatives. Evaluations on Qwen3‑4B and Qwen3‑8B across nine general benchmarks and six writing domains show that TreeSpark increases mean acceptance length by 23.4%–31.6% and mean single‑request throughput by 15.1%–25.0% compared to DSpark, with additional gains when using the proposed supervision. whyItMatters":"TreeSpark demonstrates that a unified training and inference approach can substantially boost both acceptance length and throughput in speculative decoding, offering measurable performance improvements for large language models."
arXiv:2607.08642v4 Announce Type: replace Abstract: Speculative decoding speeds up large language model inference: a small drafter proposes several tokens, and the target model verifies them in one p...
arXiv:2608.20375v1 Announce Type: new Abstract: Tree-based speculative decoding raises the mean accepted tokens of standard speculative decoding by verifying multiple draft paths, and existing tree b...
Power‑SMC is a training‑free sampling method that targets the sequence‑level power distribution used to sharpen large language model outputs for reasoning tasks. It maintains multiple candidate sequences, assigns importance weights based on how well each matches the power distribution, and periodically prunes low‑scoring candidates to keep inference latency close to standard decoding. Experiments on MATH500, GSM8K, GPQA, and HumanEval show that Power‑SMC matches or surpasses Metropolis–Hastings sampling in accuracy, preserves output diversity, and can accelerate inference speed by up to 17.6×.
LEV is a decision model that performs reasoning by looping its layers instead of generating text, allowing it to deepen its computation on a per-question basis. Trained jointly across all loop steps, a single LEV checkpoint can answer at any depth, improving accuracy—especially on hard questions—while keeping inference fast. On benchmark tests, LEV outperforms dense and cascade models by 5.4 to 8.8 accuracy points with sub‑50 ms latency, and on‑demand depth selection saves 15–24% of computation for minimal accuracy loss.
arXiv:2609.24698v1 Announce Type: new Abstract: Repeated execution of the target model during autoregressive decoding is a major source of LLM inference latency. Unlike linear speculation, which foll...
CANOPY is a multi‑fidelity tree bandit algorithm that learns where a piecewise‑smooth prior holds instead of assuming global smoothness. It uses cheap random‑path probes to certify local aggregation bias and then focuses expensive leaf evaluations on cells where smoothness is violated. The method achieves provable fixed‑budget and regret guarantees that scale with the number of discontinuities, matching smooth‑tree rates when no violations exist and approaching structure‑blind search when violations are dense.
Prefill‑only decision models evaluate every candidate in a menu in a single forward pass, avoiding decoding and reducing cost by one to two orders of magnitude compared to generative language models. The paper demonstrates that when only the candidate menu changes, the model’s post‑intervention accuracy can be predicted solely from the cached first‑pass distribution using a simple estimator that renormalizes and selects the argmax, without any labels or second pass. Across seven model families, ten datasets, and two task types, this menu‑only intervention prediction is within 4.2 points of actual accuracy, and in one family it is exact, whereas a probability‑level variant fails by 21 points, indicating the property resides in ranking rather than calibrated probabilities.
arXiv:2606. 01682v1 Announce Type: cross Abstract: Selecting the best response from multiple small-model samples using a stronger scorer is a simple inference-time strategy, but fails when the small model has already committed to incorrect reasoning paths.