This work proposes a dependent block drafter based on a low-rank latent mixture over token positions, complemented by an acceptance-oriented training objective that directly targets the expected verified length.
Abstract
Speculative decoding accelerates large language models'inference by using a lightweight drafter to propose multiple future tokens and a target model to verify them. While recent block and diffusion-style drafters can predict several positions in a single pass, their training and sampling procedures are typically optimized for greedy decoding or assume that positions in the draft block are conditionally independent. This assumption becomes brittle in non-greedy speculative decoding, where the target distribution is deliberately stochastic and multiple continuations become plausible. We study this mismatch for block diffusion drafters and show that the accepted draft length degrades as the entropy of the target sampling distribution increases. We propose a dependent block drafter based on a low-rank latent mixture over token positions, complemented by an acceptance-oriented training objective that directly targets the expected verified length. Experiments with Qwen3-4B and Qwen3-8B on GSM8K, MT-Bench, HumanEval, and creative-writing benchmarks show that our approach, namely DBLast, consistently improves accepted length over independent block sampling, especially in higher-entropy decoding regimes.
LiLiCorr, a Lightweight Likelihood-based model that Correlates the per-position marginal distributions a drafter already produces, is introduced, a Lightweight Likelihood-based model that Correlates the per-position marginal distributions a drafter already produces.
M. Rusanovsky, Yoav Miron, Roy Uziel et al.· 0 citations
LibraSpec, a training-free and plug-and-play algorithm that iteratively determines the speculative length using drafter confidence scores, is developed and it is proved that LibraSpec monotonically converges toward the optimal speculative length.
Zexun Lin, Yuan Feng, Junlin Lv et al.· 0 citations
DFly is proposed, a block-diffusion framework combining a hybrid target-conditioning backbone with a predecessor-conditioned autoregressive head, improving target-feature utilization and intra-block dependency modeling while keeping generation parallel, and DFly treats verification as a shared batch-level resource.
Hong Liu, Rui Cen, Jun-Han Shi et al.· arXiv.org· 2 citations
AdaFlash framework is proposed, comprising two components: an on-policy distillation algorithm with reverse-KL divergence tailored for diffusion drafters, bringing stable convergence and effectively reducing domain-level variance and an adaptive length head that dynamically adjusts the candidate sequence length on the fly, substantially lowering the verification cost of the target model and handling token-level variance.
Verification-Aware Training is introduced, a plug-in framework that simulates verification at every training step and turns the resulting accept and reject patterns into supervision and improves average acceptance length and wall-clock speedup across math, code, and chat benchmarks.
Geonmo Gu, Byeongho Heo, Heejae Jun et al.· 0 citations
Speculative decoding is a leading technique to reduce the cost of autoregressive generation by using a small drafter to propose several tokens, which are then verified in parallel by a larger target model. Speculative diffusion decoding (SDD) further removes sequential drafting by generating every position in a draft block in parallel with a discrete diffusion model. However, SDD still invokes the target on every block, leaving verification as a potential bottleneck. This paper recognizes that this creates a new control handle: whether to invoke the verifier at all. Thus, we study verifier skipping, a lossy policy that commits a selected draft prefix directly, and ask which confidence signal should schedule it. Interestingly, our study finds that better token predictors need not yield better schedulers: skips require contiguous high-confidence prefixes, while short skips can induce additional drafting rounds. To study this mismatch, we compare raw confidence with learned marginal and conditional survival scores under the same policy, using Strict SDD, lenience, and top-$k$ acceptance as baselines. On HumanEval with DiffuCoder-7B-Instruct and Qwen3-32B, all three confidence signals save $9.6\%$ to $13.5\%$ of verifier calls at the same observed pass@1 as Strict SDD. Surprisingly, raw confidence saves the most; marginal survival has higher positionwise AUROC than raw confidence at most positions, yet neither learned signal dominates online. Our analysis shows that verifier skipping is a useful new lossy axis and, surprisingly, its key challenge is prefix scheduling rather than token prediction alone.
