This paper introduces DominoTree, a training-free best-first draft tree scored by Domino's conditional (non-factorized) correction along each root-to-node path, made practical by restricting the per-node correction to a candidate top-M.
Abstract
Speculative decoding accelerates LLM inference by drafting tokens and verifying them in parallel. Block-diffusion drafters such as DFlash model only per-position marginals, and tree methods such as DDTree expand candidate trees from those marginals. The released Domino drafter adds a GRU-based causal correction making each draft token's distribution path-dependent, a structure DDTree's factorized formulation cannot represent. We introduce DominoTree, a training-free best-first draft tree scored by Domino's conditional (non-factorized) correction along each root-to-node path, made practical by restricting the per-node correction to a candidate top-M. We evaluate it on eight benchmarks in a single-stream harness, and in SGLang, where it runs as an out-of-tree plugin against AR, DFlash, EAGLE-3 and Domino under identical flags. DominoTree attains the highest mean accepted length in every serving cell - two model sizes, single-request and concurrent load, context to 32K - and the highest Overall accepted length at every temperature in the research harness (21 of 24 per-dataset cells). A three-arm decomposition holding drafter, budget and verifier fixed separates the gain from applying the correction at all (+10.1% accepted length) from that of recomputing it along each candidate's realized path (+4.7% more), the part this paper adds. Where the round is verify-dominated, throughput follows: up to 7.3x over AR on Qwen3-8B, beating the released Domino decoder at its CUDA-graph best at every temperature, and inside SGLang winning single-request throughput by +12% over Domino on Qwen3-8B. On HELMET long context it beats Domino by +29-36% accepted length and +10-34% throughput at every length and both model sizes. Past a memory-constrained card's admission cap the chain wins goodput, and at our longest context, where prefill dominates, our lead over EAGLE-3 narrows to a tie.
DARTree is introduced, a training-free speculative decoding method that extends a pretrained AR correction head from chains to trees, and achieves the highest average acceptance length and speedup in all four model--temperature configurations.
Tianyi Li, Yaxin Luo, Xinyi Shang et al.· 0 citations
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.
Block-diffusion drafters like dFlash generate an entire block of draft tokens in a single forward pass, drastically reducing the overhead of multiple-token drafting in speculative decoding. The crucial final step of the single-pass discrete denoising process involves using the logit distribution at each position to sample conditionally independent tokens. The resulting draft is thus a set of per-position marginals, rather than a joint distribution: no draft token is guaranteed to depend on its predecessors. Such independently sampled marginals tend to produce sequences with tokens that are individually likely, but jointly improbable under the target model's distribution, which verifies each token conditionally. This can cause early rejection and limits acceptance length. To address this, we propose xPress as a means to restore the missing causality in diffusion drafters. xPress is a lightweight causal refiner that reconciles the whole diffusion block at once through parallel refinement, restoring and propagating causal dependencies across the draft without a token-by-token loop. On Qwen3-8B, across seven math, code, and chat benchmarks, xPress raises acceptance length by about 30% on average (up to +56%) and its end-to-end decoding throughput by about 1.3 on average (up to 1.7) compared to the original dFlash diffusion drafter.
Zheng Wang, Davis Wertheimer, Y. Lim et al.· 1 citation
Speculative decoding accelerates language-model inference by drafting future tokens that the target model verifies in parallel. A diffusion-style block head such as DFlash is an attractive drafter, predicting an entire block of future tokens in one forward pass. However, it is trained on per-position marginals rather than the joint block distribution, so the tokens it emits are individually plausible yet jointly incoherent. We introduce LiLiCorr, a Lightweight Likelihood-based model that Correlates the per-position marginal distributions a drafter already produces. It keeps the top-k tokens at each position as candidates and processes them jointly, producing for each an in and an out vector. A pair of adjacent candidates matches when the earlier one's out vector has high cosine similarity with the later one's in vector. These matches capture the block's joint structure without ever materializing the full joint distribution. One lightweight network pass produces all the vectors, and the pairwise scores are then computed in parallel as batched matrix operations, leaving only a cheap greedy walk sequential. We further co-train the drafter with LiLiCorr, so it learns to propose candidates that correlate into longer accepted sequences. Over the vanilla DFlash drafter, LiLiCorr raises acceptance length on every benchmark by 9 to 19%, while its scoring head accounts for about 2.8% of the per-block latency. Against DFlash and two concurrent methods that also restore coherence at draft time, LiLiCorr delivers the highest throughput in 70 of 72 settings: nine benchmarks at two target sizes under greedy and temperature-one decoding, and a throughput sweep over six concurrencies, two input lengths and three entropy tiers, with all systems equally optimized on a common serving stack. Extending LiLiCorr to inputs an order of magnitude longer than it was trained on preserves that lead.
M. Rusanovsky, Yoav Miron, Roy Uziel et al.· 0 citations
Speculative decoding accelerates LLM inference only when drafted continuations survive target-model verification. Semi-autoregressive drafters such as DSpark predict an entire token block with one backbone forward and refine it with a lightweight Markov head. However, DSpark decodes this block as a single chain, so an early mismatch invalidates the remaining suffix and limits the benefit of large draft blocks. We show that the conditional structure already learned by DSpark can support multiple parent-consistent continuations without retraining or additional backbone passes. We introduce Parent-Conditioned Drafting Tree (PCTree), which uses the pretrained Markov head to score alternative children separately for each concrete parent and allocates a fixed verification budget to the most probable paths. This converts DSpark's linear draft into a tree while preserving its one-pass parallel backbone. Across Qwen3-{4B,8B,14B} and nine benchmarks, at $B{=}7$, measured speedup gains over autoregressive (AR) decoding, relative to matched DSpark, range from $3.1\%$ to $29.5\%$. On Qwen3-4B GSM8K at $B{=}16$, PCTree increases mean acceptance length from $9.41$ to $11.16$ and three-run mean AR speedup from $6.14{\times}$ to $6.60{\times}$. These show that parent-conditioned branching can turn conditional capacity already present in a semi-autoregressive drafter into end-to-end inference gains through an inference-only change.
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.