DeLS-Spec: Decoupled Long-Short Contexts for Parallel Speculative Drafting
Quick Answer
DeLS-Spec introduces a decoupled long-short context approach to speculative decoding, enhancing DFlash's efficiency without joint training.
Quick Take
DeLS-Spec introduces a decoupled long-short context approach to speculative decoding, enhancing DFlash's efficiency without joint training. This method achieves significant speedup and improved acceptance lengths across math, code, and dialogue benchmarks on Qwen3 models, while maintaining low training costs.
Key Points
- DeLS-Spec combines long-context and short-context logits for improved modularity.
- Local head can be trained independently, reducing overall training costs.
- Experiments show consistent speedup over DFlash across various benchmarks.
- Applicable to math, code, and dialogue tasks with enhanced performance.
- Maintains flexibility by not tying the local head to a specific DFlash checkpoint.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Speculative decoding accelerates LLM inference by drafting multiple tokens and verifying them in parallel. Block-parallel drafters such as DFlash further improve drafting efficiency by predicting an entire block in one pass, but their position-wise predictions lack explicit intra-block causal conditioning. Recent methods such as Domino and DSpark attempt to introduce such causality into block-parallel drafting, but they require training the draft model from scratch, which limits their flexibility and increases training cost. We propose DeLS-Spec, a decoupled long-short context speculative decoding method. DeLS-Spec treats the fixed DFlash model as a long-context expert and introduces a lightweight local head as a short-context expert. The local head can be trained independently with a standard next-token prediction objective, without joint training with the target model or the DFlash backbone, leading to extremely low training cost. At inference time, DeLS-Spec combines long-context and short-context logits, and the local head is not tied to a specific DFlash checkpoint, making the method more modular and flexible. Experiments on Qwen3 models show that DeLS-Spec consistently improves speedup and average acceptance length over DFlash across math, code, and dialogue benchmarks.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2607.07409 [cs.CL] |
| (or arXiv:2607.07409v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.07409 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Hong-Kai Zheng [view email]
[v1]
Wed, 8 Jul 2026 13:41:52 UTC (243 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →Quantifying Prior Dominance in Systems
The study introduces the Normalized Context Utilization (NCU) metric to evaluate Retrieval-Augmented Generation (RAG) systems, revealing that Small Language Models (SLMs) outperform larger models in factual extraction. The findings indicate that traditional scaling laws yield diminishing returns, with a commercial API frequently failing against adversarial evidence due to systemic confidence collapse.