analiza aiqbench
This paper introduces D3D, a novel diffusion-based drafting method that eliminates sequential dependencies in speculative decoding by enabling parallel generation and verification of token blocks. By decoupling the drafting process from target-model hidden states, the approach achieves significant speedups—up to 7.48× on GSM8K with Qwen3-32B—while maintaining high acceptance rates even after pruning model layers. The work advances the efficiency frontier of large language model inference through asynchronous decoding techniques grounded in machine learning theory.
D3D: Decoupled Diffusion Drafting for Asynchronous Speculative Decoding
arXiv:2611.06425v1 Announce Type: new Abstract: Target-conditioned diffusion drafters predict candidate tokens in parallel, yet each new block must wait for updated target-model hidden states. This forces drafting and verification to run sequentially, limiting speedup even though each block is generated in one pass. We introduce D3D, a standalone diffusion drafter that removes this inter-round dependency. D3D combines blockwise denoising with auxiliary autoregressive supervision to achieve long accepted prefixes without target hidden states. This independence lets D3D draft successor blocks during verification in one packed pass, enabling asynchronous speculative decoding. We progressively prune layers according to their effect on acceptance length to reduce the cost of packed drafting. Removing five of 36 layers retains over 99\% of mean acceptance length across three benchmarks with Qwen3-8B as the target. Across seven benchmarks, D3D averages 5.00$\times$ and 5.71$\times$ decoding speedup with Qwen3-8B and Qwen3-32B targets, compared with 4.03$\times$ and 3.67$\times$ for DFlash. On GSM8K with Qwen3-32B, speedup reaches 7.48$\times$. Code is publicly available at https://github.com/fudan-generative-vision/D3D.
Abstract:Target-conditioned diffusion drafters predict candidate tokens in parallel, yet each new block must wait for updated target-model hidden states. This forces drafting and verification to run sequentially, limiting speedup even though each block is generated in one pass. We introduce D3D, a standalone diffusion drafter that removes this inter-round dependency. D3D combines blockwise denoising with auxiliary autoregressive supervision to achieve long accepted prefixes without target hidden states. This independence lets D3D draft successor blocks during verification in one packed pass, enabling asynchronous speculative decoding. We progressively prune layers according to their effect on acceptance length to reduce the cost of packed drafting. Removing five of 36 layers retains over 99\% of mean acceptance length across three benchmarks with Qwen3-8B as the target. Across seven benchmarks, D3D averages 5.00$\times$ and 5.71$\times$ decoding speedup with Qwen3-8B and Qwen3-32B targets, compared with 4.03$\times$ and 3.67$\times$ for DFlash. On GSM8K with Qwen3-32B, speedup reaches 7.48$\times$. Code is publicly available at this https URL.
Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2611.06425 [cs.AI]
(or arXiv:2611.06425v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2611.06425 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 256 cuvinte
Subiecte conexe
When RL Meets Adaptive Speculative Training: A Unified Training-Serving System
arXiv · 5 aug. 2026
Shallow Queries, Mature Values: Depth-Asynchronous Self-Speculation for Looped Transformers
arXiv · 30 sept. 2026
Together AI at ICML 2026: frontier research across the full stack
Together AI · 30 iun. 2026
ASPIRE: Asynchronous Batched Self-Speculative Decoding for Long-Context LLM Inference
arXiv · 17 sept. 2026