aiqbench analysis
The paper targets a systems bottleneck in speculative decoding: for hybrid models that mix Gated DeltaNet and mixture-of-experts layers, verification cost rises sharply with verification length, unlike dense models. Its reported gains (1.30x-1.33x over DFlash, 1.75x-1.88x over autoregressive decoding) come from single-request greedy decoding on three models, so they should be checked against batched serving and other workloads. Readers building LLM inference stacks should watch whether the routing-correction and block-parallel kernel ideas generalize beyond this model family.
HybridSpec: Throughput-Aware Speculative Decoding for Hybrid GDN-MoE Models
arXiv:2611.04055v1 Announce Type: cross Abstract: In speculative decoding, parallel drafters such as DFlash make drafting cheap, so target verification dominates each round. A central choice is the verification length, the number of tokens that the target processes in each round. Verifying more draft tokens can accept more of them, but work after the first rejection is wasted. For hybrid target models that combine Gated DeltaNet (GDN) and mixture-of-experts (MoE) layers, verification time grows much faster with this length than for dense models. In a single-request profile of Qwen3.6-35B-A3B with DFlash, it grew by 71% from length 4 to 16, almost entirely in GDN and MoE layers, compared with 4%-6% for a dense Qwen3-4B target. Each GDN layer must update a recurrent state at every token and recover it at the end of the accepted prefix. The MoE cost depends on the experts that the tokens activate, which are known only once verification runs. We present HybridSpec, which selects the verification length on the GPU from draft-side acceptance estimates and measured target latency. Verification then runs this length plus one extra token up to the first MoE router, before any expert runs. There, a lightweight routing correction uses the observed expert assignments to decide whether to keep that token. The rest of verification runs only on the selected tokens. Block-parallel GDN kernels avoid storing a state after every token and reconstruct the accepted state of every GDN layer in one launch. In isolated benchmarks at verification lengths 4 to 16 with every draft token accepted, these kernels, including accepted-state reconstruction, are $1.54\times$-$4.04\times$ faster than the default kernels. With three GDN-MoE target models in single-request greedy decoding, HybridSpec is $1.75\times$-$1.88\times$ faster than autoregressive decoding and $1.30\times$-$1.33\times$ faster than DFlash.
Abstract:In speculative decoding, parallel drafters such as DFlash make drafting cheap, so target verification dominates each round. A central choice is the verification length, the number of tokens that the target processes in each round. Verifying more draft tokens can accept more of them, but work after the first rejection is wasted. For hybrid target models that combine Gated DeltaNet (GDN) and mixture-of-experts (MoE) layers, verification time grows much faster with this length than for dense models. In a single-request profile of Qwen3.6-35B-A3B with DFlash, it grew by 71% from length 4 to 16, almost entirely in GDN and MoE layers, compared with 4%-6% for a dense Qwen3-4B target. Each GDN layer must update a recurrent state at every token and recover it at the end of the accepted prefix. The MoE cost depends on the experts that the tokens activate, which are known only once verification runs. We present HybridSpec, which selects the verification length on the GPU from draft-side acceptance estimates and measured target latency. Verification then runs this length plus one extra token up to the first MoE router, before any expert runs. There, a lightweight routing correction uses the observed expert assignments to decide whether to keep that token. The rest of verification runs only on the selected tokens. Block-parallel GDN kernels avoid storing a state after every token and reconstruct the accepted state of every GDN layer in one launch. In isolated benchmarks at verification lengths 4 to 16 with every draft token accepted, these kernels, including accepted-state reconstruction, are $1.54\times$-$4.04\times$ faster than the default kernels. With three GDN-MoE target models in single-request greedy decoding, HybridSpec is $1.75\times$-$1.88\times$ faster than autoregressive decoding and $1.30\times$-$1.33\times$ faster than DFlash.
Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2611.04055 [cs.AI]
(or arXiv:2611.04055v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2611.04055 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.
Full text under the publisher's open licence · 379 words