analiza aiqbench
This research addresses a critical bottleneck in deploying Mixture-of-Experts models by optimizing router design to enhance expert coactivation during speculative decoding. By leveraging techniques like shared experts and autoregressive selection, the study achieves a 21% throughput improvement while mitigating memory transfer costs associated with parallel token verification. The findings offer a practical pathway for accelerating large-scale inference without sacrificing model accuracy.
Efficient Mixture-of-Experts with Speculative Decoding via Expert Coactivation
arXiv:2609.22471v1 Announce Type: cross Abstract: Mixture-of-Experts (MoE) models are increasingly deployed alongside Speculative Decoding (SD) to accelerate inference, but combining the two is challenging. SD improves the inference speed of dense models by verifying groups of tokens in parallel. However, the inference speedup for SD with MoEs depends heavily on the number of tokens being verified. Using more verification tokens results in more experts being transferred from DRAM to the Neural Processing Unit (NPU), which increases the memory transfer cost. This negatively impacts model runtime, as memory transfer is typically the bottleneck in inference. In this work, we investigate the impact of MoE router design during training on the speed of MoEs with SD. We find that routers with high degrees of expert coactivation result in much faster runtimes, mitigating the impact of using more verification tokens. Motivated by this observation, we assess the impact of various router design choices on expert coactivation and runtime using billion-parameter transformer models. We find that combining a global load-balancing loss, shared experts, a consistency loss, and an autoregressive expert selection mechanism during training results in significantly stronger expert coactivation. This increased coactivation translates into higher overall runtime throughput: our exploration yields a model that improves throughput by 21% over MoE baselines, while maintaining on-par accuracy with the baseline MoE.
Abstract:Mixture-of-Experts (MoE) models are increasingly deployed alongside Speculative Decoding (SD) to accelerate inference, but combining the two is challenging. SD improves the inference speed of dense models by verifying groups of tokens in parallel. However, the inference speedup for SD with MoEs depends heavily on the number of tokens being verified. Using more verification tokens results in more experts being transferred from DRAM to the Neural Processing Unit (NPU), which increases the memory transfer cost. This negatively impacts model runtime, as memory transfer is typically the bottleneck in inference. In this work, we investigate the impact of MoE router design during training on the speed of MoEs with SD. We find that routers with high degrees of expert coactivation result in much faster runtimes, mitigating the impact of using more verification tokens. Motivated by this observation, we assess the impact of various router design choices on expert coactivation and runtime using billion-parameter transformer models. We find that combining a global load-balancing loss, shared experts, a consistency loss, and an autoregressive expert selection mechanism during training results in significantly stronger expert coactivation. This increased coactivation translates into higher overall runtime throughput: our exploration yields a model that improves throughput by 21% over MoE baselines, while maintaining on-par accuracy with the baseline MoE.
Machine Learning (cs.LG); Computation and Language (cs.CL)
arXiv:2609.22471 [cs.LG]
(or arXiv:2609.22471v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.22471 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 · 309 cuvinte
Subiecte conexe
Efficient MiniMax-M3 Inference on AMD Instinct GPUs with ATOM and ATOMesh
AMD · 21 iul. 2026
S2-MoE: Enabling Efficient Self-Speculative Decoding for Mixture-of-Experts on Edge Devices
arXiv · 18 aug. 2026
[AINews] OpenAI reports Navier-Stokes singularity find in 88 hours using Astra-next, roughly 10,000 agents and 130B tokens (>$40M), a contender for second ever Millennium Prize awarded
Latent Space · 9 sept. 2026
JetBrains Releases Mellum2: A 12B MoE Model for Fast, Specialized Tasks in Multi-Model AI Pipelines
MarkTechPost · 2 iun. 2026