AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

AdaFuse: Accelerating Dynamic Adapter Inference via Token-Level Pre-Gating and Fused Kernel Optimization

Qiyang Li, Rui Kong, Yuchen Li, Hengyi Cai, Shuaiqiang Wang, Linghe Kong, Guihai Chen, Dawei Yin

AAAI 2026 · Architectures and efficiency › Mixture of experts · approach: Efficient MoE language model systems · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use token-level pre-gating with fused SGMM adapter switchingdynamic LoRA adapters, Llama2-7B · ShareGPT decodinglatency (ms/token): 8.5 (+254%) -> 3.1 (+29%)Improved
use SGMM fused kernel instead of simple merge (ablation)AdaFuse · ShareGPT decodinglatency (ms/token): 4.2 -> 3.1Improved
use simple layer-by-layer adapter mergingMoRAL · ShareGPT decodinglatency (ms/token): 8.5 -> 4.5Improved
switch from MoRAL layer-wise routing to AdaFuse pre-gating (with simple merge) (ablation)dynamic adapters · ShareGPT decodinglatency (ms/token): 4.5 -> 4.2Improved
use AdaFuse token-level pre-gating instead of block-wise PESCLlama2-7B · General ability averageaccuracy average: 60.45 -> 60.12No clear effect
use AdaFuse instead of static LoRALlama2-7B · General ability averageaccuracy average: 59.64 -> 60.12Improved
use AdaFuse instead of LoRALlama2-7B · CommonsenseQAaccuracy: 75.51 -> 79.03Improved
use AdaFuse instead of MoLALlama2-7B · ScienceQA/CommonsenseQA/OpenbookQA avgaccuracy average: 84.20 -> 83.60Hurt
use AdaFuse instead of LoRAMistral-7B · ScienceQA/CommonsenseQA/OpenbookQA avgaccuracy average: 86.06 -> 87.24Improved
use AdaFuse instead of MoRAL for memoryLlama2-7B · ShareGPT peak GPU memorymemory (GiB): 13.3 (+3%) -> 13.8 (+7%)No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy