AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

FIRM-MoE:Fine-GrainedExpert Decomposition for Resource-Adaptive MoE Inference

Keyu Chen, Qihang Zhou, Bin Qian, Zhenyu Wen, Wenchao Meng, Shibo He

AAAI 2026 · Architectures and efficiency › Mixture of experts · approach: Expert routing and specialization

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
add fine-grained (sub-expert) prefetching (ablation)equal-ratio coarse-grained baseline · TruthfulQA, Qwen1.5-MoE-A2.7Bdecoding latency: 0.483 -> 0.365 (cache 128); speedups 1.24x to 1.41xImproved
add MOL multi-layer prediction (ablation)coarse-grained baseline · TruthfulQA, Qwen1.5-MoE-A2.7Bdecoding latency: 0.483 -> 0.400 (cache 128)Improved
add HEOP on top of FG + MOL (ablation)FIRM-MoE · TruthfulQA, Qwen1.5-MoE-A2.7Bdecoding latency: 0.337 -> 0.292 (cache 128)Improved
add HEOP on top of FG + MOL at cache 512 (ablation)FIRM-MoE · TruthfulQA, Qwen1.5-MoE-A2.7B cache 512decoding latency: 0.284 -> 0.263Improved
use MOL intersection over Top4 prediction (ablation)expert prediction · Qwen1.5-MoE-A2.7Bexpert utilization / prediction accuracy: 78.8 -> 90.5 utilization; accuracy 59.8Mixed
prefetch two linear layers (n=2) instead of one (ablation)FIRM-MoE · OLMoE-1B-7B cache 256E2E throughput: 5.31 -> 4.76Hurt
prefetch two linear layers (n=2) instead of one (ablation)FIRM-MoE · DeepSeek-MoE-16B-Base cache 768E2E throughput: 3.47 -> 3.67Improved
prefetch two linear layers (n=2) instead of one (ablation)FIRM-MoE · Qwen1.5-MoE-A2.7B cache 128E2E throughput: 3.51 -> 3.75Improved
use FIRM-MoE instead of Fiddler and Llama.cppinference system · ShareGPT E2Ethroughput: 1.31x average gainImproved
shrink expert cache from 768 to 128FIRM-MoE · Qwen1.5-MoE-A2.7Blatency: less than 30% degradation, 2.85x memory reductionMixed

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