Pith. sign in

REVIEW 5 cited by

Towards MoE Deployment: Mitigating Inefficiencies in Mixture-of-Expert (MoE) Inference

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2303.06182 v2 pith:PNXNGMAU submitted 2023-03-10 cs.DC cs.ARcs.CLcs.LG

classification cs.DCcs.ARcs.CLcs.LG
keywords timesmemorybufferingexpertinefficienciesproposebalancingdeployment
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
abstract

Mixture-of-Experts (MoE) models have gained popularity in achieving state-of-the-art performance in a wide range of tasks in computer vision and natural language processing. They effectively expand the model capacity while incurring a minimal increase in computation cost during training. However, deploying such models for inference is difficult due to their large size and complex communication pattern. In this work, we provide a characterization of two MoE workloads, namely Language Modeling (LM) and Machine Translation (MT) and identify their sources of inefficiencies at deployment. We propose three optimization techniques to mitigate sources of inefficiencies, namely (1) Dynamic gating, (2) Expert Buffering, and (3) Expert load balancing. We show that dynamic gating improves maximum throughput by 6.21-11.23$\times$ for LM, 5.75-10.98$\times$ for MT Encoder and 2.58-5.71$\times$ for MT Decoder. It also reduces memory usage by up to 1.36$\times$ for LM and up to 1.1$\times$ for MT. We further propose Expert Buffering, a new caching mechanism that only keeps hot, active experts in GPU memory while buffering the rest in CPU memory. This reduces static memory allocation by up to 1.47$\times$. We finally propose a load balancing methodology that provides additional scalability to the workload.

Discussion (0). Sign in to comment.

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. HAP: Hybrid Adaptive Parallelism for Efficient Mixture-of-Experts Inference

    cs.DC 2025-08 conditional novelty 6.0 of 10

    HAP uses ILP over module-specific hybrid parallel choices to speed up MoE inference, reporting up to 1.77x versus tensor parallelism on tested GPUs.

  2. Utility-Driven Speculative Decoding for Mixture-of-Experts

    cs.DC 2025-06 conditional novelty 6.0 of 10

    Cascade is a utility-driven speculation manager that makes speculative decoding practical for MoE LLMs by disabling it when expert-activation cost exceeds token gain and hill-climbing to the best speculation length.

  3. A Replicate-and-Quantize Strategy for Plug-and-Play Load Balancing of Sparse Mixture-of-Experts LLMs

    cs.LG 2026-02 reject novelty 5.0 of 10

    Replicate-and-Quantize copies the busiest MoE expert as a quantized duplicate and compresses the least important expert, lowering a new Load-Imbalance Score by up to 1.4x while accuracy varies by roughly -1.2 to +3.0 points.

  4. Compute Can't Handle the Truth: Why Communication Tax Prioritizes Memory and Interconnects in Modern AI Infrastructure

    cs.DC 2025-07 reject novelty 4.0 of 10

    A CXL-based disaggregated memory architecture with hybrid XLink interconnects is proposed and prototyped, claiming large speedups for memory-bound AI and HPC workloads.

  5. MoE-GPS: Guidlines for Prediction Strategy for Dynamic Expert Duplication in MoE Load Balancing

    cs.LG 2025-06 conditional novelty 4.0 of 10

    MoE-GPS shows that predicting only the aggregate token-to-expert distribution, instead of exact token routing, can improve MoE inference latency by more than 23% in simulation, with guidelines for when each strategy wins.

Pith tools