DALI: A Workload-Aware Offloading Framework for Efficient MoE Inference on Local PCs

• URL: http://arxiv.org/abs/2602.03495v1
• Date: Tue, 03 Feb 2026 13:11:52 GMT
• Title: DALI: A Workload-Aware Offloading Framework for Efficient MoE Inference on Local PCs
• Authors: Zeyu Zhu, Gang Li, Peisong Wang, Zitao Mo, Minnan Pei, Zhuoran Song, Xiaoyao Liang, Jian Cheng,
• Abstract summary: Offloading MoE expert parameters to host memory and leveraging both CPU and GPU computation has emerged as a promising direction to support such models on resourceconstrained local PCs.<n>Existing prefetching techniques fail to accurately predict high-workload experts, leading to costly inaccurate prefetches.<n>We propose DALI, a workloaDAware offLoadIng framework for efficient MoE inference on local PCs.
• Score: 28.841079546977458
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Mixture of Experts (MoE) architectures significantly enhance the capacity of LLMs without proportional increases in computation, but at the cost of a vast parameter size. Offloading MoE expert parameters to host memory and leveraging both CPU and GPU computation has recently emerged as a promising direction to support such models on resourceconstrained local PC platforms. While promising, we notice that existing approaches mismatch the dynamic nature of expert workloads, which leads to three fundamental inefficiencies: (1) Static expert assignment causes severe CPUGPU load imbalance, underutilizing CPU and GPU resources; (2) Existing prefetching techniques fail to accurately predict high-workload experts, leading to costly inaccurate prefetches; (3) GPU cache policies neglect workload dynamics, resulting in poor hit rates and limited effectiveness. To address these challenges, we propose DALI, a workloaDAware offLoadIng framework for efficient MoE inference on local PCs. To fully utilize hardware resources, DALI first dynamically assigns experts to CPU or GPU by modeling assignment as a 0-1 integer optimization problem and solving it efficiently using a Greedy Assignment strategy at runtime. To improve prefetching accuracy, we develop a Residual-Based Prefetching method leveraging inter-layer residual information to accurately predict high-workload experts. Additionally, we introduce a Workload-Aware Cache Replacement policy that exploits temporal correlation in expert activations to improve GPU cache efficiency. By evaluating across various MoE models and settings, DALI achieves significant speedups in the both prefill and decoding phases over the state-of-the-art offloading frameworks.

Related papers

• Remoe: Towards Efficient and Low-Cost MoE Inference in Serverless Computing [29.98726492279776]
  Mixture-of-Experts (MoE) has become a dominant architecture in large language models.<n>MoEs incurs high inference costs due to memory-intensive parameter caching.<n>We propose Remoe, a heterogeneous MoE inference system tailored for serverless computing.
  arXiv  Detail & Related papers  (2025-12-21T10:27:50Z)
• Eliminating Multi-GPU Performance Taxes: A Systems Approach to Efficient Distributed LLMs [61.953548065938385]
  We introduce the ''Three Taxes'' (Bulk Synchronous, Inter- Kernel Data Locality, and Kernel Launch Overhead) as an analytical framework.<n>We propose moving beyond the rigid BSP model to address key inefficiencies in distributed GPU execution.<n>We observe a 10-20% speedup in end-to-end latency over BSP-based approaches.
  arXiv  Detail & Related papers  (2025-11-04T01:15:44Z)
• HybriMoE: Hybrid CPU-GPU Scheduling and Cache Management for Efficient MoE Inference [5.015541720729724]
  HybriMoE is a hybrid CPU-GPU inference framework that improves resource utilization through a novel CPU-GPU scheduling and cache management system.<n>We implement HybriMoE on top of the kTransformers framework and evaluate it on three widely used MoE-based LLMs.
  arXiv  Detail & Related papers  (2025-04-08T10:47:37Z)
• QuartDepth: Post-Training Quantization for Real-Time Depth Estimation on the Edge [55.75103034526652]
  We propose QuartDepth which adopts post-training quantization to quantize MDE models with hardware accelerations for ASICs.<n>Our approach involves quantizing both weights and activations to 4-bit precision, reducing the model size and computation cost.<n>We design a flexible and programmable hardware accelerator by supporting kernel fusion and customized instruction programmability.
  arXiv  Detail & Related papers  (2025-03-20T21:03:10Z)
• Numerical Pruning for Efficient Autoregressive Models [87.56342118369123]
  This paper focuses on compressing decoder-only transformer-based autoregressive models through structural weight pruning.<n>Specifically, we propose a training-free pruning method that calculates a numerical score with Newton's method for the Attention and modules, respectively.<n>To verify the effectiveness of our method, we provide both theoretical support and extensive experiments.
  arXiv  Detail & Related papers  (2024-12-17T01:09:23Z)
• DAOP: Data-Aware Offloading and Predictive Pre-Calculation for Efficient MoE Inference [14.676716521856813]
  Mixture-of-Experts (MoE) models face significant deployment challenges on memory-constrained devices.<n>We presentP, an on-device MoE inference engine to optimize parallel GPU- CPU execution.<n>P outperforms traditional expert caching and prefetching methods by up to 8.20x and offloading techniques by 1.35x while maintaining accuracy.
  arXiv  Detail & Related papers  (2024-12-16T07:59:21Z)
• HOBBIT: A Mixed Precision Expert Offloading System for Fast MoE Inference [54.40808356999408]
  We present HOBBIT, a mixed precision expert offloading system to enable flexible and efficient MoE inference. Our key insight is that dynamically replacing less critical cache-miss experts with low precision versions can substantially reduce expert-loading latency. HOBBIT achieves up to a 9.93x speedup in decoding compared to state-of-the-art MoE offloading systems.
  arXiv  Detail & Related papers  (2024-11-03T04:25:46Z)
• Bypass Back-propagation: Optimization-based Structural Pruning for Large Language Models via Policy Gradient [57.9629676017527]
  We propose an optimization-based structural pruning that learns the pruning masks in a probabilistic space directly by optimizing the loss of the pruned model.<n>We achieve this by learning an underlying Bernoulli distribution to sample binary pruning masks.<n>Experiments conducted on LLaMA, LLaMA-2, LLaMA-3, Vicuna, and Mistral models demonstrate the promising performance of our method in efficiency and effectiveness.
  arXiv  Detail & Related papers  (2024-06-15T09:31:03Z)
• PIM-Opt: Demystifying Distributed Optimization Algorithms on a Real-World Processing-In-Memory System [21.09681871279162]
  Modern Machine Learning (ML) training on large-scale datasets is a time-consuming workload. It relies on the optimization algorithm Gradient Descent (SGD) due to its effectiveness, simplicity, and generalization performance. processor-centric architectures suffer from low performance and high energy consumption while executing ML training workloads. Processing-In-Memory (PIM) is a promising solution to alleviate the data movement bottleneck.
  arXiv  Detail & Related papers  (2024-04-10T17:00:04Z)
• Harnessing Deep Learning and HPC Kernels via High-Level Loop and Tensor Abstractions on CPU Architectures [67.47328776279204]
  This work introduces a framework to develop efficient, portable Deep Learning and High Performance Computing kernels. We decompose the kernel development in two steps: 1) Expressing the computational core using Processing Primitives (TPPs) and 2) Expressing the logical loops around TPPs in a high-level, declarative fashion. We demonstrate the efficacy of our approach using standalone kernels and end-to-end workloads that outperform state-of-the-art implementations on diverse CPU platforms.
  arXiv  Detail & Related papers  (2023-04-25T05:04:44Z)
• SmartDeal: Re-Modeling Deep Network Weights for Efficient Inference and Training [82.35376405568975]
  Deep neural networks (DNNs) come with heavy parameterization, leading to external dynamic random-access memory (DRAM) for storage. We present SmartDeal (SD), an algorithm framework to trade higher-cost memory storage/access for lower-cost computation. We show that SD leads to 10.56x and 4.48x reduction in the storage and training energy, with negligible accuracy loss compared to state-of-the-art training baselines.
  arXiv  Detail & Related papers  (2021-01-04T18:54:07Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.