Searcharxiv⌕ Search

arXiv subjects

Sangyeob Kim

Publications and source records attributed to Sangyeob Kim.

5 recordsLinked to original sources

CECAR: Cache & Expert Co-Aware Routing Accelerates On-Device Inference of MoE LLMs

Despite growing interest in on-device LLM deployment, large-scale Mixture-of-Experts (MoE) models remain impractical on resource-constrained devices, as sparse expert activation still requires all expert weights to be memory-resident. To address this, we identify three key observations for on-device MoE inference: (a) cache hit rates are fundamentally bounded even under Belady's optimal policy; (b) these bounds are insufficient for low-latency on-device inference; and (c) prior cache-aware routing based on binary cache presence can induce routing instability. Based on these insights, we propose CECAR (Cache & Expert Co-Aware Routing), a MoE inference system that jointly considers expert routing and cache management. Unlike prior cache-aware routing methods, CECAR uses an ML-based cache policy to predict the expected reuse distance of each expert and prioritizes routing and execution toward experts with smaller predicted reuse distances. On the Qwen3-30B-A3B model, CECAR achieves 12.5 tokens/s on a consumer-grade GPU with only 16 GB of VRAM, where non-resident experts are fetched from SSD, delivering a 4.5x speedup over conventional MoE inference and exceeding the decode speed of baselines that keep the entire model fully resident in DRAM. Code is available at https://github.com/hoonki2002/CECAR.

cs.LG↗

DPA-MoE: A Runtime Precision Control Framework for MoE Inference

Mixture-of-Experts (MoE) models reduce computation by routing each token to a small set of experts, but each selected expert is still used at a fixed precision. We present DPA-MoE, a runtime precision control framework that makes routed experts any-precision. DPA-MoE encodes expert feed-forward weights as prefix-decodable chunks using SharedScale AnyBCQ, profiles per-expert reconstruction residuals offline, and allocates chunks at inference time using live router gate scores. This post-routing allocation lets a single encoded MoE checkpoint operate at multiple effective weight budgets without retraining or re-encoding. Experiments on Qwen3-30B-A3B and DeepSeekMoE-16B show that DPA-MoE improves KL divergence and downstream accuracy over uniform chunk allocation at matched budgets, with modest steady-state throughput overhead in our prototype.

cs.CL↗

FlashMoE: Reducing SSD I/O Bottlenecks via ML-Based Cache Replacement for Mixture-of-Experts Inference on Edge Devices

Recently, Mixture-of-Experts (MoE) models have gained attention for efficiently scaling large language models. Although these models are extremely large, their sparse activation enables inference to be performed by accessing only a fraction of the model at a time. This property opens the possibility of on-device inference of MoE, which was previously considered infeasible for such large models. Consequently, various systems have been proposed to leverage this sparsity and enable efficient MoE inference for edge devices. However, previous MoE inference systems like Fiddler[8] or DAOP[13] rely on DRAM-based offloading and are not suitable for memory constrained on-device environments. As recent MoE models grow to hundreds of gigabytes, RAM-offloading solutions become impractical. To address this, we propose FlashMoE, a system that offloads inactive experts to SSD, enabling efficient MoE inference under limited RAM. FlashMoE incorporates a lightweight ML-based caching strategy that adaptively combines recency and frequency signals to maximize expert reuse, significantly reducing storage I/O. In addition, we built a user-grade desktop platform to demonstrate the practicality of FlashMoE. On this real hardware setup, FlashMoE improves cache hit rate by up to 51% over well-known offloading policies such as LRU and LFU, and achieves up to 2.6x speedup compared to existing MoE inference systems.

cs.LG↗

Two-Step Spike Encoding Scheme and Architecture for Highly Sparse Spiking-Neural-Network

This paper proposes a two-step spike encoding scheme, which consists of the source encoding and the process encoding for a high energy-efficient spiking-neural-network (SNN) acceleration. The eigen-train generation and its superposition generate spike trains which show high accuracy with low spike ratio. Sparsity boosting (SB) and spike generation skipping (SGS) reduce the amount of operations for SNN. Time shrinking multi-level encoding (TS-MLE) compresses the number of spikes in a train along time axis, and spike-level clock skipping (SLCS) decreases the processing time. Eigen-train generation achieves 90.3% accuracy, the same accuracy of CNN, under the condition of 4.18% spike ratio for CIFAR-10 classification. SB reduces spike ratio by 0.49x with only 0.1% accuracy loss, and the SGS reduces the spike ratio by 20.9% with 0.5% accuracy loss. TS-MLE and SLCS increases the throughput of SNN by 2.8x while decreasing the hardware resource for spike generator by 75% compared with previous generators.

cs.AR↗

GST: Group-Sparse Training for Accelerating Deep Reinforcement Learning

Deep reinforcement learning (DRL) has shown remarkable success in sequential decision-making problems but suffers from a long training time to obtain such good performance. Many parallel and distributed DRL training approaches have been proposed to solve this problem, but it is difficult to utilize them on resource-limited devices. In order to accelerate DRL in real-world edge devices, memory bandwidth bottlenecks due to large weight transactions have to be resolved. However, previous iterative pruning not only shows a low compression ratio at the beginning of training but also makes DRL training unstable. To overcome these shortcomings, we propose a novel weight compression method for DRL training acceleration, named group-sparse training (GST). GST selectively utilizes block-circulant compression to maintain a high weight compression ratio during all iterations of DRL training and dynamically adapt target sparsity through reward-aware pruning for stable training. Thanks to the features, GST achieves a 25 \%p $\sim$ 41.5 \%p higher average compression ratio than the iterative pruning method without reward drop in Mujoco Halfcheetah-v2 and Mujoco humanoid-v2 environment with TD3 training.

cs.LG↗