arXiv subjects

Shaoyuan Chen

Publications and source records attributed to Shaoyuan Chen.

7 recordsLinked to original sources

arXiv2026

DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation

Speculative decoding accelerates Large Language Model (LLM) inference by decoupling draft generation from target verification. While recent parallel drafters efficiently propose long token sequences in a single forward pass, they suffer from rapid acceptance decay due to a lack of inter-token dependencies. Furthermore, indiscriminately verifying these extended blocks wastes critical batch capacity on tokens with high rejection risks, severely degrading throughput in high-concurrency serving systems. We introduce DSpark, a speculative decoding framework that unifies high-throughput parallel generation with adaptive, load-aware verification. To maintain draft quality, DSpark utilizes a semi-autoregressive architecture, coupling a parallel backbone with a lightweight sequential module, to introduce intra-block dependency modeling and mitigate suffix decay. To optimize system efficiency, DSpark employs confidence-scheduled verification, dynamically tailoring the verification length for each request based on estimated prefix survival probabilities and engine-specific throughput profiles. On offline benchmarks across diverse domains, DSpark substantially improves the accepted length over state-of-the-art autoregressive and parallel drafters. When deployed within the DeepSeek-V4 serving system under live user traffic, DSpark successfully mitigates verification waste. Compared to the established production baseline (MTP-1), DSpark accelerates per-user generation speeds by 60 to 85 percent at matched throughput levels. More importantly, by preventing severe throughput degradation under strict interactivity constraints, it enables performance tiers that were previously unattainable, shifting the Pareto frontier of our serving system.

cs.AI↗

arXiv2026

Surviving Partial Rank Failures in Wide Expert-Parallel MoE Inference

Mixture-of-Experts (MoE) serving relies on wide expert parallelism (EP) to aggregate the memory capacity and bandwidth of many GPUs within one inference instance. This efficiency comes with a systems cost: every decoding step depends on token dispatch and combination across all active EP ranks, so even one rank failure can disrupt the entire service. Existing EP stacks handle such failures poorly because they treat membership as a fixed configuration established at initialization. The same rank set determines communicator state, expert placement, and the routing metadata baked into CUDA execution graphs, leaving the system with no way to shrink around a failure while keeping the instance valid. This paper argues that partial-failure tolerance should instead be formulated as a live EP validity problem. We present EEP, a communication and runtime substrate that represents membership as explicit, mutable runtime state. EEP repairs the specific state invalidated by a fault: it restores peer reachability without rebuilding the communication substrate, repairs lost expert coverage through a bandwidth-aware hierarchy, and reintegrates repaired ranks without forcing healthy ranks to recapture their CUDA graphs. We implement EEP in an EP serving stack integrated with SGLang and evaluate it under steady-state serving, failure recovery, and rank reintegration. The results show that explicit mutable membership preserves the steady-state fast path, staying within 4.4% of a fixed-membership DeepEP baseline under static serving, while turning a local rank fault from whole-instance downtime into two bounded interruptions. On a single-rank failure workload, EEP incurs an 11s recovery pause and an 8s reintegration pause, and restores throughput to within 95% of the pre-fault level within 52s, whereas a fixed-membership full-restart baseline remains unavailable until 348s.

cs.DC↗

arXiv2026

DualPath: Breaking the Storage Bandwidth Bottleneck in Agentic LLM Inference

The performance of multi-turn, agentic LLM inference is increasingly dominated by KV-Cache storage I/O rather than computation. In prevalent disaggregated architectures, loading the massive KV-Cache from external storage creates a fundamental imbalance: storage NICs on prefill engines become bandwidth-saturated, while those on decoding engines remain idle. This asymmetry severely constrains overall system throughput. We present DualPath, an inference system that breaks this bottleneck by introducing dual-path KV-Cache loading. Beyond the traditional storage-to-prefill path, DualPath enables a novel storage-to-decode path, in which the KV-Cache is loaded into decoding engines and then efficiently transferred to prefill engines via RDMA over the compute network. DualPath combines this optimized data path -- which inherently avoids network congestion and avoids interference with latency-critical model execution communications -- with a global scheduler that dynamically balances load across prefill and decode engines. Our evaluation on three models with production agentic workloads demonstrates that DualPath improves offline inference throughput by up to 1.87$\times$ on our in-house inference system. It can also improve online serving throughput by an average factor of 1.96$\times$ without violating SLO.

cs.DC↗

arXiv2025

OTARo: Once Tuning for All Precisions toward Robust On-Device LLMs

Large Language Models (LLMs) fine-tuning techniques not only improve the adaptability to diverse downstream tasks, but also mitigate adverse effects of model quantization. Despite this, conventional quantization suffers from its structural limitation that hinders flexibility during the fine-tuning and deployment stages. Practical on-device tasks demand different quantization precisions (i.e. different bit-widths), e.g., understanding tasks tend to exhibit higher tolerance to reduced precision compared to generation tasks. Conventional quantization, typically relying on scaling factors that are incompatible across bit-widths, fails to support the on-device switching of precisions when confronted with complex real-world scenarios. To overcome the dilemma, we propose OTARo, a novel method that enables on-device LLMs to flexibly switch quantization precisions while maintaining performance robustness through once fine-tuning. OTARo introduces Shared Exponent Floating Point (SEFP), a distinct quantization mechanism, to produce different bit-widths through simple mantissa truncations of a single model. Moreover, to achieve bit-width robustness in downstream applications, OTARo performs a learning process toward losses induced by different bit-widths. The method involves two critical strategies: (1) Exploitation-Exploration Bit-Width Path Search (BPS), which iteratively updates the search path via a designed scoring mechanism; (2) Low-Precision Asynchronous Accumulation (LAA), which performs asynchronous gradient accumulations and delayed updates under low bit-widths. Experiments on popular LLMs, e.g., LLaMA3.2-1B, LLaMA3-8B, demonstrate that OTARo achieves consistently strong and robust performance for all precisions.

cs.LG↗

arXiv2025

Efficient Heterogeneous Large Language Model Decoding with Model-Attention Disaggregation

Transformer-based large language models (LLMs) exhibit impressive performance in generative tasks but also introduce significant challenges in real-world serving due to inefficient use of the expensive, computation-optimized accelerators. Although disaggregated serving architectures have been proposed to split different phases of LLM inference, the efficiency of decoding phase is still low. This is caused by the varying resource demands of different operators in the transformer-based LLMs. Specifically, the attention operator is memory-intensive, exhibiting a memory access pattern that clashes with the strengths of modern accelerators, especially for long context requests. To enhance the efficiency of LLM decoding, we introduce model-attention disaggregation. This approach leverages a collection of cheap, memory-optimized devices for the attention operator while still utilizing high-end accelerators for other parts of the model. This heterogeneous setup ensures that each component is tailored to its specific workload, maximizing overall performance and cost efficiency. Our comprehensive analysis and experiments confirm the viability of splitting the attention computation over multiple devices. Also, the communication bandwidth required between heterogeneous devices proves to be manageable with prevalent networking technologies. To further validate our theory, we develop and deploy Lamina, an LLM inference system that incorporates model-attention disaggregation in a distributed heterogeneous cluster. Experimental results indicate that Lamina can provide 16.1 ~ 90.1% higher estimated throughput than existing solutions with similar costs.

cs.LG↗

arXiv2024

Federated Knowledge Transfer Fine-tuning Large Server Model with Resource-Constrained IoT Clients

The training of large models, involving fine-tuning, faces the scarcity of high-quality data. Compared to the solutions based on centralized data centers, updating large models in the Internet of Things (IoT) faces challenges in coordinating knowledge from distributed clients by using their private and heterogeneous data. To tackle such a challenge, we propose KOALA (Federated Knowledge Transfer Fine-tuning Large Server Model with Resource-Constrained IoT Clients) to impel the training of large models in IoT. Since the resources obtained by IoT clients are limited and restricted, it is infeasible to locally execute large models and also update them in a privacy-preserving manner. Therefore, we leverage federated learning and knowledge distillation to update large models through collaboration with their small models, which can run locally at IoT clients to process their private data separately and enable large-small model knowledge transfer through iterative learning between the server and clients. Moreover, to support clients with similar or different computing capacities, KOALA is designed with two kinds of large-small model joint learning modes, namely to be homogeneous or heterogeneous. Experimental results demonstrate that compared to the conventional approach, our method can not only achieve similar training performance but also significantly reduce the need for local storage and computing power resources.

cs.LG↗

arXiv2020

Joint Transceiver Design Based on Dictionary Learning Algorithm for SCMA

With the explosively increasing demands on the network capacity, throughput and number of connected wireless devices, massive connectivity is an urgent problem for the next generation wireless communications. In this paper, we propose a grant-free access protocol for massive connectivity that utilizes a large number of antennas in a base station (BS) and is expected to be widely deployed in cellular networks. The scheme consists of a sparse structure in sparse code multiple access (SCMA) and receiver processing based on dictionary learning (DL). A large number of devices can transmit data without any scheduling process. Unlike existing schemes, whose signal schedulings require a lot of overhead, the scheduling overhead required by the proposed scheme is negligible, which is attractive for resource utilization and transmission power efficiency. The numerical results show that the proposed scheme has promising performance in massive connectivity scenario of cellular networks.

eess.SP↗