arXiv subjects

Chuansai Zhou

Publications and source records attributed to Chuansai Zhou.

3 recordsLinked to original sources

arXiv2026

MoE Proxy Models for Low-Cost Failure Reproduction and Diagnosis in LLM RL Post-Training

Reinforcement learning (RL) post-training of large language models (LLMs) is computationally intensive and involves complex system pipelines with substantial debugging overhead. In practice, factors such as framework adaptation, numerical precision, and operator implementation can cause failures, including gradient overflow and loss divergence. Reproducing such failures directly on large models requires considerable time and computational resources. This paper systematically analyzes failures encountered during large-scale RL training on the Huawei Ascend platform, summarizes representative failure types, and identifies three model-side factors relevant to fault reproduction. Based on these factors, we propose a proxy-model construction method for low-cost fault investigation and auxiliary diagnosis. It employs structure-preserving, clustering-based expert pruning to select representative experts while retaining the model's backbone architecture, routing mechanism, and basic task capabilities. Our experimental results show that the proxy models reduce accelerator requirements by 50%-87.5% and achieve up to a 33.3x reduction in per-step NPU-hour cost, while preserving major training dynamics and reproducing fault responses consistent with the original models. Overall, the proxy models can serve as low-cost surrogates for fault reproduction, targeted validation, and auxiliary diagnosis in RL post-training.

cs.LG↗

arXiv2026

Towards Efficient Agents: A Co-Design of Inference Architecture and System

The rapid development of large language model (LLM)-based agents has unlocked new possibilities for autonomous multi-turn reasoning and tool-augmented decision-making. However, their real-world deployment is hindered by severe inefficiencies that arise not from isolated model inference, but from the systemic latency accumulated across reasoning loops, context growth, and heterogeneous tool interactions. This paper presents AgentInfer, a unified framework for end-to-end agent acceleration that bridges inference optimization and architectural design. We decompose the problem into four synergistic components: AgentCollab, a hierarchical dual-model reasoning framework that balances large- and small-model usage through dynamic role assignment; AgentSched, a cache-aware hybrid scheduler that minimizes latency under heterogeneous request patterns; AgentSAM, a suffix-automaton-based speculative decoding method that reuses multi-session semantic memory to achieve low-overhead inference acceleration; and AgentCompress, a semantic compression mechanism that asynchronously distills and reorganizes agent memory without disrupting ongoing reasoning. Together, these modules form a Self-Evolution Engine capable of sustaining efficiency and cognitive stability throughout long-horizon reasoning tasks. Experiments on the BrowseComp-zh and DeepDiver benchmarks demonstrate that through the synergistic collaboration of these methods, AgentInfer reduces ineffective token consumption by over 50%, achieving an overall 1.8-2.5 times speedup with preserved accuracy. These results underscore that optimizing for agentic task completion-rather than merely per-token throughput-is the key to building scalable, efficient, and self-improving intelligent systems.

cs.CL↗

arXiv2020

Detecting Suspected Epidemic Cases Using Trajectory Big Data

Emerging infectious diseases are existential threats to human health and global stability. The recent outbreaks of the novel coronavirus COVID-19 have rapidly formed a global pandemic, causing hundreds of thousands of infections and huge economic loss. The WHO declares that more precise measures to track, detect and isolate infected people are among the most effective means to quickly contain the outbreak. Based on trajectory provided by the big data and the mean field theory, we establish an aggregated risk mean field that contains information of all risk-spreading particles by proposing a spatio-temporal model named HiRES risk map. It has dynamic fine spatial resolution and high computation efficiency enabling fast update. We then propose an objective individual epidemic risk scoring model named HiRES-p based on HiRES risk maps, and use it to develop statistical inference and machine learning methods for detecting suspected epidemic-infected individuals. We conduct numerical experiments by applying the proposed methods to study the early outbreak of COVID-19 in China. Results show that the HiRES risk map has strong ability in capturing global trend and local variability of the epidemic risk, thus can be applied to monitor epidemic risk at country, province, city and community levels, as well as at specific high-risk locations such as hospital and station. HiRES-p score seems to be an effective measurement of personal epidemic risk. The accuracy of both detecting methods are above 90\% when the population infection rate is under 20\%, which indicates great application potential in epidemic risk prevention and control practice.

stat.AP↗