Yubin Xia

Yunpeng Dong, Jingkai He, Yuze Hou et al.

LLM-powered AI agents require high-frequency state exploration (e.g., test-time tree search and reinforcement learning), relying on rapid checkpoint and rollback (C/R) of the complete sandbox state, including files and process state (e.g., memory, contexts, etc.). Existing mechanisms duplicate the entire state, causing hundreds of milliseconds to seconds of latency per C/R, which severely bottlenecks deep search and large-scale fan-outs. This paper observes that subsequent checkpoints in AI agents are highly similar. Therefore, instead of full duplication, a sandbox should only duplicate the changes between consecutive checkpoints (Key Insight). However, it is non-trivial to realize the idea, mainly due to the missing OS supports. This paper proposes a new OS-level abstraction, DeltaState, to enable the change-based transactional C/R for AI agents with two co-designed OS mechanisms. First, DeltaFS enables change-based filesystem C/R by organizing the file states into layers and dynamically freezing the writable layer and inserting a new one during checkpoint, reducing file updates to copy-on-write, and making rollback a simple layer switch. Second, DeltaCR enables change-based process state C/R using incremental dumps, and accelerates rollback by bypassing traditional pipelines to directly fork() from a frozen template process. We then present DeltaBox, a novel agent sandbox achieving millisecond level C/R through the two new mechanisms. Evaluations on SWE-bench and RL micro-benchmarks show DeltaBox completes checkpoint and rollback in millisecond-level latency (14ms and 5ms, respectively), empowering agents to explore substantially more nodes under fixed time budgets.

Zilong Bai, Ruiji Zhang, Linqing Chen et al.

In recent years, large language models(LLMs) have attracted significant attention due to their exceptional performance across a multitude of natural language process tasks, and have been widely applied in various fields. However, the application of large language models in the Intellectual Property (IP) domain is challenging due to the strong need for specialized knowledge, privacy protection, processing of extremely long text in this field. In this technical report, we present for the first time a low-cost, standardized procedure for training IP-oriented LLMs, meeting the unique requirements of the IP domain. Using this standard process, we have trained the PatentGPT series models based on open-source pretrained models. By evaluating them on the open-source IP-oriented benchmark MOZIP, our domain-specific LLMs outperforms GPT-4, indicating the effectiveness of the proposed training procedure and the expertise of the PatentGPT models in the IP domain. Remarkably, our model surpassed GPT-4 on the 2019 China Patent Agent Qualification Examination, scoring 65 and matching human expert levels. Additionally, the PatentGPT model, which utilizes the SMoE architecture, achieves performance comparable to that of GPT-4 in the IP domain and demonstrates a better cost-performance ratio on long-text tasks, potentially serving as an alternative to GPT-4 within the IP domain.

Yinpeng Wu, Yitong Chen, Lixiang Wang et al.

Device-side Large Language Models (LLMs) have witnessed explosive growth, offering higher privacy and availability compared to cloud-side LLMs. During LLM inference, both model weights and user data are valuable, and attackers may even compromise the OS kernel to steal them. ARM TrustZone is the de facto hardware-based isolation technology on mobile devices, used to protect sensitive applications from a compromised OS. However, protecting LLM inference with TrustZone incurs significant overhead due to its inflexible isolation of memory and the NPU. To address these challenges, this paper introduces FlexServe, a fast and secure LLM serving system for mobile devices. It first introduces a Flexible Resource Isolation mechanism to construct Flexible Secure Memory (Flex-Mem) and Flexible Secure NPU (Flex-NPU). Both memory pages and the NPU can be efficiently switched between unprotected and protected modes. Based on these mechanisms, FlexServe designs a fast and secure LLM inference framework within TrustZone's secure world. The LLM-Aware Memory Management and Secure Inference Pipeline are introduced to accelerate inference. A Multi-Model Scheduler is proposed to optimize multi-model workflows. We implement a prototype of FlexServe and compare it with two TrustZone-based strawman designs. The results show that FlexServe achieves an average $10.05\times$ speedup in Time to First Token (TTFT) compared to the strawman, and an average $2.44\times$ TTFT speedup compared to an optimized strawman with pipeline and secure NPU enabled. For multi-model agent workflows, the end-to-end speedup is up to $24.30\times$ and $4.05\times$ compared to the strawman and optimized strawman, respectively.

Yunfei Feng, Xi Zhao, Cheng Zhang et al.

Mobile agents can autonomously complete user-assigned tasks through GUI interactions. However, existing mainstream evaluation benchmarks, such as AndroidWorld, operate by connecting to a system-level Android emulator and provide evaluation signals based on the state of system resources. In real-world mobile-agent scenarios, however, many third-party applications do not expose system-level APIs to determine whether a task has succeeded, leading to a mismatch between benchmarks and real-world usage and making it difficult to evaluate model performance accurately. To address these issues, we propose MobiFlow, an evaluation framework built on tasks drawn from arbitrary third-party applications. Using an efficient graph-construction algorithm based on multi-trajectory fusion, MobiFlow can effectively compress the state space, support dynamic interaction, and better align with real-world third-party application scenarios. MobiFlow covers 20 widely used third-party applications and comprises 240 diverse real-world tasks, with enriched evaluation metrics. Compared with AndroidWorld, MobiFlow's evaluation results show higher alignment with human assessments and can guide the training of future GUI-based models under real workloads.

Siyang Yao, Erhu Feng, Yubin Xia

Hallucination detection in large language models (LLMs) requires balancing accu racy, efficiency, and robustness to distribution shift. Black-box consistency methods are effective but demand repeated inference; single-pass white-box probes are effi cient yet treat answer representations in isolation, often degrading sharply under domain shift. We propose QAOD (Question-Answer Orthogonal Decomposition), a single-pass framework that projects away the question-aligned direction from the answer representation to obtain a question-orthogonal component that suppresses domain-conditioned variation. To identify informative signals, QAOD further selects layers via diversity-penalized Fisher scoring and discriminative neurons via Fisher importance. To address both in-domain detection and cross-domain generalization, we design two complementary probing strategies: pairing the or thogonal component with question context yields a joint probe that maximizes in-domain discriminability, while using the orthogonal component alone preserves domain-agnostic factuality signals for robust transfer. QAOD's joint probe achieves the best in-domain AUROC across all evaluated model-dataset pairs, while the orthogonal-only probe delivers the strongest OOD transfer, surpassing the best white-box baseline by up to 21% on BioASQ at under 25% of generation cost.

Le Chen, Erhu Feng, Yubin Xia et al.

LLM agents increasingly adopt skills as a reusable unit of composition. While skills are shared across diverse agent platforms, current systems treat them as raw context, causing the same skill to behave inconsistently for different agents. This fragility undermines skill portability and execution efficiency. To address this challenge, we analyze 118,000 skills and draw inspiration from traditional compiler design. We treat skills as code and LLMs as heterogeneous processors. To make portability actionable, we decompose a skill's requirements into a set of primitive capabilities, and measure how well each model-harness pair supports them. Based on these capability profiles, we propose SkillRT, a compilation and runtime system designed for portable and efficient skill execution. At compile time, SkillRT performs capability-based compilation, environment binding, and concurrency extraction. At runtime, SkillRT applies JIT code solidification and adaptive recompilation for performance optimization. We evaluate SkillRT across eight LLMs of varying scales and three agent harnesses, covering SkillsBench and representative skill tasks. Results demonstrate that SkillRT significantly improves task completion rates across different models and environments while reducing token consumption by up to 40%. In terms of performance, SkillRT achieves up to 3.2x speedup with enhanced parallelism, and 19-50x latency reduction through code solidification.

Jingkai He, Tianjian Li, Erhu Feng et al.

With the rapid advancement of large language models (LLMs), reinforcement learning (RL) has emerged as a pivotal methodology for enhancing the reasoning capabilities of LLMs. Unlike traditional pre-training approaches, RL encompasses multiple stages: rollout, reward, and training, which necessitates collaboration among various worker types. However, current RL systems continue to grapple with substantial GPU underutilization, due to two primary factors: (1) The rollout stage dominates the overall RL process due to test-time scaling; (2) Imbalances in rollout lengths (within the same batch) result in GPU bubbles. While prior solutions like asynchronous execution and truncation offer partial relief, they may compromise training accuracy for efficiency. Our key insight stems from a previously overlooked observation: rollout responses exhibit remarkable similarity across adjacent training epochs. Based on the insight, we introduce RhymeRL, an LLM RL system designed to accelerate RL training with two key innovations. First, to enhance rollout generation, we present HistoSpec, a speculative decoding inference engine that utilizes the similarity of historical rollout token sequences to obtain accurate drafts. Second, to tackle rollout bubbles, we introduce HistoPipe, a two-tier scheduling strategy that leverages the similarity of historical rollout distributions to balance workload among rollout workers. We have evaluated RhymeRL within a real production environment, demonstrating scalability from dozens to thousands of GPUs. Experimental results demonstrate that RhymeRL achieves a 2.6x performance improvement over existing methods, without compromising accuracy or modifying the RL paradigm.

Le Chen, Dahu Feng, Erhu Feng et al.

With the rapid advancement of artificial intelligence technologies such as ChatGPT, AI agents, and video generation, contemporary mobile systems have begun integrating these AI capabilities on local devices to enhance privacy and reduce response latency. To meet the computational demands of AI tasks, current mobile SoCs are equipped with diverse AI accelerators, including GPUs and Neural Processing Units (NPUs). However, there has not been a comprehensive characterization of these heterogeneous processors, and existing designs typically only leverage a single AI accelerator for LLM inference, leading to suboptimal use of computational resources and memory bandwidth. In this paper, we first summarize key performance characteristics of heterogeneous processors, SoC memory bandwidth, etc. Drawing on these observations, we propose different heterogeneous parallel mechanisms to fully exploit both GPU and NPU computational power and memory bandwidth. We further design a fast synchronization mechanism between heterogeneous processors that leverages the unified memory architecture. By employing these techniques, we present HeteroInfer, the fastest LLM inference engine in mobile devices which supports GPU-NPU heterogeneous execution. Evaluation shows that HeteroInfer delivers a 1.34x to 6.02x end-to-end speedup over state-of-the-art GPU-only and NPU-only LLM engines, while maintaining negligible interference with other applications.

Erhu Feng, Wenbo Zhou, Zibin Liu et al.

AI agents, empowered by Large Language Models (LLMs) and communication protocols such as MCP and A2A, have rapidly evolved from simple chatbots to autonomous entities capable of executing complex, multi-step tasks, demonstrating great potential. However, the LLMs' inherent uncertainty and heavy computational resource requirements pose four significant challenges to the development of safe and efficient agents: reliability, privacy, cost and performance. Existing approaches, like model alignment, workflow constraints and on-device model deployment, can partially alleviate some issues but often with limitations, failing to fundamentally resolve these challenges. This paper proposes a new paradigm called AgentRR (Agent Record & Replay), which introduces the classical record-and-replay mechanism into AI agent frameworks. The core idea is to: 1. Record an agent's interaction trace with its environment and internal decision process during task execution, 2. Summarize this trace into a structured "experience" encapsulating the workflow and constraints, and 3. Replay these experiences in subsequent similar tasks to guide the agent's behavior. We detail a multi-level experience abstraction method and a check function mechanism in AgentRR: the former balances experience specificity and generality, while the latter serves as a trust anchor to ensure completeness and safety during replay. In addition, we explore multiple application modes of AgentRR, including user-recorded task demonstration, large-small model collaboration and privacy-aware agent execution, and envision an experience repository for sharing and reusing knowledge to further reduce deployment cost.

Jiahui Sun, Zhichao Hua, Yubin Xia

Comprehensive evaluation of mobile agents can significantly advance their development and real-world applicability. However, existing benchmarks lack practicality and scalability due to the extensive manual effort in defining task reward signals and implementing evaluation codes. We propose AutoEval, an evaluation framework which tests mobile agents without any manual effort. Our approach designs a UI state change representation which can be used to automatically generate task reward signals, and employs a Judge System for autonomous evaluation. Evaluation shows AutoEval can automatically generate reward signals with high correlation to human-annotated signals, and achieve high accuracy (up to 94%) in autonomous evaluation comparable to human evaluation. Finally, we evaluate state-of-the-art mobile agents using our framework, providing insights into their performance and limitations.

Linqing Chen, Weilei Wang, Yubin Xia et al.

In this paper, we propose a novel system that integrates state-of-the-art, domain-specific large language models with advanced information retrieval techniques to deliver comprehensive and context-aware responses. Our approach facilitates seamless interaction among diverse components, enabling cross-validation of outputs to produce accurate, high-quality responses enriched with relevant data, images, tables, and other modalities. We demonstrate the system's capability to enhance response precision by leveraging a robust question-answering model, significantly improving the quality of dialogue generation. The system provides an accessible platform for real-time, high-fidelity interactions, allowing users to benefit from efficient human-computer interaction, precise retrieval, and simultaneous access to a wide range of literature and data. This dramatically improves the research efficiency of professionals in the biomedical and pharmaceutical domains and facilitates faster, more informed decision-making throughout the R\&D process. Furthermore, the system proposed in this paper is available at https://synapse-chat.patsnap.com.

Cheng Zhang, Erhu Feng, Xi Zhao et al.

With the rapid advancement of Vision-Language Models (VLMs), GUI-based mobile agents have emerged as a key development direction for intelligent mobile systems. However, existing agent models continue to face significant challenges in real-world task execution, particularly in terms of accuracy and efficiency. To address these limitations, we propose MobiAgent, a comprehensive mobile agent system comprising three core components: the MobiMind-series agent models, the AgentRR acceleration framework, and the MobiFlow benchmarking suite. Furthermore, recognizing that the capabilities of current mobile agents are still limited by the availability of high-quality data, we have developed an AI-assisted agile data collection pipeline that significantly reduces the cost of manual annotation. Compared to both general-purpose LLMs and specialized GUI agent models, MobiAgent achieves state-of-the-art performance in real-world mobile scenarios.

Zhenliang Xue, Hanpeng Hu, Xing Chen et al.

Large multimodal models (LMMs) have demonstrated excellent capabilities in both understanding and generation tasks with various modalities. While these models can accept flexible combinations of input data, their training efficiency suffers from two major issues: pipeline stage imbalance caused by heterogeneous model architectures, and training data dynamicity stemming from the diversity of multimodal data. In this paper, we present PipeWeaver, a dynamic pipeline scheduling framework designed for LMM training. The core of PipeWeaver is dynamic interleaved pipeline, which searches for pipeline schedules dynamically tailored to current training batches. PipeWeaver addresses issues of LMM training with two techniques: adaptive modality-aware partitioning and efficient pipeline schedule search within a hierarchical schedule space. Meanwhile, PipeWeaver utilizes SEMU (Step Emulator), a training simulator for multimodal models, for accurate performance estimations, accelerated by spatial-temporal subgraph reuse to improve search efficiency. Experiments show that PipeWeaver can enhance LMM training efficiency by up to 97.3% compared to state-of-the-art systems, and demonstrate excellent adaptivity to LMM training's data dynamicity.

Zibin Liu, Cheng Zhang, Xi Zhao et al.

Large Language Model (LLM) agents are increasingly deployed to automate complex workflows in mobile and desktop environments. However, current model-centric agent architectures struggle to self-evolve post-deployment: improving personalization, capability, and efficiency typically requires continuous model retraining/fine-tuning, which incurs prohibitive computational overheads and suffers from an inherent trade-off between model accuracy and inference efficiency. To enable iterative self-evolution without model retraining, we propose MOBIMEM, a memory-centric agent system. MOBIMEM first introduces three specialized memory primitives to decouple agent evolution from model weights: (1) Profile Memory uses a lightweight distance-graph (DisGraph) structure to align with user preferences, resolving the accuracy-latency trade-off in user profile retrieval; (2) Experience Memory employs multi-level templates to instantiate execution logic for new tasks, ensuring capability generalization; and (3) Action Memory records fine-grained interaction sequences, reducing the reliance on expensive model inference. Building upon this memory architecture, MOBIMEM further integrates a suite of OS-inspired services to orchestrate execution: a scheduler that coordinates parallel sub-task execution and memory operations; an agent record-and-replay (AgentRR) mechanism that enables safe and efficient action reuse; and a context-aware exception handling that ensures graceful recovery from user interruptions and runtime errors. Evaluation on AndroidWorld and top-50 apps shows that MOBIMEM achieves 83.1% profile alignment with 23.83 ms retrieval time (280x faster than GraphRAG baselines), improves task success rates by up to 50.3%, and reduces end-to-end latency by up to 9x on mobile devices.

Tianhao Zhu, Dahu Feng, Erhu Feng et al.

With the widespread adoption of Large Language Models (LLMs), the demand for high-performance LLM inference services continues to grow. To meet this demand, a growing number of AI accelerators have been proposed, such as Google TPU, Huawei NPU, Graphcore IPU, and Cerebras WSE, etc. Most of these accelerators adopt multi-core architectures to achieve enhanced scalability, but lack the flexibility of SIMT architectures. Therefore, without careful configuration of the hardware architecture, as well as deliberate design of tensor parallelism and core placement strategies, computational resources may be underutilized, resulting in suboptimal inference performance. To address these challenges, we first present a multi-level simulation framework with both transaction-level and performance-model-based simulation for multi-core NPUs. Using this simulator, we conduct a systematic analysis and further propose the optimal solutions for tensor parallelism strategies, core placement policies, memory management methods, as well as the selection between PD-disaggregation and PD-fusion on multi-core NPUs. We conduct comprehensive experiments on representative LLMs and various NPU configurations. The evaluation results demonstrate that, our solution can achieve 1.32x-6.03x speedup compared to SOTA designs for multi-core NPUs across different hardware configurations. As for LLM serving, our work offers guidance on designing optimal hardware architectures and serving strategies for multi-core NPUs across various LLM workloads.

Qingyuan Liu, Liyan Chen, Yanning Yang et al.

Large Language Models (LLMs) increasingly require processing long text sequences, but GPU memory limitations force difficult trade-offs between memory capacity and bandwidth. While HBM-based acceleration offers high bandwidth, its capacity remains constrained. Offloading data to host-side DIMMs improves capacity but introduces costly data swapping overhead. We identify that the critical memory bottleneck lies in the decoding phase of multi-head attention (MHA) exclusively, which demands substantial capacity for storing KV caches and high bandwidth for attention computation. Our key insight reveals this operation uniquely aligns with modern DIMM-based processing-in-memory (PIM) architectures, which offers scalability of both capacity and bandwidth. Based on this observation and insight, we propose L3, a hardware-software co-designed system integrating DIMM-PIM and GPU devices. L3 introduces three innovations: First, hardware redesigns resolve data layout mismatches and computational element mismatches in DIMM-PIM, enhancing LLM inference utilization. Second, communication optimization enables hiding the data transfer overhead with the computation. Third, an adaptive scheduler coordinates GPU-DIMM-PIM operations to maximize parallelism between devices. Evaluations using real-world traces show L3 achieves up to 6.1$\times$ speedup over state-of-the-art HBM-PIM solutions while significantly improving batch sizes.

Linqing Chen, Weilei Wang, Zilong Bai et al.

Large language models (LLMs) have revolutionized Natural Language Processing (NLP) by minimizing the need for complex feature engineering. However, the application of LLMs in specialized domains like biopharmaceuticals and chemistry remains largely unexplored. These fields are characterized by intricate terminologies, specialized knowledge, and a high demand for precision areas where general purpose LLMs often fall short. In this study, we introduce PharmaGPT, a suite of domain specilized LLMs with 13 billion and 70 billion parameters, specifically trained on a comprehensive corpus tailored to the Bio-Pharmaceutical and Chemical domains. Our evaluation shows that PharmaGPT surpasses existing general models on specific-domain benchmarks such as NAPLEX, demonstrating its exceptional capability in domain-specific tasks. Remarkably, this performance is achieved with a model that has only a fraction, sometimes just one-tenth-of the parameters of general-purpose large models. This advancement establishes a new benchmark for LLMs in the bio-pharmaceutical and chemical fields, addressing the existing gap in specialized language modeling. It also suggests a promising path for enhanced research and development, paving the way for more precise and effective NLP applications in these areas.

Zhenliang Xue, Yixin Song, Zeyu Mi et al.

Large language models (LLMs) on smartphones enable real-time AI assistance and privacy-preserving, offline operation. However, resource constraints of smartphones limit current deployments to small language models (SLMs), significantly compromising their capabilities. This paper introduces PowerInfer-2, a smartphone-based framework that enables fast inference for LLMs exceeding the memory capacity. The key insight is decomposing matrix operations into neuron clusters as the basic processing unit, which enables flexible scheduling and efficient I/O-computation pipelining. PowerInfer-2 leverages this neuron-cluster-based design in both computation and storage. For computation, neuron clusters with dense activations are processed on NPU, while sparse clusters use CPU. The storage engine provides a fine-grained pipeline mechanism that coordinates cluster-level computation and I/O operations, enhanced by a segmented neuron cache to reduce I/O activities. PowerInfer-2 achieves up to a 27.8x speed increase compared to state-of-the-art frameworks. PowerInfer-2 is the first system to serve a 47B LLM on a smartphone, achieving 11.68 tokens/s. Notably, these performance improvements preserve model quality with negligible accuracy degradation.

Youren Shen, Hongliang Tian, Yu Chen et al.

Intel Software Guard Extensions (SGX) enables user-level code to create private memory regions called enclaves, whose code and data are protected by the CPU from software and hardware attacks outside the enclaves. Recent work introduces library operating systems (LibOSes) to SGX so that legacy applications can run inside enclaves with few or even no modifications. As virtually any non-trivial application demands multiple processes, it is essential for LibOSes to support multitasking. However, none of the existing SGX LibOSes support multitasking both securely and efficiently. This paper presents Occlum, a system that enables secure and efficient multitasking on SGX. We implement the LibOS processes as SFI-Isolated Processes (SIPs). SFI is a software instrumentation technique for sandboxing untrusted modules (called domains). We design a novel SFI scheme named MPX-based, Multi-Domain SFI (MMDSFI) and leverage MMDSFI to enforce the isolation of SIPs. We also design an independent verifier to ensure the security guarantees of MMDSFI. With SIPs safely sharing the single address space of an enclave, the LibOS can implement multitasking efficiently. The Occlum LibOS outperforms the state-of-the-art SGX LibOS on multitasking-heavy workloads by up to 6,600X on micro-benchmarks and up to 500X on application benchmarks.

Yubin Xia, Qingyuan Liu, Cheng Tan et al.

Decentralized Internet is booming. People are fascinated by its promise that users can truly own their data. However, in a decentralized Internet, completing a task usually involves multiple nodes with mutual distrust. Such distrust might eventually become a major obstacle for the growth of the decentralized Internet. In this paper, we analyze the distrust using a simple model and highlight the properties required to faithfully accomplish one task in a decentralized Internet. We also introduce our draft solution -- PIXIU, a framework to mitigate the distrust among different nodes. In PIXIU, we design and utilize trust-λ and decentralized executor to achieve the above-needed properties.