论文库

Rapidly evolving LLM architectures and heterogeneous mobile hardware make supporting new models in on-device inference engines a recurring and costly system challenge. Each model may require coordinated changes across conversion, graph construction, operators, and hardware backends. Existing coding agents often miss these cross-layer dependencies and struggle to diagnose failures across host--device execution paths. We present EngineAgent, an agentic system for automating on-device LLM model enablement. Its key insight is to separate reusable engine and hardware knowledge from model-specific implementation. Before enablement, EngineAgent constructs a hierarchy-aware engine index that captures engine abstraction layers and cross-layer dependencies, and pre-validates backend-specific build, execution, and debugging workflows using already supported models. For each new model, a builder uses this knowledge to coordinate cross-layer changes, while an auditor checks implementation completeness and drives iterative debugging through device-side validation. Across 12 tasks spanning three mobile LLMs and four engine--backend pairs on CPUs, GPUs, and NPUs, EngineAgent achieves 75% end-to-end success with either of two underlying LLMs, versus 33.3% for the strongest coding-agent baseline. Success requires compilation, on-device execution, and output-logit agreement with reference implementations. EngineAgent also reduces average per-task time and model-API cost relative to baselines using the same underlying LLMs.

Local LLM agents often work on shared hosts. Unlike multi-tenant cloud clusters, these hosts often have tightly bounded memory, CPU, and I/O resources. Concurrent agents frequently suffer from resource contention, leading to failures such as OOM and reduced resource efficiency. Considering that agents' resource demands are bursty and shell command delay is imperceptible to agents, there is a huge scheduling opportunity to arrange when and in what order commands are actually executed. Thus, we present ToolSched, a Shell MCP Server for shared local hosts that serves multiple agents and schedules their operations. ToolSched embeds a resource model into the MCP interface: agents estimate resource demands along with command requests. These hints, together with real-time host resource conditions, determine when commands are actually executed. Internally, ToolSched applies a scheduling algorithm to stagger commands, reducing contention and improving resource utilization. Prediction correction and checkpoint-based recovery mitigate estimation bias and resource errors. Our evaluation on Terminal Bench shows that, compared with unscheduled execution, ToolSched improves resource utilization by 148.4% and end-to-end command throughput by 29.4%, demonstrating its effectiveness and practicality.

Modern LLMs strain device memory through large model weights and attention states, particularly in resource-constrained scenarios. Existing systems separately provision memory for model weights and attention states, imposing fixed fixed boundaries that cannot adapt to the changing memory demands of workloads. Under heavy workloads, insufficient state capacity limits request concurrency and reduces throughput; under light workloads, idle memory in one region cannot be reallocated to another, wasting device memory. Meanwhile, modern LLM architectures introduce heterogeneous tensors with diverse layouts and lifetimes, further complicating memory management. These problems motivate unified memory management for heterogeneous tensors. However, a shared pool alone does not make memory directly reusable across tensors with different allocation granularities and execution layouts. This paper presents Rimuru, a modern LLM serving system that unifies memory management through elastic expert offloading. Rimuru establishes a shared page pool to enables elastic memory sharing across heterogeneous tensors. To accommodate different allocation granularities, Rimuru introduces memory views that allow each tensor type to interpret the shared memory space using its own page size and page table. Furthermore, Rimuru combines huge-page alignment with a multi-level page allocator that splits and merges pages, enabling cross-type capacity reuse. A page-table-aware fused MoE kernel directly accesses non-contiguous expert weights, avoiding copy overhead. Building on these mechanisms, Rimuru employs a workload-aware adjuster that dynamically reallocates pages across tensor types according to runtime demands. Implemented on top of vLLM, Rimuru remains compatible with modern LLMs. Evaluations show that Rimuru achieves up to 19.03x baseline throughput and reduces TTFT by 98.1% on average.