Searcharxiv⌕ Search

arXiv subjects

Weili Xu

Publications and source records attributed to Weili Xu.

12 recordsLinked to original sources

Evolving Infra: Agent-Adaptive LLM Serving at Scale

Agentic workloads are a major source of LLM inference demand. Existing prefill--decode (PD) disaggregated serving systems offer limited flexibility for runtime adaptation of compute allocation and KV cache management, and rely on coarse signals that only partly capture the diverse and dynamic agent workload. These limitations lead to load imbalance between prefill and decode nodes and eviction of KV caches that agents will reuse. We present Evolving Infra, an agent-adaptive serving framework that adapts compute allocation to agent workloads and KV cache management to agent context management at runtime. We decouple compute allocation from key--value (KV) cache management, allowing nodes to switch efficiently between prefill and decode. Building on this design, an online planner matches the PD split to workload composition, while an agent-aware distributed KV manager guides cache eviction and transfer with signals from agent context management. Across coding, math, and computer-use tasks with GLM-5.2 and Kimi K2.6 on up to 256 GB300 GPUs, Evolving Infra reduces mean end-to-end request latency and P95 TTFT by up to 4.1x and 14x over Dynamo, NVIDIA's production framework. We open-source the system implementation of Evolving Infra at https://github.com/mit-han-lab/EvolvingInfra.

cs.DC↗

InteractionBench: A Real-Time Interaction Benchmark for Streaming Video Systems

A video assistant must speak when its instruction warrants a response and stay silent otherwise. We introduce a benchmark that evaluates this decision for the complete system of model, memory, and response controller. InteractionBench covers query responses, event triggers, and ongoing updates in 1,060 interactions over 812 videos, with 69 negative streams and 53 suites that pair counted events with look-alike near misses. It scores content accuracy, timing accuracy, and silence compliance on the video clock. Timely speech costs silence across systems. Polled Qwen3-VL-8B reaches 77.8 timing accuracy but 10.9 silence compliance. A native real-time interaction system reaches 29.2 silence compliance at 66.8 timing accuracy, yet emits on 89.9% of negative streams. No open-weight system clears a third of the near-miss suites. Fewer replies help only when chosen, as random deletion merely trades timing for silence. Offline scores miss these failures and mispredict online behavior. Adding restraint is costly, as the native system's controller adds little by itself and agentic systems add it only at about 30 s per poll.Project page: https://www.enxinsong.com/projects/interactionbench/ Code: https://github.com/Espere-1119-Song/InteractionBench Data: https://huggingface.co/datasets/InteractionBench/InteractionBench

cs.CV↗

QATFactory: A Versatile, Deployment-Aligned Framework for Quantization-aware Training and Distillation of LLMs

Large language model (LLM) inference is increasingly moving toward lower precision to realize the throughput of hardware accelerators, but aggressive post-training quantization (PTQ) can degrade model quality. We present QATFactory, an open-source framework for deployment-aligned quantization-aware distillation (QAD) and reinforcement learning (QARL). QATFactory simulates deployment-time quantization while performing matrix multiplications in BF16, allowing models to adapt to quantization noise without requiring training hardware that natively supports the target format; for example, it supports NVFP4 training on H100 GPUs, which lack FP4 Tensor Cores. The framework supports NVFP4, MXFP4, and $\text{llama}.\text{cpp}$'s Q4_K format; dense and mixture-of-experts models; and both full-parameter and LoRA-based training. It exports checkpoints directly to vLLM and $\text{llama}.\text{cpp}$ without an additional lossy conversion step or added inference overhead. With QATFactory, we conduct extensive experiments on models ranging from 8B to 230B parameters and evaluate exported checkpoints in production inference engines. Across models and formats, QAD consistently improves deployed-model quality over strong PTQ baselines. On Qwen3.5-9B, QAD achieves average benchmark accuracies of 68.9% under NVFP4 and 66.0% under MXFP4, outperforming the best PTQ results of 65.4% and 56.4%, respectively. Through our experiments, we found that although both FP4 formats quantize weights and activations at deployment, the best training strategy is format-dependent: NVFP4 generally performs better when only weights are quantized during training, whereas MXFP4 benefits from quantizing both weights and activations. At a fixed training token budget, training on fewer 32K sequences improves average accuracy by 1.9 points over training on more 4K sequences.

cs.LG↗

ThunderAgent: A Simple, Fast and Program-Aware Agentic Inference System

Large language models(LLMs) are now used to power complex multi-turn agentic workflows. Existing systems run agentic inference by loosely assembling isolated components: an LLM inference engine (e.g., vLLM) and a tool orchestrator (e.g., Kubernetes). Although agentic workflows involve multiple LLM and tool requests, these systems schedule and allocate resources separately on a per-request basis, without end-to-end knowledge of the workflow. This leads to sub-optimal management of KV cache and tool execution environments. To address the challenges, we propose ThunderAgent, a fast, simple, and program-aware agentic inference system. We first abstract agentic workflows as LLM Programs, enabling a unified view of heterogeneous resources, including KV caches, system states, and external tool assets such as disk memory and network ports. Built upon this abstraction, ThunderAgent introduces a program-aware scheduler and a tool resource manager designed to maximize KV cache hit rates, mitigate memory imbalances, and enable asynchronous environment preparation. Evaluations across coding, routing, and scientific discovery agents demonstrate that ThunderAgent achieves 1.5-3.6x throughput improvements in serving, 1.8-3.9x in RL rollout, and up to 4.2x disk memory savings compared to state-of-the-art inference systems. To facilitate reproducibility and support future development, we open-source the system implementations of the whole ThunderAgent at: https://github.com/Agentic-Kinetics/ThunderAgent.

cs.OS↗

$M^3-Verse$: A "Spot the Difference" Challenge for Large Multimodal Models

Modern Large Multimodal Models (LMMs) have demonstrated extraordinary ability in static image and single-state spatial-temporal understanding. However, their capacity to comprehend the dynamic changes of objects within a shared spatial context between two distinct video observations, remains largely unexplored. This ability to reason about transformations within a consistent environment is particularly crucial for advancements in the field of spatial intelligence. In this paper, we introduce $M^3-Verse$, a Multi-Modal, Multi-State, Multi-Dimensional benchmark, to formally evaluate this capability. It is built upon paired videos that provide multi-perspective observations of an indoor scene before and after a state change. The benchmark contains a total of 270 scenes and 2,932 questions, which are categorized into over 50 subtasks that probe 4 core capabilities. We evaluate 16 state-of-the-art LMMs and observe their limitations in tracking state transitions. To address these challenges, we further propose a simple yet effective baseline that achieves significant performance improvements in multi-state perception. $M^3-Verse$ thus provides a challenging new testbed to catalyze the development of next-generation models with a more holistic understanding of our dynamic visual world. You can get the construction pipeline from https://github.com/Wal-K-aWay/M3-Verse_pipeline and full benchmark data from https://www.modelscope.cn/datasets/WalKaWay/M3-Verse.

cs.CV↗

MeMix: Writing Less, Remembering More for Streaming 3D Reconstruction

Reconstruction is a fundamental task in 3D vision and a fundamental capability for spatial intelligence. Particularly, streaming 3D reconstruction is central to real-time spatial perception, yet existing recurrent online models often suffer from progressive degradation on long sequences due to state drift and forgetting, motivating inference-time remedies. We present MeMix, a training-free, plug-and-play module that improves streaming reconstruction by recasting the recurrent state into a Memory Mixture. MeMix partitions the state into multiple independent memory patches and updates only the least-aligned memory patches while exactly preserving others. This selective update mitigates catastrophic forgetting while retaining $O(1)$ inference memory, and requires no fine-tuning or additional learnable parameters, making it directly applicable to existing recurrent reconstruction models. Across standard benchmarks (ScanNet, 7-Scenes, KITTI, etc.), under identical backbones and inference settings, MeMix reduces reconstruction completeness error by 15.3% on average (up to 40.0%) across 300--500 frame streams on 7-Scenes. The code is available at https://dongjiacheng06.github.io/MeMix/

cs.CV↗

OptAgent: an Agentic AI framework for Intelligent Building Operations

The urgent need for building decarbonization calls for a paradigm shift in future autonomous building energy operation, from human-intensive engineering workflows toward intelligent agents that interact with physics-grounded digital environments. This study proposes an end-to-end agentic AI-enabled Physics-Informed Machine Learning (PIML) environment for scalable building energy modeling, simulation, control, and automation. The framework consists of (1) a modular and physics-consistent PIML digital environment spanning building thermal dynamics, Heating, Ventilation, and Air Conditioning (HVAC), and distributed energy resources (DER) for grid-interactive energy management; and (2) an agentic AI layer with 11 specialist agents and 72 Model Context Protocol (MCP) tools that enable end-to-end execution of multi-step energy analytics. A representative case study demonstrates multi-domain, multi-agent coordination for assessing how system and control upgrades affect energy use, operating cost, thermal comfort, and flexibility. In addition, a large-scale benchmark (about 4000 runs) systematically evaluates workflow performance in terms of accuracy, token consumption, execution time, and inference cost. The results quantify the impacts of intelligence mode design, model size, task complexity, and orchestrator-specialist coordination, and provide key lessons for building future agentic AI systems in real-world building energy applications. This work establishes a scalable, physics-grounded foundation for deploying agentic AI in decarbonized and grid-interactive building operations.

eess.SY↗

Automatic Building Code Review: A Case Study

Building officials, particularly those in resource-constrained or rural jurisdictions, face labor-intensive, error-prone, and costly manual reviews of design documents as projects increase in size and complexity. The growing adoption of Building Information Modeling (BIM) and Large Language Models (LLMs) presents opportunities for automated code review (ACR) solutions. This study introduces a novel agent-driven framework that integrates BIM-based data extraction with automated verification using both retrieval-augmented generation (RAG) and Model Context Protocol (MCP) agent pipelines. The framework employs LLM-enabled agents to extract geometry, schedules, and system attributes from heterogeneous file types, which are then processed for building code checking through two complementary mechanisms: (1) direct API calls to the US Department of Energy COMcheck engine, providing deterministic and audit-ready outputs, and (2) RAG-based reasoning over rule provisions, enabling flexible interpretation where coverage is incomplete or ambiguous. The framework was evaluated through case demonstrations, including automated extraction of geometric attributes (such as surface area, tilt, and insulation values), parsing of operational schedules, and validation of lighting allowances under ASHRAE Standard 90.1-2022. Comparative performance tests across multiple LLMs showed that GPT-4o achieved the best balance of efficiency and stability, while smaller models exhibited inconsistencies or failures. Results confirm that MCP agent pipelines outperform RAG reasoning pipelines in rigor and reliability. This work advances ACR research by demonstrating a scalable, interoperable, and production-ready approach that bridges BIM with authoritative code review tools.

cs.SE↗

AuroraLong: Bringing RNNs Back to Efficient Open-Ended Video Understanding

The challenge of long video understanding lies in its high computational complexity and prohibitive memory cost, since the memory and computation required by transformer-based LLMs scale quadratically with input sequence length. We propose AuroraLong to address this challenge by replacing the LLM component in MLLMs with a linear RNN language model that handles input sequence of arbitrary length with constant-size hidden states. To further increase throughput and efficiency, we combine visual token merge with linear RNN models by reordering the visual tokens by their sizes in ascending order. Despite having only 2B parameters and being trained exclusively on public data, AuroraLong achieves performance comparable to Transformer-based models of similar size trained on private datasets across multiple video benchmarks. This demonstrates the potential of efficient, linear RNNs to democratize long video understanding by lowering its computational entry barrier. To our best knowledge, we are the first to use a linear RNN based LLM backbone in a LLaVA-like model for open-ended video understanding.

cs.CV↗

Exploring Gen-AI applications in building research and industry: A review

This paper investigates the transformative potential of Generative AI (Gen-AI) technologies, particularly large language models, within the building industry. By leveraging these advanced AI tools, the study explores their application across key areas such as automated compliance checking and building design assistance. The research highlights how Gen-AI can automate labor-intensive processes, significantly improving efficiency and reducing costs in building practices. The paper first discusses the two widely applied fundamental models-Transformer and Diffusion model-and summarizes current pathways for accessing Gen-AI models and the most common techniques for customizing them. It then explores applications for text generation, such as compliance checking, control support, data mining, and building simulation input file editing. Additionally, it examines image generation, including direct generation through diffusion models and indirect generation through language model-supported template creation based on existing Computer-Aided Design or other design tools with rendering. The paper concludes with a comprehensive analysis of the current capabilities of Gen-AI in the building industry, outlining future directions for research and development, with the goal of paving the way for smarter, more effective, and responsive design, construction, and operational practices.

cs.AI↗

Video-MMLU: A Massive Multi-Discipline Lecture Understanding Benchmark

Recent advancements in language multimodal models (LMMs) for video have demonstrated their potential for understanding video content, yet the task of comprehending multi-discipline lectures remains largely unexplored. We introduce Video-MMLU, a massive benchmark designed to evaluate the capabilities of LMMs in understanding Multi-Discipline Lectures. We evaluate over 90 open-source and proprietary models, ranging from 0.5B to 40B parameters. Our results highlight the limitations of current models in addressing the cognitive challenges presented by these lectures, especially in tasks requiring both perception and reasoning. Additionally, we explore how the number of visual tokens and the large language models influence performance, offering insights into the interplay between multimodal perception and reasoning in lecture comprehension.

cs.CV↗

GRID: a Student Project to Monitor the Transient Gamma-Ray Sky in the Multi-Messenger Astronomy Era

The Gamma-Ray Integrated Detectors (GRID) is a space mission concept dedicated to monitoring the transient gamma-ray sky in the energy range from 10 keV to 2 MeV using scintillation detectors onboard CubeSats in low Earth orbits. The primary targets of GRID are the gamma-ray bursts (GRBs) in the local universe. The scientific goal of GRID is, in synergy with ground-based gravitational wave (GW) detectors such as LIGO and VIRGO, to accumulate a sample of GRBs associated with the merger of two compact stars and study jets and related physics of those objects. It also involves observing and studying other gamma-ray transients such as long GRBs, soft gamma-ray repeaters, terrestrial gamma-ray flashes, and solar flares. With multiple CubeSats in various orbits, GRID is unaffected by the Earth occultation and serves as a full-time and all-sky monitor. Assuming a horizon of 200 Mpc for ground-based GW detectors, we expect to see a few associated GW-GRB events per year. With about 10 CubeSats in operation, GRID is capable of localizing a faint GRB like 170817A with a 90% error radius of about 10 degrees, through triangulation and flux modulation. GRID is proposed and developed by students, with considerable contribution from undergraduate students, and will remain operated as a student project in the future. The current GRID collaboration involves more than 20 institutes and keeps growing. On August 29th, the first GRID detector onboard a CubeSat was launched into a Sun-synchronous orbit and is currently under test.

astro-ph.IM↗