arXiv subjects

Yanyu Ren

Publications and source records attributed to Yanyu Ren.

12 recordsLinked to original sources

arXiv2026

TideRL: Boosting Agentic RL Goodput with Readiness-Aware Scheduling

Reinforcement learning (RL) for large language models is moving toward multi-turn agentic workloads, where rollout tasks repeatedly pause for external environments, resume with growing contexts, and finish at highly variable times. In this setting, RL training goodput, measured by training throughput, matters more than raw GPU occupancy: GPU waiting and repeated prefill recomputation are pure overhead. We present TideRL, a readiness-aware elastic RL system with Continuous Task Batching, Resource-Aware Ref-Actor Pipelining, and Elastic Resource Scaling. CTB preserves useful rollout state, $\textrm{RA}^2\textrm{P}$ selects between decoupled streaming and colocated aggregation from the ready backlog and arrival interval, and ERS moves ranks between rollout and training using the same readiness signals. Across text-only and multi-modal agentic workloads, TideRL improves RL training goodput by up to 5.6$\times$ over synchronous baselines and over 33% over asynchronous baselines, while reaching similar task performance. It also improves KV cache hit rate by 1.58$\times$, reduces per-step training time by up to 44.3%, and cuts total waiting time by up to 77.6%.

cs.LG↗

arXiv2026

An Empirical Study of Coordination Mode as the First-Class Citizen in From-Scratch Multi-Agent Coding

Multi-agent vibe coding promises to accelerate software development, yet existing benchmarks rely on synthetic environments that ignore practical time and monetary costs, conflate reasoning with communication, and reward only superficial completion. We introduce multi-agent from-scratch evaluation benchmark, MSEval, evaluating multi-agent coding on real-world tasks. Grounded in 10 authentic, full-stack projects across 10 domains, MSEval scores performance using hierarchical requirements and deterministic rubrics. Its execution engine, LegoGent, tests 10 collaboration topologies where agents coordinate via periodic sync intervals and deploy through native CI/CD pipelines. Concurrently, the automated grader TAgent dynamically probes implementations to jointly measure functional success, latency, and prefix-cached token cost. Across 100 runs, MSEval reveals that organizational topology rivals model capability in shaping the speed--cost--quality trade-off. For identical tasks and models, varying the topology shifts scores by over 30 points and doubles wall-clock time. Structured pipelines converge fastest with the highest quality, whereas heavy managerial oversight degrades performance. Ultimately, MSEval establishes a rigorous, reproducible standard for measuring how multi-agent teams actually build software. The benchmark is released at https://github.com/robinren03/MSEval.

cs.AI↗

arXiv2026

SCALECUA: Scaling Computer Use Agents with Verifiable Task Synthesis and Efficient Online RL

Computer use agents (CUAs) are emerging as a powerful interface for automating complex digital workflows through visual perception and GUI execution. Online reinforcement learning with verifiable rewards (RLVR) has emerged as a key direction for scaling their capabilities. However, this paradigm is bottlenecked by verifiable data scarcity and online RL inefficiency. To break these barriers, we introduce ScaleCUA, a unified framework that scales online RL for CUAs via verifiable task synthesis and efficient training. At the data level, we design VeriGen, an end-to-end framework for generating verifiable RL tasks through iterative docker interactions and a multi-agent feedback loop. Scaled to 100+ concurrent agent workers via a shared docker interaction probe, this pipeline produces 24K+ verifiable tasks and nearly 3K high-quality RL tasks. To maximize sample efficiency, we propose Frontier Sampling, which tracks per-task capability and allocates rollouts to the current learning frontier. On the training side, we further design Visual Context Segmentation, a sliding window over recent visual context that balances rollout and training-engine pressure, yielding a 2.83x training speedup over step-wise decomposition. Together, ScaleCUA achieves 68.7% on OSWorld and 54.0% on ScienceBoard, establishing new state-of-the-art performance among open-source computer use agents. Code, models, and datasets are available at https://github.com/THUDM/SCALE-CUA.

cs.AI↗

arXiv2025

Communication-Efficient Serving for Video Diffusion Models with Latent Parallelism

Video diffusion models (VDMs) perform attention computation over the 3D spatio-temporal domain. Compared to large language models (LLMs) processing 1D sequences, their memory consumption scales cubically, necessitating parallel serving across multiple GPUs. Traditional parallelism strategies partition the computational graph, requiring frequent high-dimensional activation transfers that create severe communication bottlenecks. To tackle this issue, we exploit the local spatio-temporal dependencies inherent in the diffusion denoising process and propose Latent Parallelism (LP), the first parallelism strategy tailored for VDM serving. \textcolor{black}{LP decomposes the global denoising problem into parallelizable sub-problems by dynamically rotating the partitioning dimensions (temporal, height, and width) within the compact latent space across diffusion timesteps, substantially reducing the communication overhead compared to prevailing parallelism strategies.} To ensure generation quality, we design a patch-aligned overlapping partition strategy that matches partition boundaries with visual patches and a position-aware latent reconstruction mechanism for smooth stitching. Experiments on three benchmarks demonstrate that LP reduces communication overhead by up to 97\% over baseline methods while maintaining comparable generation quality. As a non-intrusive plug-in paradigm, LP can be seamlessly integrated with existing parallelism strategies, enabling efficient and scalable video generation services.

cs.DC↗

arXiv2025

DMA: Online RAG Alignment with Human Feedback

Retrieval-augmented generation (RAG) systems often rely on static retrieval, limiting adaptation to evolving intent and content drift. We introduce Dynamic Memory Alignment (DMA), an online learning framework that systematically incorporates multi-granularity human feedback to align ranking in interactive settings. DMA organizes document-, list-, and response-level signals into a coherent learning pipeline: supervised training for pointwise and listwise rankers, policy optimization driven by response-level preferences, and knowledge distillation into a lightweight scorer for low-latency serving. Throughout this paper, memory refers to the model's working memory, which is the entire context visible to the LLM for In-Context Learning. We adopt a dual-track evaluation protocol mirroring deployment: (i) large-scale online A/B ablations to isolate the utility of each feedback source, and (ii) few-shot offline tests on knowledge-intensive benchmarks. Online, a multi-month industrial deployment further shows substantial improvements in human engagement. Offline, DMA preserves competitive foundational retrieval while yielding notable gains on conversational QA (TriviaQA, HotpotQA). Taken together, these results position DMA as a principled approach to feedback-driven, real-time adaptation in RAG without sacrificing baseline capability.

cs.AI↗

arXiv2025

ComputerRL: Scaling End-to-End Online Reinforcement Learning for Computer Use Agents

We introduce ComputerRL, a framework for autonomous desktop intelligence that enables agents to operate complex digital workspaces skillfully. ComputerRL features the API-GUI paradigm, which unifies programmatic API calls and direct GUI interaction to address the inherent mismatch between machine agents and human-centric desktop environments. Scaling end-to-end RL training is crucial for improvement and generalization across diverse desktop tasks; however, it remains challenging due to environmental inefficiency and instability during extended training. To support scalable and robust training, we develop a distributed RL infrastructure capable of orchestrating thousands of parallel virtual desktop environments to accelerate large-scale online RL. Furthermore, we propose Entropulse, a training strategy that alternates reinforcement learning with supervised fine-tuning, effectively mitigating entropy collapse during extended training runs. We employ ComputerRL on open models GLM-4-9B-0414 and GLM-4.1V-9B-Thinking, and evaluate them on the OSWorld benchmark. The AutoGLM-OS-9B achieves a new state-of-the-art accuracy of 48.9%, demonstrating significant improvements for general agents in desktop automation. Our code and the new OfficeWorld benchmark are available at https://github.com/thudm/ComputerRL. The algorithm and framework are adopted in building AutoGLM (Liu et al., 2024b).

cs.AI↗

arXiv2025

DeepFaith: A Domain-Free and Model-Agnostic Unified Framework for Highly Faithful Explanations

Explainable AI (XAI) builds trust in complex systems through model attribution methods that reveal the decision rationale. However, due to the absence of a unified optimal explanation, existing XAI methods lack a ground truth for objective evaluation and optimization. To address this issue, we propose Deep architecture-based Faith explainer (DeepFaith), a domain-free and model-agnostic unified explanation framework under the lens of faithfulness. By establishing a unified formulation for multiple widely used and well-validated faithfulness metrics, we derive an optimal explanation objective whose solution simultaneously achieves optimal faithfulness across these metrics, thereby providing a ground truth from a theoretical perspective. We design an explainer learning framework that leverages multiple existing explanation methods, applies deduplicating and filtering to construct high-quality supervised explanation signals, and optimizes both pattern consistency loss and local correlation to train a faithful explainer. Once trained, DeepFaith can generate highly faithful explanations through a single forward pass without accessing the model being explained. On 12 diverse explanation tasks spanning 6 models and 6 datasets, DeepFaith achieves the highest overall faithfulness across 10 metrics compared to all baseline methods, highlighting its effectiveness and cross-domain generalizability.

cs.LG↗

arXiv2025

Studies of Directed Flow with Event Plane Method in the HIRFL-CSR External-target Experiment

The Cooling-Storage-Ring External-target Experiment (CEE) at Heavy Ion Research Facility in Lanzhou (HIRFL) is designed to study the properties of nuclear matter created in heavy-ion collisions at a few hundred MeV/$u$ to 1 GeV/$u$ beam energies, facilitating the research of quantum chromodynamics phase structure in the high-baryon-density region. Collective flow is one of the most important observables in heavy-ion collision experiments to study the bulk behavior of the created matter. Even though the standard event plane method has been widely used for collective flow measurements, it remains crucial to validate and optimize this method for the CEE spectrometer. In this paper, we study the experimental procedures of measuring directed flow in $^{238}$U+$^{238}$U collisions at 500 MeV/$u$ using event planes reconstructed by Multi Wire Drift Chamber and Zero Degree Calorimeter, respectively. Jet AA Microscopic (JAM) transport generator is used to generate events, and the detector response is simulated by the CEE Fast Simulation (CFS) package. Finally, the optimal kinetic region for proton directed flow measurements is discussed for the future CEE experiment.

physics.ins-det↗

arXiv2024

Pistis-RAG: Enhancing Retrieval-Augmented Generation with Human Feedback

RAG systems face limitations when semantic relevance alone does not guarantee improved generation quality. This issue becomes particularly evident due to the sensitivity of large language models (LLMs) to the ordering of few-shot prompts, which can affect model performance. To address this challenge, aligning LLM outputs with human preferences using structured feedback, such as options to copy, regenerate, or dislike, offers a promising method for improvement. This feedback is applied to the entire list of inputs rather than giving specific ratings for individual documents, making it a Listwide Labels Learning-to-Rank task. To address this task, we propose Pistis-RAG, a new RAG framework designed with a content-centric approach to better align LLMs with human preferences. Pistis-RAG effectively utilizes human feedback, enhancing content ranking and generation quality. To validate our framework, we use public datasets to simulate human feedback, allowing us to evaluate and refine our method effectively. Experimental results indicate that Pistis-RAG improves alignment with human preferences relative to the baseline RAG system, showing a 6.06% increase in MMLU (English) and a 7.08% increase in C-EVAL (Chinese) accuracy metrics. These results highlight Pistis-RAG's effectiveness in overcoming the limitations associated with traditional RAG approaches.

cs.IR↗

arXiv2023

An Empirical Study of NetOps Capability of Pre-Trained Large Language Models

Nowadays, the versatile capabilities of Pre-trained Large Language Models (LLMs) have attracted much attention from the industry. However, some vertical domains are more interested in the in-domain capabilities of LLMs. For the Networks domain, we present NetEval, an evaluation set for measuring the comprehensive capabilities of LLMs in Network Operations (NetOps). NetEval is designed for evaluating the commonsense knowledge and inference ability in NetOps in a multi-lingual context. NetEval consists of 5,732 questions about NetOps, covering five different sub-domains of NetOps. With NetEval, we systematically evaluate the NetOps capability of 26 publicly available LLMs. The results show that only GPT-4 can achieve a performance competitive to humans. However, some open models like LLaMA 2 demonstrate significant potential.

cs.CL↗

arXiv2020

The effect of single-particle space-momentum angle distribution on two-pion HBT correlation in high-energy heavy-ion collisions

Using several source models, we analyze the transverse momentum dependence of HBT radii in the relativistic heavy-ion collisions. The results indicate that the single-particle space-momentum angle distribution plays an important role in the transverse momentum dependence of HBT radii. In a cylinder source, we use several formulas to describe the transverse momentum dependence of HBT radii and the single-particle space-momentum angle distribution. We also make a numerical connection between them in the transverse plane.

nucl-th↗

arXiv2018

Analysis of the image of pion-emitting sources in source center of mass frame

In this paper, we try a method to extract the image of pion-emitting source function in the center-of-mass frame of source (CMFS). We choose the identical pion pairs according to the difference of their energy and use these pion pairs to build the correlation function. The purpose is to reduce the effect of $\triangle E \triangle t$, thus the corresponding imaging result can tend to the real source function. We examine the effect of this method by comparing its results with real source functions extracted from models directly.

nucl-th↗