Daily Papers · 2026-10-10
📄 论文日报 | 来源：Hugging Face Daily Papers
🗓 Episode：2026.10.09 | 代码世界实时渲染进化智能体；LLM文本游戏经验学习有限
1. AgentGarten：面向进化智能体的代码世界
- 英文标题：AgentGarten: Code Worlds for Evolving Agents
- arXiv：2610.12374
💡 一句话解读：这篇研究要解决的是：智能体在虚拟世界里学习时，环境既要规则一致、又要画面逼真，很难两全。作者提出 AgentGarten 框架，把模拟器和游戏引擎与一个共享的神经渲染器结合，实时生成逼真画面，并让智能体把每轮经验写成“攻略”传给下一代，从而只需 4 轮就能学会，远快于传统强化学习。
交互式虚拟世界让智能体能够通过探索和交互来学习。智能体能学到什么受限于其所练习的环境，而这些环境必须既忠实（状态、规则和动态一致）又真实（观测符合真实世界的视觉分布）。在多样化的世界中同时做到这两点仍是一个瓶颈。我们提出 AgentGarten，一个将模拟器和游戏引擎与共享神经渲染器耦合起来以构建实时交互环境的框架。其模拟后端维护持久的世界状态并执行程序定义的交互规则，而渲染器则根据通过通用接口导出的结构化条件生成视觉观测。为构建该神经渲染器，我们将预训练视频模型适配到几何条件上，用我们提出的对抗强制（Adversarial Forcing）对其进行蒸馏，并针对实时交互优化推理。对抗强制通过精确重放使历史预填充可微，从而让后续预测的损失能够更新渲染器对先前观测的编码方式，并加入真实数据对抗监督以提升视觉质量。在 AgentGarten 中，智能体通过视觉观测感知世界，实时与之交互，并通过将每轮经验蒸馏成可供后续智能体继承和改进的攻略来提升自身。我们的实证研究表明学习效率大幅提升，智能体仅需 4 轮学习，而传统强化学习对应方法需要数百万轮。由于新世界可以写成代码并通过同一接口渲染，环境的数量和难度可以与其智能体一同扩展，这是朝着通过交互经验持续进化的智能体迈出的一步。
📄 English Abstract
Interactive virtual worlds allow agents to learn through exploration and interaction. What agents can learn is bounded by the environments they practice in, which must be faithful, with consistent state, rules, and dynamics, and realistic, with observations that follow the real-world visual distributions. Achieving both across diverse worlds remains a bottleneck. We introduce AgentGarten, a framework that couples simulators and game engines with a shared neural renderer to build real-time interactive environments. Its simulation backends maintain persistent world state and execute program-defined interaction rules, while the renderer generates visual observations from structured conditions exported through a common interface. To build the neural renderer, we adapt a pretrained video model to geometry conditions, distill it with our proposed Adversarial Forcing, and optimize inference for real-time interaction. Adversarial Forcing makes history prefilling differentiable through exact replay, so that losses on later predictions update how the renderer encodes prior observations, and adds real-data adversarial supervision to improve its visual quality. In AgentGarten, agents perceive the world through visual observations, interact with it in real time, and improve by distilling each round of experience into playbooks that subsequent agents inherit and refine. Our empirical study demonstrates a substantial gain in learning efficiency, with agents learning from just 4 rounds compared with millions for a conventional reinforcement learning counterpart. As new worlds can be written as code and rendered through the same interface, environments can scale in both number and difficulty alongside their agents, a step toward agents that keep evolving through interactive experience.
2. Learn2Play Bench：LLM 智能体在陌生环境中从经验中学习的效果如何？
- 英文标题：Learn2Play Bench: How Well Do LLM Agents Learn from Experience in Unfamiliar Environments?
- arXiv：2610.08215
💡 一句话解读：现有测试分不清大模型是真从经验里学了新东西，还是靠原有知识推理。为此，作者提出 Learn2Play 基准，让智能体在规则完全陌生的环境里通过互动学习，从而单独考察其从经验中学习的能力。
从经验中学习对于 LLM 智能体适应陌生且动态的环境至关重要。因此，评估这一能力对于理解智能体如何有效地获取和使用新知识十分重要。现有基准虽试图评估该能力，但它们主要评估的是规则已在指令中给出、或对预训练模型而言已熟悉的任务，因而难以区分智能体是从交互中学习，还是仅凭已有知识进行推理。为解决这一问题，我们提出了 Learn2Play。
📄 English Abstract
Learning from experience is essential for LLM agents to adapt to unfamiliar and dynmaic environments. Evaluating this ability is therefore important for understanding how effectively agents acquire and use new knowledge. Existing benchmarks have sought to evaluate this ability, but they primarily evaluate tasks whose rules are provided in the instructions or already familiar to pretrained models, making it difficult to distinguish learning from interactions from reasoning with existing knowledge. To address this, we introduce Learn2Play
3. TokenRouter：面向 Token 级 LLM 路由的高效服务系统
- 英文标题：TokenRouter: Efficient Serving System for Token-Level LLM Routing
- arXiv：2610.12242
💡 一句话解读：大模型路由能按需把任务分给不同模型，但细到每个词（token）级别时，现有系统会卡顿、延迟高。TokenRouter 让开发者只写单个请求的路由逻辑，系统自动为每个模型开子服务并异步调度，还用数学公式算出最佳批处理参数，吞吐量提升 2~64 倍。
大语言模型（LLM）路由将推理工作分配到不同模型上，从而推进 LLM 服务的成本-质量帕累托前沿。尽管在会话或查询级别的粗粒度路由已在生产系统中广泛采用，但近期算法工作表明，细粒度的 token 级路由能带来显著的效率与质量收益。然而，高效服务 token 级路由推理对现有系统构成了重大挑战：当前系统基于单 LLM 假设，在 token 级路由下会遭遇严重的步调不同步和频繁的批处理准入延迟，并且给开发者带来很高的实现复杂度。为解决这些挑战，我们设计了 TokenRouter，一个高效且对开发者友好的 token 级路由 LLM 推理服务系统。TokenRouter 遵循“以请求为中心编程、以模型为中心执行”的原则：开发者从单个请求的角度描述路由逻辑，而运行时为每个 LLM 启动一个子服务器并异步分派请求。每个子服务器采用延迟批处理调度器，其最优超参数由系统的数学吞吐量模型推导得出。在多种路由算法、工作负载和模型组合下，TokenRouter 的解码吞吐量比现有系统高 2.01–64.15 倍，大幅提升了 token 级 LLM 路由的服务效率。我们的代码可在 https://github.com/thu-nics/TokenRouter 获取。
📄 English Abstract
Large language model (LLM) routing distributes inference work across different models, advancing the cost-quality Pareto frontier of LLM serving. While coarse-grained routing at the session or query level has been widely adopted in production systems, recent algorithmic work shows that fine-grained token-level routing can yield substantial efficiency and quality gains. However, efficiently serving token-level routed inference poses significant challenges to existing systems. Built on single-LLM assumptions, current systems suffer from severe step desynchronization and frequent batch admission delays under token-level routing, and they also impose high implementation complexity on developers. To address these challenges, we design TokenRouter, an efficient and developer-friendly serving system for token-level routed LLM inference. TokenRouter follows the principle of request-centric programming, model-centric execution: developers describe routing logic from the perspective of a single request, while the runtime launches a subserver for each LLM and dispatches requests asynchronously. Each subserver employs a delayed-batching scheduler, whose optimal hyperparameters are derived from a mathematical throughput model of the system. Across diverse routing algorithms, workloads, and model pairs, TokenRouter achieves 2.01-64.15x higher decoding throughput than existing systems, substantially advancing the serving efficiency of token-level LLM routing. Our code is available at https://github.com/thu-nics/TokenRouter.
4. 从轨迹到智能体世界：用于交互式环境模拟的智能体语言世界模型
- 英文标题：From Traces to Agentic Worlds: Agentic Language World Models for Interactive Environment Simulation
- arXiv：2610.06100
💡 一句话解读：这篇研究要解决的问题是：训练和测试AI智能体时，往往需要逼真的模拟环境，但原始系统可能无法访问或难以重建。它提出的方法是：不重建可执行环境，而是让一个“世界模型智能体”充当环境，通过历史交互记录构建可复用的环境知识库，在运行时推断每个动作的观察结果和状态影响。
逼真的环境副本对于训练和评估LLM智能体（LLM agents）越来越有价值，但原始系统可能无法访问或难以复现。我们探索了智能体语言世界建模（agentic language world modeling）：世界模型智能体（world model agent）不重建可执行环境，而是作为任务智能体（task agent）的环境，支持忠实且有状态的模拟。我们用Trace2Env实例化了这一范式，这是一个无需学习的框架，适用于原始系统不可用但历史交互轨迹（interaction traces）仍可访问的场景。Trace2Env将这些轨迹重建为可复用的环境世界书（worldbook），其中包含环境模式（environment schemas）、有依据的证据（grounded evidence）和归纳的行为知识（induced behavioral knowledge）。在运行时，世界模型智能体主动查阅世界书以及持久的情节状态（episodic state），以推断每个动作的观察结果和持久状态影响。在九个环境中，Trace2Env在下一观察保真度（next-observation fidelity）和长时程交互一致性（long-horizon interaction consistency）上均优于传统的基于提示的LWM。在多轮交互中，针对Trace2Env生成的任务智能体动作在真实环境中重放时更常保持有效，这表明其模拟动态能更好地在连续轮次中保留先前动作的后果。这些结果确立了智能体语言世界建模作为构建逼真环境副本的替代方向，而无需重建原始可执行系统。
📄 English Abstract
Realistic environment replicas are increasingly valuable for training and evaluating LLM agents, yet the original systems may be inaccessible or impractical to reproduce. We explore agentic language world modeling: rather than rebuilding an executable environment, a world model agent serves as the environment for a task agent and supports faithful and stateful simulation. We instantiate this paradigm with Trace2Env, a learning-free framework for settings where the original system is unavailable but historical interaction traces remain accessible. Trace2Env reconstructs these traces into a reusable environment worldbook containing environment schemas, grounded evidence, and induced behavioral knowledge. At runtime, the world model agent actively consults the worldbook together with persistent episodic state to infer each action's observation and lasting state effects. Across nine environments, Trace2Env improves both next-observation fidelity and long-horizon interaction consistency over conventional prompt-based LWMs. In multi-turn interaction, task agent actions generated against Trace2Env remain valid more often when replayed in the real environment, indicating that its simulated dynamics better preserve the consequences of earlier actions across successive turns. These results establish agentic language world modeling as an alternative direction for building realistic environment replicas without reconstructing the original executable system.
5. SuperNav：面向任意场景中任意任务的智能体导航系统
- 英文标题：SuperNav: An Agentic Navigation System for Any Task in Any Scene
- arXiv：2610.12126
💡 一句话解读：通用服务机器人需要能在陌生环境里听懂各种指令并导航，但现有方法用导航数据微调大模型，导致遇到新指令或新环境就不好使。SuperNav 让大模型只负责理解指令、看懂场景和做决策，把具体移动交给导航工具执行，从而保住大模型的通用能力。
通用服务机器人需要能在陌生环境中处理多样化的人类请求，同时兼顾任务通用性和场景通用性。现有的一些方法通过微调多模态大语言模型（MLLM）来预测导航动作，这使其行为依赖于导航训练数据的覆盖范围，可能限制其对新请求和新环境的泛化能力。我们的核心思路是让 MLLM 专注于解读请求、理解场景和做出决策，同时保留其通用能力，并将运动执行委托给导航工具。为实现这一想法，我们提出了 SuperNav。
📄 English Abstract
General-purpose service robots need navigation systems that can handle diverse human requests in unfamiliar environments, combining task generality with scene generality. Some existing methods fine-tune multimodal large language models (MLLMs) to predict navigation actions, making their behavior dependent on the coverage of navigation training data and potentially limiting generalization to new requests and environments. Our key insight is to let the MLLM focus on interpreting requests, understanding scenes, and making decisions while preserving its general-purpose capabilities and delegating motion execution to navigation tools. To realize this idea, we introduce SuperNav
6. MiMo-V2.6：将强化学习扩展至自我改进
- 英文标题：MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement
- arXiv：2610.11959
💡 一句话解读：这篇研究想解决的是：如何让大型AI模型通过强化学习不断自我提升，而不是只靠人工喂数据。做法是把强化学习的计算规模大幅扩大——用更大批次、更多样复杂的任务环境、更精细的评分机制，同时冻结部分网络并防止模型“钻空子”作弊，从而让模型在长任务中表现更准、更省字。
强化学习（RL）是推动大型基础模型走向自我改进的核心训练范式。本报告介绍 MiMo-V2.6 系列，这是一个全模态模型家族，通过扩展强化学习算力来推进模型智能的前沿。在强化学习之前，我们在广泛的多模态语料上进行中期训练，以提供充足的探索空间，并在预训练的混合 SWA 架构上构建坚实基础，以支持后续的规模扩展。我们从三个维度扩展强化学习算力：（1）更大的批次和更高的吞吐量，采用异步训练，每步消耗 1,568 个样本和 2.7–3.7B 个 token，上下文长度最高可达 1M；（2）更多样、更复杂的环境，在多种智能体框架的混合下覆盖代码、通用、视觉和网络领域；（3）更多的评分算力，通过分组式智能体评分，为长时程任务提供更准确的奖励信号，并引导模型给出更短、更节省 token 的解决方案。为了在规模化下保持训练稳定，我们冻结了 MoE 路由器，并建立了多层防御以对抗奖励黑客行为。我们还为混合任务智能体强化学习构建了基础设施，包括统一的轨迹表示、高并发多框架 rollout、解耦的控制平面与数据平面，以及训练-推理一致性。我们开源了训练动态、强化学习环境和强化学习框架，以促进复现以及关于规模化强化学习和模型自我改进的进一步研究。
📄 English Abstract
Reinforcement learning (RL) is the central training paradigm for advancing large foundation models towards self-improvement. This report introduces the MiMo-V2.6 series, an omni-modal family that pushes the frontier of model intelligence by scaling RL compute. Prior to RL, we conduct mid-training on a broad multimodal corpus to provide ample exploration space, and build a solid infrastructure on the pretrained hybrid-SWA architecture to support subsequent scale-up. We scale RL compute along three dimensions: (1) larger batches and higher throughput, with an asynchronous training that consumes 1,568 samples and 2.7-3.7B tokens per step at context lengths of up to 1M; (2) more diverse and complex environments, spanning code, general, visual, and cyber domains under a mixture of agent harnesses; and (3) more grader compute, via groupwise agentic grading that yields more accurate reward signals for long-horizon tasks and steers the model towards shorter, more token-efficient solutions. To keep training stable at scale, we freeze the MoE router and establish a multi-layer defense against reward hacking. We further build infrastructure for mixed-task agentic RL, including a unified trajectory representation, high-concurrency multi-framework rollout, decoupled control and data planes, and training-inference consistency. We open-source the training dynamics, RL environments, and RL framework to facilitate reproduction and further research on scaled RL and model self-improvement.
7. 让机器人上下文学习变简单：面向操作任务的普惠方案
- 英文标题：In-context Robot Learning Made Simple: A Democratized Recipe for Manipulation Tasks
- arXiv：2609.38173
💡 一句话解读：机器人看一遍演示就能学会做事，但演示里同时包含动作、物体含义、空间关系等信息，机器人到底该学哪部分并不清楚。本文先把这个任务定义清楚，再提出一个简单、易复现的框架（含视觉提示编码器和低成本数据采集方法），不用大规模预训练也能在仿真和真实环境中表现良好。
我们研究机器人上下文学习（ICL），这是一种新兴范式，能让机器人从视觉演示中推断并执行任务。尽管前景广阔，该问题本身仍缺乏明确定义：一个视觉演示同时传达了动作轨迹、物体语义、操作可供性（affordance）、空间关系和任务目标，使得机器人究竟应遵循哪些信息并不清楚。在本工作中，我们首先给出了机器人 ICL 的清晰问题定义，明确其学习目标并解决了这一根本性的提示歧义。基于该定义，我们开发了一个极简且可复现的 ICL 框架（SimpleICL），包含视觉提示编码器和低成本数据采集协议。无需大规模预训练或专门的数据基础设施，我们的框架在仿真和真实环境中均取得了强劲性能。大量实验进一步揭示了机器人 ICL 的几个关键特性，包括语义区分和任务相关解耦。我们将完全开源数据和训练流程，以促进机器人 ICL 的系统性和可复现研究。项目页面见 https://simpleicl.github.io/simpleicl。
📄 English Abstract
We study robotic in-context learning (ICL), an emerging paradigm that enables robots to infer and execute tasks from visual demonstrations. Despite its growing promise, the problem itself remains under-defined: a visual demonstration simultaneously conveys action trajectories, object semantics, manipulation affordances, spatial relations, and task goals, making it unclear what information the robot is actually expected to follow. In this work, we first provide a clear problem definition of robot ICL that explicitly defines its learning target and resolves this fundamental prompt ambiguity. Building on this definition, we develop a minimalist and reproducible ICL framework (SimpleICL) with a visual prompt encoder and a low-cost data collection protocol. Without massive pre-training or specialized data infrastructure, our framework achieves strong performance in both simulation and real-world environments. Extensive experiments further reveal several key properties of robot ICL, including semantic discrimination and task-relevant disentanglement. We will fully open-source our data and training pipeline to facilitate systematic and reproducible research on robot ICL. The project page can be found at https://simpleicl.github.io/simpleicl.
8. 面向细粒度具身交互的多智能体自我中心世界模型
- 英文标题：Multi-Agent Egocentric World Model with Fine-Grained Embodied Interaction
- arXiv：2610.12299
💡 一句话解读：这篇研究要解决的是：现有世界模型大多只模拟单个智能体，而现实中多个智能体需要在同一环境里进行精细互动，比如用手操作物体。作者提出ME-World，让多个智能体各自的第一人称视角视频能同步生成，并保证它们看到的环境和互动结果一致。
以自我为中心的世界模型根据智能体的动作预测第一人称观测，但大多只关注单个智能体。真实的具身场景通常涉及多个智能体在共享环境中行动和交互。现有多智能体世界模型依赖粗粒度动作（如移动、相机控制或离散指令），细粒度具身交互仍未得到充分探索。我们将多智能体自我中心世界建模定义为：多个智能体在共享世界中通过细粒度动作交互时，同步生成各自的自我流。这需要跨视角动作一致性、共享环境一致性，以及交互引发的状态更新的一致性传播。我们提出多智能体自我中心世界模型（ME-World），它在共享令牌序列中联合去噪多个自我流，以所有智能体的目标视角位姿为每个流提供条件，并利用共享环境记忆来支撑生成。我们在真实和合成多智能体数据上训练和评估，并引入共享世界一致性指标，用于衡量环境、更新和身份一致性。实验表明，ME-World在共享世界一致性、动作控制、身份保持和视频质量上均优于现有方法。
📄 English Abstract
Egocentric world models predict first-person observations conditioned on an agent's actions, but most focus on a single agent. Real embodied settings often involve multiple agents that act and interact within a shared environment. Existing multi-agent world models rely on coarse actions like locomotion, camera control, or discrete commands, leaving fine-grained embodied interactions underexplored. We formulate multi-agent egocentric world modeling as synchronized ego-stream generation for multiple agents interacting through fine-grained actions in a shared world. This requires cross-view action consistency, shared-environment consistency, and consistent propagation of interaction-induced state updates. We propose Multi-agent Egocentric World Model (ME-World), which jointly denoises multiple ego streams in a shared token sequence, conditions each stream on all agents' target-view poses, and grounds generation with shared environment memory. We train and evaluate on real and synthetic multi-agent data and introduce shared-world consistency metrics for environment, update, and identity consistency. Experiments show ME-World improves shared-world consistency, action control, identity preservation, and video quality over existing methods.
9. DreamTrue：基于反事实后训练的动作忠实机器人世界模型
- 英文标题：DreamTrue: Action-Faithful Robot World Model with Counterfactual Post-Training
- arXiv：2610.12468
💡 一句话解读：机器人要预测自己动作后会发生什么，但现有数据标定不准、失败案例太少，导致预测不听话、总往成功方向编。DreamTrue 把动作画进画面里对齐，并造出各种“如果那样做会怎样”的假想场景来训练，再用人类打分教模型分辨物理上是否合理。结果动作跟随达到最好，交互出错率从48%降到6%。
我们提出 DreamTrue，一个多视角、跨本体的机器人世界模型，用于动作忠实且物理合理的视频预测。在现有机器人数据集上训练此类模型面临两个障碍：标定不精确会损害动作跟随能力，而对失败交互覆盖有限则会使预测偏向成功结果。为提升跨本体的动作跟随，我们将动作轨迹渲染为图像空间条件，并引入离线几何标定使这些条件与目标视频对齐。为扩大交互覆盖，我们引入反事实后训练，修改记录的动作轨迹并在更广泛的动作与接触配置下生成未来视频。为在缺乏配对真实未来视频的情况下对这些预测提供反馈，我们构建了一个人工标注视频数据集，涵盖机器人、物体和交互缺陷，并用其训练具身视频奖励模型。其评分引导强化学习后训练朝向更物理合理的交互结果。在 AgiBot 上，DreamTrue 达到了最先进的动作跟随性能，同时将人工评估的交互缺陷率从 48.12% 降至 6.25%。值得注意的是，我们的模型在 AgiBot World Challenge 2026 的世界模型赛道中排名第一。
📄 English Abstract
We present DreamTrue, a multi-view, cross-embodiment robot world model for action-faithful and physically plausible video prediction. Training such a model on existing robot datasets faces two obstacles: imprecise calibration can impair action following, while limited coverage of unsuccessful interactions can bias predictions toward successful outcomes. To improve action following across embodiments, we render action trajectories into image-space conditions and introduce offline geometric calibration to align these conditions with the target videos. To broaden interaction coverage, we introduce counterfactual post-training, modifying recorded action trajectories and generating future videos under a wider range of actions and contact configurations. To provide feedback on these predictions without paired ground-truth futures, we construct a human-annotated video dataset covering robot, object, and interaction defects and use it to train an embodied video reward model. Its scores guide reinforcement-learning post-training toward more physically plausible interaction outcomes. On AgiBot, DreamTrue attains state-of-the-art action following, while reducing the human-assessed interaction defect rate from from 48.12% to 6.25%. Notably, our model ranks first in the world model track of the AgiBot World Challenge 2026. The project page can be found at https://brave-eai.github.io/DreamTrue.
10. OuroWorld：将任意3D世界变为多样且无限循环的3D动态影像
- 英文标题：OuroWorld: Bringing Any 3D World Alive as Diverse, Endlessly Looping 3D Cinemagraphs
- arXiv：2610.12461
💡 一句话解读：静态3D场景无法动起来，现有方法只能做流体般运动。OuroWorld用视觉语言模型推断合理动态，生成参考视频并扩展为多视角，再通过抗不一致的周期变形场让任意3D场景变成无缝循环的动态影像。
近期的3D世界模型能生成逼真、可探索的场景，但这些场景在时间上是冻结的。OuroWorld是一个无需掩码（mask-free）的框架，可将任何静态3D高斯泼溅（3D Gaussian Splatting）场景转变为3D动态影像（3D cinemagraph）：一种从任意视角观看都能无缝循环、具有生动多样运动的动态场景。一个视觉语言模型（vision-language model）推断出合理的动态，并引导视频模型合成参考视频，我们将其提升并补全为多视角视频。为了从不完美的监督中学习，我们提出了不一致鲁棒的周期4DGS（Inconsistency-Robust Periodic 4DGS）：一个傅里叶级数变形场通过构造保证循环性，而一个锚定在参考视图的接地漂移场（Grounded Drift Field）吸收跨视图不一致性。与先前仅限于类流体运动的欧拉方法不同，我们能够捕捉一般变形、物体运动和光照变化。我们引入了一种无需真实标注的评估方法，涵盖生动性、自然性、循环接缝连贯性和场景质量。在39个重建和生成的场景上，OuroWorld优于所有基线，并在70.8%至99.0%的用户研究比较中获胜。项目页面：this https URL
📄 English Abstract
Recent 3D world models generate photorealistic, explorable scenes that remain frozen in time. OuroWorld is a mask-free framework that turns any static 3D Gaussian Splatting scene into a 3D cinemagraph: a dynamic scene with vivid, diverse motion looping seamlessly from any viewpoint. A vision-language model infers plausible dynamics and guides a video model to synthesize a reference video, which we lift and complete into multi-view videos. To learn from this imperfect supervision, we propose Inconsistency-Robust Periodic 4DGS: a Fourier-series deformation field guarantees looping by construction, while a Grounded Drift Field anchored at the reference view absorbs cross-view inconsistency. Unlike prior Eulerian methods limited to fluid-like motion, we capture general deformation, object motion, and illumination change. We introduce a ground-truth-free evaluation covering vividness, naturalness, loop seam coherence, and scene quality. On 39 reconstructed and generated scenes, OuroWorld outperforms all baselines and wins 70.8%-99.0% of user-study comparisons. Project page: this https URL
11. MC-Sparse：解构并弥合扩散 Transformer 中的稠密-稀疏注意力差距
- 英文标题：MC-Sparse: Deconstructing and Closing the Dense-Sparse Attention Gap in Diffusion Transformers
- arXiv：2610.06801
💡 一句话解读：扩散Transformer生成视频或3D内容时，注意力计算太慢。现有稀疏注意力方法在高稀疏度下会降低生成质量。本文分析出三个原因：分组限制、选错交互、丢弃信息。提出MC-Sparse，不训练模型，通过缓存查询分组、精确选出的键值索引和稠密-稀疏残差，在去噪步骤中复用，实现更高保真度和加速。
稀疏注意力是降低扩散Transformer在长序列生成任务（如视频和高分辨率3D资产生成）中延迟的主要方法。然而，现有方法在高稀疏度下会降低生成质量和保真度。通过受控的预言机比较，我们将这种退化归因于三个来源：令牌分组带来的约束、不准确的交互选择以及丢弃令牌时损失的注意力贡献。基于此分析，我们提出了元缓存稀疏注意力（MC-Sparse），这是一个无需训练的框架，它选择单个键值（KV）令牌，同时将相似的查询组织成瓦片对齐的组以实现高效的GPU执行。MC-Sparse缓存元数据，包括查询组、使用精确注意力概率选择的KV索引以及稠密与稀疏注意力输出之间的残差，并在后续去噪步骤中重用它们。在视频和3D生成模型上，MC-Sparse比现有稀疏注意力基线实现了更高的稠密注意力输出保真度和更大的去噪加速，且无明显质量下降。相对于稠密注意力，它在Minimax-H3-Base上实现了1.80倍的去噪加速，在3D资产生成上实现了2.32倍的加速，两者质量损失均可忽略不计。
📄 English Abstract
Sparse attention is a primary approach to reducing the latency of diffusion transformers in long-sequence generation tasks, such as video and high-resolution 3D asset generation. However, existing methods can degrade generation quality and fidelity at high sparsity levels. Through controlled oracle comparisons, we trace this degradation to three sources: constraints imposed by token grouping, inaccurate interaction selection, and the attention contributions lost when tokens are discarded. Guided by this analysis, we propose Meta-Cached Sparse Attention (MC-Sparse), a training-free framework that selects individual key-value (KV) tokens while organizing similar queries into tile-aligned groups for efficient GPU execution. MC-Sparse caches metadata comprising query groups, KV indices selected using exact attention probabilities, and residuals between dense and sparse attention outputs, and reuses them across subsequent denoising steps. Across video and 3D generation models, MC-Sparse achieves higher fidelity to dense-attention outputs and larger denoising speedups than existing sparse-attention baselines, without visible quality degradation. Relative to dense attention, it delivers a $1.80\times$ denoising speedup on Minimax-H3-Base and a $2.32\times$ speedup on 3D asset generation, both with negligible quality loss.
12. 超越时空先验：面向稠密对应匹配的可泛化方法
- 英文标题：Beyond Spatio-Temporal Priors: A Generalizable Approach for Dense Correspondence Matching
- arXiv：2610.12421
💡 一句话解读：图像编辑等任务中，画面可能被改得面目全非却仍保留同一身份，而传统匹配方法依赖“运动平滑、物体刚硬”等假设，遇到这种变换就失效。本文提出 FreeMatching，融合生成与语义基础表示，并用多种数据监督训练，无需密集标注即可提升匹配质量。
稠密对应匹配长期以来受限于简化的时空先验（如运动平滑和刚性几何）。这些假设对经典任务有效，但在图像编辑与参考引导生成（IEG）中会失效，因为此类变换可在保持视觉身份的同时打破物理连续性。为在此类变换间建立保持身份的对应关系，我们提出 FreeMatching——一个可泛化框架，它结合生成式与语义基础表示，并利用来自经典数据集、跟踪视频和合成场景的异构监督。教师引导的迭代细化进一步提升了 IEG 中的对应匹配，且无需稠密对应标注。实验表明，单个 FreeMatching 模型在具有挑战性的 IEG 图像对上大幅提升了匹配质量，同时在经典基准上保持有竞争力的性能。此外，我们展示了其作为评估身份保持的定量指标的实用性，其得分与人类判断相关。代码见 https://github.com/luping-liu/FreeMatching。
📄 English Abstract
Dense correspondence matching has historically been bounded by simplifying spatio-temporal priors, such as smooth motion and rigid geometry. While effective for classical tasks, these assumptions break down in image editing and reference-guided generation (IEG), where transformations can preserve visual identity while breaking physical continuity. To establish identity-preserving correspondence across such transformations, we introduce FreeMatching, a generalizable framework combining generative and semantic foundation representations with heterogeneous supervision from classical datasets, tracked videos, and synthetic scenes. Teacher-guided iterative refinement further improves correspondence in IEG without dense correspondence annotations. Experimentally, a single FreeMatching model substantially improves correspondence quality on challenging IEG image pairs while retaining competitive performance on classical benchmarks. Furthermore, we demonstrate its utility as a quantitative metric for evaluating identity preservation, with scores that correlate with human judgment. The code is available at https://github.com/luping-liu/FreeMatching.
13. TestPrism：重新思考超越单一参考的测试评估
- 英文标题：TestPrism: Rethinking Test Evaluation Beyond a Single Reference
- arXiv：2610.12289
💡 一句话解读：现有评估AI生成测试的方法只用一份标准答案来检验，会漏掉其他正确写法、高估测试质量。作者构建了TestPrism数据集（含多个正确与错误实现），要求测试能同时接受所有正确解、拒绝所有错误解，并据此提出TestHelix方法，通过合成测试与修复对、交叉验证和自我改进来提升测试质量。
大语言模型（LLM）编程智能体在各类编程任务中的测试生成能力已有提升。然而，用单一参考解来评估测试的常见做法忽视了其他有效的实现方式，并可能高估测试质量。我们提出TestPrism，包含来自17个来源的300个测试任务和3000个候选实现，有效解与无效解各占一半。其核心指标“联合成功函数”（Joint Success Function）要求生成的测试在初始程序状态下失败、接受每一个有效候选实现，并拒绝每一个无效候选实现。在十四种基线编程智能体配置中，联合成功函数仅达到28.00%，而单一参考解成功率则达59.67%。我们的分析揭示了遗漏行为、不支持的断言以及测试构造缺陷等问题。为解决这些弱点，我们提出TestHelix，它将测试与修复对的异构合成、同伴交叉验证以及递归自我改进（RSI）相结合。在两个模型上，TestHelix在TestHelix评估中将联合成功函数较原生测试框架对照提升了8.67至9.00个百分点。
📄 English Abstract
Large language model (LLM) coding agents have advanced test generation across diverse programming tasks. However, the common practice of evaluating tests against a single reference solution overlooks alternative valid implementations and can overstate test quality. We introduce TestPrism, comprising 300 test tasks from 17 sources and 3000 candidate implementations, evenly split between valid and invalid solutions. Its primary metric, Joint Success Function, requires the generated tests to fail on the initial program state, accept every valid candidate, and reject every invalid candidate. Across fourteen baseline coding agent configurations, Joint Success Function reaches only 28.00%, whereas single reference success reaches 59.67%. Our analysis reveals missed behaviors, unsupported assertions, and faulty test construction. To address these weaknesses, we introduce TestHelix, which combines heterogeneous synthesis of test and repair pairs with peer cross validation and recursive self improvement (RSI). Across two models, TestHelix improves Joint Success Function by 8.67 to 9.00 percentage points over the native harness comparators in the TestHelix evaluation
14. 通过组合偏好奖励与评分规则奖励对前沿文本到图像模型进行后训练
- 英文标题：Post-Training Frontier Text-to-Image Models by Composing Preference and Rubric Rewards
- arXiv：2610.02967
💡 一句话解读：这篇研究想解决的是：文生图模型虽然画得漂亮，但很难通过后续训练让它更符合人的喜好，因为单一的评分标准无法全面反映人类偏好。作者的做法是把两种互补的奖励信号组合起来——一种学人类整体审美偏好，另一种按规则检查是否忠实于提示词，并设计了一种比简单加权平均更有效的组合方式，训练后模型在竞技场排行榜上大幅提升。
近期的文本到图像生成模型已取得出色的视觉质量，但通过后训练（post-training）进一步提升仍然困难，因为没有任何单一奖励信号能够涵盖人类偏好的全部范围。在本工作中，我们为开放域文本到图像生成开发了一套简单而有效的后训练方案，其基础是互补奖励信号的组合。我们的奖励系统包含两个主要部分：一是偏好奖励（preference reward），使用 Bradley-Terry 目标在大规模人类偏好数据上训练，以捕捉整体的人类审美与感知偏好；二是基于评分规则（rubric-based）的奖励，显式评估提示词忠实度及其他理想属性，同时为防止奖励被钻空子（reward hacking）提供保障。一个关键挑战是如何组合这些异构的奖励信号。我们表明，朴素的加权平均会导致次优的优化行为，并提出了一种简单的奖励组合策略，能更有效地在偏好优化与评分规则满足之间取得平衡。在 Arena 文本到图像排行榜（https://arena.ai/）上，我们经强化学习训练的 Flux2dev 比基础模型高出 69 个 Elo 分，而后训练的 Ideogram-4 超越了排行榜上所有开源模型，达到 1223.5 的 Elo 分。（关于最先进性能的说法基于截至 2026 年 9 月 4 日的 Arena 排行榜快照。）我们的结果表明，前沿生成模型训练所需的有效奖励，既要广泛覆盖用户意图，又要在优化过程中对钻空子行为具有鲁棒性。为支持可复现研究，我们发布了 Arena-T2I-Training，这是训练数据的 1K 子集，可复现全量训练的部分收益，我们希望这一资源能促进未来文本到图像模型后训练的研究。
📄 English Abstract
Recent text-to-image generation models have achieved remarkable visual quality, but improving them through post-training remains challenging because no single reward signal captures the full range of human preference. In this work, we develop a simple and effective post-training recipe for open-domain text-to-image generation based on the composition of complementary reward signals. Our reward system consists of two main components: a preference reward, trained on large-scale human preference data using a Bradley-Terry objective to capture overall human aesthetic and perceptual preferences, and rubric-based rewards, which explicitly evaluate prompt faithfulness and other desirable properties while providing safeguards against reward hacking. A key challenge is how to combine these heterogeneous reward signals. We show that a naive weighted average leads to suboptimal optimization behavior, and propose a simple reward composition strategy that more effectively balances preference optimization with rubric satisfaction. In the Arena text-to-image leaderboard (https://arena.ai/), our RL-trained Flux2dev achieves an Elo rating 69 points above the base model, and our post-trained Ideogram-4 surpasses every open-source model on the leaderboard, reaching an Elo of 1223.5. (Claims of state-of-the-art performance are based on the Arena leaderboard snapshot as of September 4, 2026.) Our results suggest that effective rewards for frontier generative-model training require broad coverage of user intent and robustness to exploitation under optimization. To support reproducible research, we release Arena-T2I-Training, a 1K subset of training data that recovers some gains of full-scale training, providing a resource that we hope will facilitate future work on post-training for text-to-image models.
15. SparseDecoding：面向准确高效大语言模型推理的解码感知剪枝
- 英文标题：SparseDecoding: Decoding-Aware Pruning for Accurate and Efficient LLM Inference
- arXiv：2610.12327
💡 一句话解读：大模型生成文字时速度慢，因为要从内存反复读取大量参数。现有剪枝方法用自然文本算重要性，但模型实际生成时输入的是自己产生的词，两者分布不一致，导致剪枝后效果变差；而且这些方法主要优化矩阵乘法，对生成时更关键的矩阵-向量乘法支持不足。本文提出SparseDecoding，让剪枝过程感知解码时的分布，并针对解码操作优化，从而在加速的同时保持准确率。
大语言模型（LLM）推理的解码阶段受内存带宽限制，导致显著延迟。基于Hessian的逐层免训练网络剪枝是解决该问题的主流方案，因为剪枝减少了解码时从内存读取的非零参数数量。然而，这类方法通常使用预先收集的自然序列计算Hessian，而解码时模型输入的是自生成的token，两者之间存在分布偏移。在自然序列上计算的Hessian与在生成序列上计算的Hessian不同。我们观察到这种差异导致生成过程中的激活分布偏离剪枝时使用的分布，进一步损害剪枝后模型的性能。此外，大多数能带来实际加速的现有LLM剪枝方法主要针对稀疏矩阵-矩阵（SpMM）乘法，对主导解码的稀疏矩阵-向量（SpMV）操作支持有限。为了解决这些问题，我们提出了SparseDecoding。
📄 English Abstract
The memory-bound nature of the decoding stage of large language model (LLM) inference incurs significant latency. Layer-wise training-free network pruning approaches guided by the Hessian have been a prominent solution to this problem, as pruning reduces the number of nonzero parameters read from memory during decoding. Nevertheless, typical methods in this line compute the Hessian using pre-collected natural sequences, whereas the model is fed self-generated tokens during decoding, creating a distribution shift between the two sequences. The Hessian calculated on the natural sequence is different from that calculated on the generated sequence. We observe that this discrepancy causes the activation distribution during generation to deviate from that used for pruning, further hurting the pruned model performance. Moreover, most existing LLM pruning methods that bring actual speedup primarily target the sparse matrix-matrix (SpMM) multiplication, providing limited support for the sparse matrix-vector (SpMV) operations, which dominate decoding. To solve these problems, we introduce SparseDecoding