论文

介绍 Gemini for Google (GfG)，一款面向 Google 内部软件工程生态的 Gemini 专用适配版本，涵盖从构建万亿 token 的专有数据集到采用可缓解灾难性遗忘的中段训练策略的完整过程。G Gemini for Google (GfG)}, an adaptation of Gemini specialized for Google's internal software engineering ecosystem, is introduced, from curating a trillion-token proprietary dataset to implementing a mid-training strategy that mitigates catastrophic forgetting.

对八款主流开源 MLOps 框架的实践使用与功能增强需求进行调查，结果显示 MLOps 框架很少被直接开箱即用，也较少集成进 GitHub Workflows，开发者更多通过其 API 在项目中实现自定义功能。Investigating the practical use and desired feature enhancements of eight popular open-source MLOps frameworks indicates that users mainly ask for enhancements to core features of the frameworks, but also better API exposure and CI/CD integration.

实验表明，DataEvolver 显著提升了数据质量，相比在原始数据上训练，下游 LLM 性能平均提升 10%，凸显了 LLM 与数据迭代协同演化的新机遇。Experiments show that DataEvolver substantially improves data quality and achieves an average 10\% gain in downstream LLM performance compared with training on original data, highlighting new opportunities for the iterative co-evolution of LLMs and data.

本文提出 LoRAFusion，一种面向 LLM 的高效 LoRA 微调系统，可消除不必要的内存访问，在不付出重算或同步代价的前提下保持 compute-bound GEMM 的性能，并引入面向多任务微调的自适应批处理算法。LoRAFusion is introduced, an efficient LoRA fine-tuning system for LLMs that eliminates unnecessary memory accesses and preserves the performance of compute-bound GEMMs without incurring the cost of recomputation or synchronization and introduces an adaptive batching algorithm for multi-job fine-tuning.

本文提出 MatryoshkaLoRA，一种受 Matryoshka 启发、面向 LoRA 的通用训练框架，通过在已有 LoRA adapter 之间插入一个固定的、经精心设计的对角矩阵来按比例缩放其子秩，从而学习到准确的层次化低秩表示。MatryoshkaLoRA is proposed, a general, Matryoshka-inspired training framework for LoRA that learns accurate hierarchical low-rank representations by inserting a fixed, carefully crafted diagonal matrix between the existing LoRA adapters to scale their sub-ranks accordingly.

这些结果表明，Kueue、DAS 与 GAIE 等互补组件构成了一个高性能的协同平台，证明了 Kubernetes 能够作为承载高要求 GenAI 工作负载的统一底座。These findings illustrate that these complementary components (Kueue, DAS, and GAIE) form a cohesive, high-performance platform, proving Kubernetes' capability to serve as a unified foundation for demanding GenAI workloads.

CONFLUX 是一个面向胸部 CT 的潜在扩散模型：由 3D 变分自编码器压缩每个体数据，整流流 Transformer 在潜在空间中生成，并以分类器从生成体中恢复所请求病征的可靠性作为奖励。CONFLUX, a latent diffusion model for chest computed tomography (CT): a 3D variational autoencoder compresses each volume, and a rectified-flow transformer generates in the latent space, that rewards how reliably a classifier recovers the requested findings from each generated volume.

首个面向预训练胸部 X 光报告生成器的无需训练 best-of-N 采样方案，显式建模"既往-当前-过渡"纵向先验，全面优于随机选择。This work presents the first training-free best-of-N sampling scheme for pre-trained chest X-ray report generators that is explicitly aware of this longitudinal prior to current transition, and outperforms random selection across the board.

本文提出了 Graph-PRefLexOR，这是一族基于图结构的推理模型，使用 Group Relative Policy Optimization 进行微调，将推理过程组织为显式阶段，分别用于机理探索、图构建、模式提取和假设合成，确立了面向图结构的强化学习作为通向可解释 AI 系统的路径，可应用于材料设计及其他科学领域的科学假设生成。Graph-PRefLexOR is developed, a family of graph-native reasoning models fine-tuned with Group Relative Policy Optimization to organize reasoning into explicit phases for mechanism exploration, graph construction, pattern extraction, and hypothesis synthesis, establishing graph-native reinforcement learning as a pathway toward interpretable AI systems for scientific hypothesis generation in materials design and other scientific applications.

本文提出 Monotonic Inference Policy Update (MIPU)，一种两步式 LLM 强化学习框架：构建采样器引用的候选更新，并使用推理侧差距代理选择性接受同步候选；实验表明 MIPU 提升了平均推理性能与训练稳定性。Monotonic Inference Policy Update (MIPU) is introduced, a two-step LLM RL framework that constructs sampler-referenced candidate updates and selectively accepts synchronized candidates using an inference-side gap proxy, and experiments show that MIPU improves average reasoning performance and training stability.

本文提出 MultiDepth-3k (MD-3k)，一个用于衡量深度层偏好与多层空间关系准确率 (ML-SRA) 的稀疏双层序数基准；领先的深度基础模型在标准 RGB 输入下表现出不同的层偏好，表明同一分层几何可在不同模型中被差异化地解析。MultiDepth-3k (MD-3k), a sparse two-layer ordinal benchmark for measuring depth-layer preference and multi-layer spatial relationship accuracy (ML-SRA), is introduced and leading depth foundation models exhibit diverse layer preferences under standard RGB input, showing that the same layered geometry can be resolved differently across models.

水电隧道检测对基础设施完整性至关重要，但人工方式效率低下且具有危险性。本文提出 FLISP（Fast LiDAR-IMU Synchronized Path Planner），一种面向 UGV-UAV 协同检测的无地图规划框架。不同于传统基于地图的范式，FLISP 具有三项核心贡献：(1) 统一架构，由单套 UGV 搭载的 LiDAR-IMU 驱动两平台的同步路径生成；(2) 平台特定的求解器，采用增强型萤火虫算法用于 UGV 避障，以及动态迭代优化器用于 UAHydropower tunnel inspection is critical for infrastructure integrity yet remains inefficient and hazardous using manual methods. We propose FLISP (Fast LiDAR-IMU Synchronized Path Planner), a mapless planning framework for cooperative UGV-UAV inspection. Unlike traditional map-based paradigms, FLISP features three core contributions: (1) a unified architecture where a single UGV-mounted LiDAR-IMU suite drives synchronized path generation for both platforms; (2) platform-specific solvers utilizing an enhanced Firefly Algorithm for UGV obstacle avoidance and a dynamic iterative optimizer for UA

实验表明，LISA 不仅能持续加速训练收敛并提升最终合成结果，还能促使侧网络特征在条件建模中更加解耦，且几乎无额外训练成本，推理成本为零。Experiments demonstrated that LISA can not only consistently accelerate the training convergence and improve final synthetic results, but also encourage the side network's features to be more disentangled for conditional modeling with negligible additional training cost and zero extra inference cost.

介绍 Stanford EDGAR Filings Dataset（SEFD），一个将 SEC filings 开放重建为版面保真 MultiMarkdown 的数据集，用于金融语言建模与评估；同时推出两个基于 SEFD 的基准：EDGAR-Forecast，用于评估模型知识截止后基于 filings 的数值预测；EDGAR-OCR，用于评估复杂金融表格的转录质量。The Stanford EDGAR Filings Dataset (SEFD), an open reconstruction of SEC filings into layout-faithful MultiMarkdown for financial language modeling and evaluation, is introduced and two SEFD-derived benchmarks are introduced: EDGAR-Forecast, which evaluates filing-grounded numerical forecasting after model knowledge cutoffs, and EDGAR-OCR, which evaluates transcription of complex financial tables.

本文提出 MedEasy，一个多 Agent 系统，通过患者对话、临床操作、决策提交、文档记录与反馈来组织虚拟患者练习，为使用案例特定标准连接情境化实践的 AI 辅助职业训练系统贡献了设计启示。MedEasy is presented, a multi-agent system that organizes virtual-patient practice through patient dialogue, clinical actions, decision submission, documentation, and feedback that contributes design implications for AI-supported professional training systems that use case-specific standards to connect situated practice.

提出 OmniTacTune，一种策略无关的真实世界 RL 流程，通过残差修正将触觉反馈适配到预训练视觉策略，并在多种接触丰富任务、视觉基础策略与触觉表征间实现泛化。OmniTacTune is introduced, a policy-agnostic real-world RL pipeline that adapts tactile feedback to pretrained visual policies through residual correction and generalizes across diverse contact-rich tasks, visual base policies, and tactile representations.

给出一个具体且有实证支撑的像素级 EO 基础模型扩展方案：训练大型 encoder，按下游性能筛选，再蒸馏为灵活的学生模型。A concrete, empirically grounded recipe for scaling pixel-wise EO foundation models: train large encoders, select by downstream performance, and distil into flexible student models is given.

借助强大的全景先验，Canvas360 构建了一个统一的上下文全景生成框架，通过 token 级拼接支持多样化下游任务，在任务覆盖范围与建模灵活性上均超越已有方法。Empowered by strong panoramic priors, Canvas360 enables a unified in-context panoramic generation framework that supports diverse downstream tasks via token-level concatenation, surpassing prior methods in both task coverage and modeling flexibility.

提出 PAST-TIDE，在官方排行榜上 Subtask A 取得 0.75 的 macro-F1，Subtask B 取得 0.74，表明对预训练模型仅做极少的架构改动即可在低资源场景下保持竞争力。PAST-TIDE is introduced, PAST-TIDE achieves macro-F1 scores of 0.75 for Subtask A and 0.74 for Subtask B on the official leaderboard, indicating that minimal architectural additions to a pre-trained model can remain competitive in low-resource settings.

通过将疾病特异性上下文整合到分子生成中，DrugGen-2 推动了 AI 辅助药物发现，为 de novo 设计和药物再利用提供了强大工具，可同时考虑疾病与分子靶点之间的复杂相互作用。By integrating disease-specific context into molecular generation, DrugGen-2 advances AI-assisted drug discovery, offering a powerful tool for de novo design and drug repurposing that accounts for the complex interplay between diseases and molecular targets.

本工作描述了如何在开源实现中满足使用有限项数的截断态向量来模拟 peaked circuits 的要求，并讨论了其性能与局限性。This work describes how the requirements to simulate peaked circuits using a truncated state vector with a limited number of terms were met in an open-source implementation, and discusses its performance and limitations.

论文提出 PanoWorld，通过固定朝向将相机轨迹简化为平移，并借助 Dense Panoramic Ray-Conditioning 与 Geometry-aware Memory Augmentation 同时支持当前动作建模与长程记忆。PanoWorld is proposed, which simplifies camera trajectories into translations via fixed headings for both current-action modeling and long-range memory through Dense Panoramic Ray-Conditioning and Geometry-aware Memory Augmentation through Dense Panoramic Ray-Conditioning and Geometry-aware Memory Augmentation.

一种简单方法 Self-Guided TTT (S-TTT)，可同时提升 Qwen3-4B-Thinking-2507 与 Llama-3.1-8B-Instruct 的准确率，相对改进最高达 15%。A simple method, Self-Guided TTT (S-TTT), which improves accuracy for both Qwen3-4B-Thinking-2507 and Llama-3.1-8B-Instruct, achieving up to a 15% relative improvement.

提出 Direct On-Policy Distillation（Direct-OPD），该方法迁移教师模型由 RL 引起的策略偏移，而非在目标模型上运行稀疏奖励 RL，并一致地利用更弱的教师模型来提升更强的目标模型Direct On-Policy Distillation (Direct-OPD) is proposed, which transfers the teacher's RL-induced policy shift instead of running sparse-reward RL on the target model and consistently leverages weaker teachers to improve stronger target models.

提出 Proxy OPD——一种异步后训练框架，迁移奖励驱动的策略改进而非绝对策略分布，将相对策略更新确立为可大规模、按奖励进行后训练的高复用、可调节资产。Proxy OPD is introduced, an asynchronous post-training framework that transfers reward-induced policy improvements rather than absolute policy distributions and establishes relative policy updates as highly reusable, adjustable assets for scalable, reward-based post-training.

一个全栈系统，可从第一人称人类视频扩展灵巧 VLA 的预训练，并支持数据高效的真实机器人后训练，在 40+ 多种任务上稳健执行自由形式指令，展现出失败恢复、灵巧性与泛化能力。A full-stack system that scales dexterous VLA pre-training from egocentric human videos and enables data-efficient real-robot post-training that robustly executes free-form instructions across 40+ diverse tasks, demonstrating failure recovery, dexterity, and generalization.

提出 SpectraReward，一种无需训练的将预训练 MLLM 转化为即用型奖励模型的奖励函数，用于图像生成强化学习；并引入 Self-SpectraReward，这是统一多模态模型的一种特例，其中策略自身的理解分支充当其生成分支的奖励模型。SpectraReward is proposed, a training-free reward function that turns pretrained MLLMs into off-the-shelf reward models for image-generation reinforcement learning, and Self-SpectraReward is introduced, a special case for unified multimodal models where the policy's own understanding branch serves as the reward model for its generation branch.

本文对 OPD 的作用、病态与调控进行系统研究，厘清 OPD 作为探索催化剂的角色，并证实调控良好的信号质量（而非单纯的教师模型规模）才是 OPD 中成功探索的主导因素。A systematic study examining the role, pathologies, and regulations of OPD, clarifying the role of OPD as an exploration catalyst and confirming that well-regulated signal quality, rather than mere teacher scale, governs successful exploration in OPD.

Xiaomi-Robotics-1是一个强大的机器人基础策略，能在复杂灵巧任务上以高数据效率高效微调，并在多个仿真基准上超越SOTA方法。Xiao-Robotics-1 serves as a strong robot foundation policy that can be efficiently fine-tuned on complex, dexterous tasks with high data efficiency and across multiple simulation benchmarks, Xiaomi-Robotics-1 outperforms state-of-the-art methods.

提出xHC（扩展型超连接），是首个在N=4之外实现有意义扩展的HC系列方法；引入xHC-Flash，将每个子层的内存访问量从73.5C降至40C（与mHC在N=4时的34C相当），同时保留完整xHC带来的性能增益。This work proposes xHC (Expanded Hyper-Connections), the first HC-family method to achieve meaningful expansion beyond $N{=}4$ and introduces xHC-Flash, which reduces the per-sublayer memory traffic from $73.5C$ to $40C, comparable to the $34C$ required by mHC at $N{=}4$, while retaining the gains of full xHC.

BLIP-2 在多种视觉-语言任务上取得 SOTA 性能，可训练参数远少于现有方法，并展现出遵循自然语言指令进行零样本图生文的新兴能力。BLIP-2 achieves state-of-the-art performance on various vision-language tasks, despite having significantly fewer trainable parameters than existing methods, and is demonstrated's emerging capabilities of zero-shot image-to-text generation that can follow natural language instructions.

本文从参考图像中识别低秩背景特征子空间，并以闭式方式将参考与查询特征投影到其正交补空间，从而获得更清晰的语义匹配，并在 PACO-Part、FSS-1000 以及 ISIC2018 等跨域数据集的无训练方法中达到新的 SOTA，表明显式去除背景子空间是一次性定位中极为有效的原则。This paper identifies the low-rank background feature subspace from the reference image and project the reference and query features onto its orthogonal complement in closed form, yielding cleaner semantic matching, and establishes a new state of the art among training-free methods on PACO-Part, FSS-1000, and cross-domain datasets such as ISIC2018, demonstrating that explicit background subspace removal is a highly effective principle for one-shot localization.

本文提出 Token-Level Off-Policy Labeling (TOPL)，一种将后训练重构为 token 级正确性预测任务的离线策略训练范式。其核心思路是：通过训练模型区分响应中的好 token 与坏 token，自然引导模型生成好 token，同时避免直接训练模型生成离线策略 token 所带来的缺陷。在文档摘要任务上的实验表明，TOPL 在 11 个数据集上针对多种序列级与 token 级方法实现了强大的分布外泛化能力。We propose Token-Level Off-Policy Labeling (TOPL), an off-policy training paradigm that reframes post-training as a token-level correctness prediction task. Our key intuition is that by training the model to distinguish good and bad tokens in a response, we naturally guide the model towards generating good tokens, while avoiding the pitfalls that come with directly training the model to generate off-policy tokens. Experiments on document summarization tasks show that TOPL achieves strong out-of-distribution generalization across 11 datasets against a diverse set of sequence-level and token-lev

大量实验表明，Distilled RL 在 pass@1 和 pass@k 上均显著优于标准 RL 和 OPD，并能将教师模型中此前无法获得的知识有效迁移至学生模型。Extensive experiments show that Distilled RL substantially outperforms standard RL and OPD in terms of both pass@1 and pass@k, and can effectively transfer previously unavailable knowledge from a teacher model to a student model.

本文针对中亚低资源语言构建鲁棒基础模型的挑战，提出 GigaAM Multilingual：一个基于 HuBERT 风格目标在 2M 小时音频上预训练的 Conformer encoder，并在预训练阶段引入 cluster 级的数据均衡策略，在微调阶段引入域感知采样方法，以缓解头部语言主导问题。This work addresses the challenge of building robust foundation models for underrepresented Central Asian languages by presenting GigaAM Multilingual, a Conformer encoder pre-trained on 2M hours of audio using a HuBERT-style objective, and introduces a cluster-level data balancing strategy during pre-training and a domain-aware sampling method during fine-tuning to mitigate head-language dominance.

本文研究 SkewAdam——一种基于以下观察构建的优化器：MoE 的三类参数群体在规模与梯度统计上差异足够大，不应共享相同的状态；研究表明优化器状态的存放位置至少与状态容量同等重要。SkewAdam, an optimizer built on the observation that the three parameter populations of an MoE differ enough in size and gradient statistics that they should not receive the same state, is studied, suggesting where optimizer state lives matters at least as much as how much of it there is.