论文

证明：对于任意固定的目标误差水平 $\delta$ 与任意松弛量 $\varepsilon>0$，数量级为 $p/\psi^2$ 的样本量足以在任意小的 $\psi$ 下完成支撑恢复。It is proved that, for every fixed target error level $\delta$ and every slack $\varepsilon>0$, a sample size of order $p/\psi^2$ is sufficient for support recovery for arbitrarily small $\psi$.

本文提出了一种简单的 flow-control 框架，通过控制 prompt 加入 LLM 活跃集合的速率，实现更高的 token 与 request 吞吐量、更低的平均与尾部 latency，以及更稳定的 KV cache 利用率。A simple flow-control framework is proposed that controls the rate at which prompts join the active set in large language models and achieves higher token and request throughput, lower average and tail latency, and more stable KV cache utilization.

提出 Negative Self-Distillation（NSD），一个通过偏离有缺陷的推理而非模仿特权解来优化 LLM 的新框架，并一致优于 OPSD 及其他无标签、自举式强化学习（RL）基线。Negative Self-Distillation (NSD) is introduced, a new framework that optimizes LLMs by diverging from flawed reasoning rather than imitating privileged solutions, and consistently outperforms OPSD and other label-free, self-bootstrapping reinforcement learning (RL) baselines.

提出 Generative Verification（GenV），通过复用语言模型的原生词表空间，将离线 Z3 等价性预言机蒸馏为无参考、连续参考等价分数，并从理论上证明基于结构、仅判决的验证启发式在这些欺骗性合法轨迹上的检测能力在数学上有界于随机水平。Generative Verification (GenV) is introduced, which distills an offline Z3-equivalence oracle into a reference-free, continuous reference-equivalence score by repurposing the language model's native vocabulary space and theoretically proves that structural, verdict-only verification heuristics are mathematically bounded to chance-level detection on these deceptively valid traces.

表明将 L2 推理泛化到未见语言的关键路径在于更广泛的语言覆盖、现成可用的多语言非推理数据，以及足够强的英语推理骨干，表明推理是一种与语言无关的行为，可通过精心数据混合在类型多样的语言间迁移。It is shown the path to generalizing L2 reasoning to held-out languages goes through broader language coverage, readily available multilingual non-reasoning data, and a sufficient English reasoning backbone, indicating that reasoning is a language-agnostic behavior that can be transferred across typologically diverse languages through careful data mixing.

结果表明，在评估 harness 中使用 Adaptive Bridge 可将关键的订阅者尾端 p95 延迟从最高 15 s 降至 1.55 ms，覆盖所有损伤严重度，同时保留发布者配置的吞吐量。The results show that using the Adaptive Bridge in the evaluation harness reduces the critical subscriber tail p95 latency from up to 15 s to 1.55 ms across all impairment severities while preserving the publisher's configured throughput.

提出了一个能力门控（competence gate），根据已解决的结果估计领域级的源权重，将不确定的估计向全局权重收缩，并重新校准合并后的预测，为基于已测边际价值的选择性模型使用提供了实用方案。A competence gate is introduced that estimates domain-level source weights from resolved outcomes, shrinks uncertain estimates toward a global weight, and recalibrates the pooled forecast and provides a practical approach for selective model use based on measured marginal value.

本文提出 CATS，一种 self-speculative decoding 框架，在内存受限设备上结合 memory budget 与 parameter offloading pattern 进行级联式 verify 与 correction，在设备峰值显存与单独运行 target model 相当的前提下，最大化 token acceptance rate 与端到端加速比。CATS, a self-speculative decoding framework that conducts cascaded verification and correction based on the memory budget and parameter offloading patterns on memory-limited devices, is proposed, which maximizes token acceptance rate and end-to-end speedup while keeping the peak memory footprint on the device equal to that of the target model alone.

展示并论证了算法无损性与其在有限精度算术下实现之间存在差距，主张应从精确生成轨迹与下游任务性能两个层面评估无损投机解码A gap between algorithmic losslessness and its implementation under finite-precision arithmetic is demonstrated and motivated, to motivate evaluating lossless speculative decoding at the level of exact generation trajectories as well as downstream task performance.

本文提出 DualPath，一种 inference 系统，通过引入 dual-path KV-Cache loading 打破瓶颈，并实现一条新的 storage-to-decode 路径：KV-Cache 先加载到 decode engine，再通过 compute network 上的 RDMA 高效转发至 prefill engine。DualPath is presented, an inference system that breaks this bottleneck by introducing dual-path KV-Cache loading and enables a novel storage-to-decode path, in which the KV-Cache is loaded into decoding engines and then efficiently transferred to prefill engines via RDMA over the compute network.

本文提出 RelateAnything，一个 53M 参数的模型，输入一张图像和来自任意来源的区域，针对推理时以字符串形式提供的谓词词汇返回带分数的关系。This work presents RelateAnything, a 53M-parameter model taking an image and regions from any source and returning scored relations over a predicate vocabulary supplied at inference as strings, a 53M-parameter model taking an image and regions from any source and returning scored relations over a predicate vocabulary supplied at inference as strings.

本文识别出 Value Flattening 是标准 PPO 中 critic 学习的一种重要但被忽视的失效模式，并提出一种简单稀疏监督策略可以缓解该问题；引入 SParse Proximal Policy Optimization，在每个响应中仅对少数间隔良好的状态施加 value loss，以同时缓解两种效应。Value Flattening is identified as an important yet overlooked failure mode of critic learning in standard PPO and a simple sparse supervision strategy can mitigate it; SParse Proximal Policy Optimization is introduced, which applies the value loss to only a few well-separated states in each response to mitigate both effects.

本文提出 Edge0，一个流式 MoE 推理引擎，通过 prerouter 缩小差距：每层 head 提前一个 token 预测下一层的 routing，并将该预测直接作为 routing 使用，使分阶段 expert 集合等于路由集合，无任何丢弃。Edge0 is presented, a streaming MoE inference engine that closes the gap with a prerouter: a per-layer head predicts the next layer's routing one token ahead, and the prediction is consumed as the routing itself, so the staged expert set equals the routed set and nothing is dropped.

本文提出 Fathom，一种 key scan，其中每个查询决定读取每个 key 通道的比特数；在真实的 coding agent 会话中，Fathom 以 92 bit 达到最准确的 136 bit scan 的步骤一致性。Fathom is presented, a key scan in which each query decides how many bits of each key channel to read, and on real coding-agent sessions Fathom reaches the step agreement of the most accurate 136-bit scan at 92 bits.

BraTS-GoAT 在异质人群上使用传统 3D nnU-Net 对 1,351 个标注案例进行肿瘤分割评估，采用五折交叉验证，每折 1,000 epochs，并应用 test-time mirroring。BraTS-GoAT evaluates tumor segmentation across heterogeneous populations using a conventional 3D nnU-Net on 1,351 labeled cases using five-fold cross-validation and 1,000 epochs per fold and applied test-time mirroring.

提出 WeVisDoc，一个面向鲁棒端到端文档解析的两阶段数据驱动框架，通过异构数据与保持结构的退化合成来扩展语义、结构与外观覆盖度，并指导有针对性的数据构建与目标 token 预算的重分配。WeVisDoc is presented, a two-stage data-centric framework for robust end-to-end document parsing that broadens semantic, structural, and appearance coverage through heterogeneous data and structure-preserving degradation synthesis and guides targeted data construction and reallocation of the target-token budget.

构建了 AMPLE-Math，一个包含 5,319 道数学问题、每题对应六种共享同一答案的推理视角的可复用套件，并通过对比表明 OPSD 能借助直答推理与带思维链推理所共享的参数，提升对已有推理能力的调用效率。AMPLE-Math, a reusable suite of 5,319 mathematical problems with six reasoning views that share the same answer, is constructed and AMPLE-Math, a reusable suite of 5,319 mathematical problems with six reasoning views that share the same answer, is compared, suggesting that OPSD can improve access to existing reasoning capabilities through parameters shared by direct-response and thinking-enabled inference.

我们研究 on-policy 蒸馏 (OPD) 中的长度膨胀现象，即学生回答会变得过长，甚至耗尽生成预算。我们发现基础学生模型与训练后教师模型之间的终止 token 不匹配是该行为的重要来源。在 Qwen3、Llama 和 Gemma 上，两个模型可能将停止概率分配到不同的 EOS token 上，即便它们声明的停止集合相同。这种不匹配会抑制学生偏好的终止动作，同时无法可靠地传递教师偏好的替代动作。我们证明对齐 t...We study length inflation in on-policy distillation (OPD), where student responses can become excessively long and even exhaust the generation budget. We identify termination-token mismatch between base students and post-trained teachers as an important source of this behavior. Across Qwen3, Llama, and Gemma, the two models can place their stopping probability on different EOS tokens, even when their declared stopping sets are identical. This mismatch can suppress the student's preferred termination action without reliably transferring the teacher-preferred alternative. We show that aligning t

结果表明，仅靠候选数量不足以刻画多候选 test-time scaling 的系统成本；评测应同时报告候选数量与准确率，以及生成调度和 GPU 层级的系统指标。The results show that candidate count alone is not enough to describe the systems cost of multi-candidate test-time scaling and Evaluations should report not only candidate count and accuracy, but also generation schedule and GPU-level systems metrics.

IntBMoE 是一种 block-conditioned MoE，将三者解耦，通过密集专家组合与稀疏 block 执行配对实现；在图像分类任务上，相较代表性的稀疏与密集 MoE 基线均取得稳定提升。IntBMoE is a block-conditioned MoE that decouples all three by pairing dense expert composition with sparse block execution, and shows consistent gains over representative sparse and dense MoE baselines on image classification.

本文提出 TTKV，一种 KV cache 管理框架，将人类记忆系统映射到具备异构容量与精度的 KV cache 上，在 128K 上下文任务上将跨层流量降低 5.94 倍。TTKV is proposed, a KV cache management framework that maps the human memory system onto the KV cache with heterogeneous capacity and precision, and reduces cross-tier traffic by 5.94x on 128K-context tasks.

在更大规模 dense 模型与 MoE 模型上的比较进一步凸显了在持续生成过程中调度 prompt 处理的重要性：在并发负载下，vllm-metal 的 packed prefill-decode 路径维持了低于 omlx 的首 token 延迟。Comparisons on larger dense and MoE models reinforce the importance of scheduling prompt processing alongside ongoing generation: vllm-metal's packed prefill-decode path maintains lower first-token latency than omlx under concurrent load.

表明 Kimi Delta Attention (KDA) 可通过单次 delta-rule 变换与其通道门控提供的反射组合实现 2D 旋转，并刻画 CKDA 的表达能力，证明每个正交的对角加秩-1矩阵恰为一个 CKDA 转移矩阵。This work shows that Kimi Delta Attention (KDA) can realize 2D rotations by combining a single delta-rule transformation with a second reflection supplied by its channel-wise gate, and characterize the expressivity of CKDA and prove that every orthogonal diagonal-plus-rank-one matrix is exactly a CKDA transition matrix.

本文提出 KV Policy (KVP)，一种仅基于 key 与 value 向量、在预计算生成轨迹上训练的轻量级 per-head RL Agent 框架，证明学习预测未来 token 效用是自适应 KV cache 管理中强大且可扩展的范式。KV Policy (KVP), a framework of lightweight per-head RL agents trained on pre-computed generation traces using only key and value vectors, is introduced, demonstrating that learning to predict future token utility is a powerful and scalable paradigm for adaptive KV cache management.

本文展示了在一对架构匹配的 Qwen3.5 4B-to-9B 兄弟模型之间实现有用的持久 hybrid-state transfer，是首次在大小不同的混合语言模型之间完成持久循环推理状态的跨模型交接，且无需 target prefix replay。This work demonstrates useful persistent hybrid-state transfer across one architecture-matched Qwen3.5 4B-to-9B sibling pair, the first demonstrated cross-model handoff of persistent recurrent inference state between differently sized hybrid language models without target prefix replay.

TritonForge 是一个面向自动化 Triton kernel 优化的 profiling 引导框架，融合 kernel 分析、运行时 profiling 与迭代式代码转换以简化优化流程，并为自动化 GPU 性能优化领域的未来研究奠定基础。TritonForge, a profiling-guided framework for automated Triton kernel optimization that integrates kernel analysis, runtime profiling, and iterative code transformation to streamline the optimization process and provides a foundation for future research in automated GPU performance optimization.

本文探讨容器化、微服务、动态调度等 Cloud Native 技术如何从根本上提升 LLM 推理服务，并展示 Cloud Native 系统在高需求场景下实现更高效资源分配、降低延迟与提升吞吐的能力。This article explores how Cloud Native technologies, such as containerization, microservices, and dynamic scheduling, can fundamentally improve LLM inference serving and demonstrates how a Cloud Native system enables more efficient resource allocation, reduces latency, and enhances throughput in high-demand scenarios.

本文提出一种方法，通过逐层剔除后 WER（Word Error Rate）的变化对编码器层进行排序，并给出剪枝后的模型——即一个层数更少的更浅编码器。This work presents an approach that ranks encoder layers by the leave-one-layer-out change in Word Error Rate (WER), and presents the pruned model, which is simply a more shallow encoder with fewer layers.

本文首次对 vLLM 启动延迟进行了详细的性能表征，并构建了一个轻量级分析模型，能够针对给定硬件配置准确预测 vLLM 的启动延迟，为大规模推理环境中的资源规划提供了可操作的指导。This paper presents the first detailed performance characterization of vLLM startup latency and develops a lightweight analytical model that accurately predicts vLLM's startup latency for a given hardware configuration, providing actionable guidance for resource planning in large-scale inference environments.

本文提出 Neural Spectral Capacity (NSC)，一种以每个权重矩阵的奇异值谱为依据的闭式标量，在七类 Transformer 与 CNN 系列上的排序效果优于 #Params、#FLOPs 及代表性免训练代理指标。This work proposes Neural Spectral Capacity (NSC), a closed-form scalar grounded in the singular-value spectrum of each weight matrix, which outperforms #Params, #FLOPs, and representative training-free proxies in ranking across seven Transformer and CNN families.

本文提出 PISA，一种采用金字塔式 Top-K 选择策略的 block-sparse attention 机制，并为训练与推理开发了硬件感知的 Triton kernel，将层级路由与 LogSumExp 评分融合，无需显式构造 query-key score 矩阵。PISA is proposed, a block-sparse attention mechanism that employs a pyramid Top-$K selection strategy, and develops hardware-aware Triton kernels for both training and inference, fusing hierarchical routing and LogSumExp scoring without materializing the query-key score matrix.

本文提出 PsPLUG，一个轻量级插件式模块，在刻画目标风格后学习用户专属的残差，能更好地保留用户偏好，并对个性化与风格遵循之间的平衡提供更精确的控制。PsPLUG is proposed, a lightweight plug-in that learns a user-specific residual after accounting for the requested style, and better preserves user preferences while providing precise control over the balance between personalization and style adherence.

本文提出首个通过 continuous depth batching (CDB) 实现 depth-adaptive looped LM 的高效方法：在 loop 步骤之间重组 batch，动态调度架构中的 looped 与非 looped 部分，管理 looped KV-caching，并提前预测将退出 loop 的 token，以便异步准备 batch。This work introduces the first efficient method for depth-adaptive looped LMs via continuous depth batching (CDB), which forms new batches between loop steps, and dynamically schedules looped and non-looped parts of the architecture, manages looped KV-caching, and predicts which tokens will exit the loop in advance so it can prepare batches asynchronously.

使用一种分层旋转位置编码（hRoPE），将段落、句子和 token 索引表示为独立通道，保持 token 序列固定，对段落坐标 $p_1$ 进行干预，并使用 token 距离精确估计器测量跨段落注意力。A hierarchical rotary positional encoding (hRoPE) that represents paragraph, sentence, and token indices as separate channels, hold the token sequence fixed, intervene on the paragraph coordinate $p_1$, and measure cross-paragraph attention with a token-distance-exact estimator is used.

Intent2Tc 是一个由语言模型驱动的闭环框架，将业务级流量整形意图转换为声明式子意图，进而生成经过验证的可执行 Linux traffic control (tc) 配置，并展示了该框架的实际适用性。Intent2Tc is presented, a closed-loop language-model-driven framework that translates business-level traffic-shaping intents into declarative sub-intents and subsequently into validated, executable Linux traffic control (tc) configurations and demonstrates the practical applicability of the proposed framework.

本文提出 softmax reparameterization，一种训练后方法，在量化前搜索功能等价的输出头，并展示了在总体 logit 误差增大的情况下保真度仍可提升。This work introduces softmax reparameterization, a post-training method that searches over functionally equivalent output heads before quantization and shows how fidelity can improve despite greater total logit error.