论文

结果为兼容的开源权重语言模型检查点建立了一种被动、无需数据的溯源信号，且该投影配对信号出现在六个及更多语言模型系列中The results establish a passive, data-free provenance signal for compatible open-weight language-model checkpoints, and the projection-pairing signal appears across six language-model families and beyond.

提出 AdaPop（自适应流行度）方法，将局部 token 置信度与源自外部代理的逐事实流行度相关指数相结合，并通过双上升控制器在每个 epoch 调整 retain 惩罚来自动平衡遗忘与保留。The AdaPop (Adaptive Popularity) method is proposed, which combines local token confidence with a per-fact popularity-dependent exponent derived from an external proxy, and automates the forget-retain balance via a dual-ascent controller that adjusts the retain penalty each epoch.

结果表明，广泛的 SFT 带来模型大部分能力提升；当失败检测精准时，turn-local 监督可发挥作用，且观察到的迁移主要集中在同族模型之间。The results suggest that broad SFT brings most of the model's capability improvement; turn-local supervision can be effective when failure detection is precise, with observed transfer concentrated primarily within-family.

数据驻留约束迫使企业自托管 LLM，但不断引入新模型而不下线旧模型会扩张服务集群，分散有限的 GPU 池。我们通过沿指令遵循、函数调用和内部任务分布三个维度，针对生产错误分析所发现的质量差距，将 200 多个内部应用的流量整合到单一模型上。质量通过按生产流量分层的离线基准进行跟踪，并由确定性验证器或经过校准的 LLM 评判器打分。不同于针对Data-residency constraints force enterprises to self-host LLMs, but continuous adoption of newer models without decommissioning their predecessors expands the serving fleet, fragmenting a finite GPU pool. We consolidate traffic from over 200 internal applications onto a single model by closing quality gaps identified through production error analysis along three axes: instruction following, function-calling, and internal task distribution. Quality is tracked by offline benchmarks stratified to production traffic and scored by deterministic verifiers or calibrated LLM judges. Rather than optimi

本文表明强化学习（RL）后训练已具备实现有效 step-level 评分所需的要素，从而完全无需额外的奖励模型训练，并在通用随机 Markov 决策过程下推导出一种隐式 advantage，称为 progress advantage。This work shows that reinforcement learning (RL) post-training already provides the ingredients for effective step-level scoring, eliminating the need for dedicated reward model training altogether, and derives an implicit advantage under a general stochastic Markov decision process, which is term progress advantage.

本文提出 Anchor-Align，通过两个目标增强 BC：Vision-Language Anchoring 从冻结 VLM 副本中蒸馏逐层表示以防止该漂移；Language-Action Alignment 将每个动作目标转换为离散的运动方向标签，并在同一机器人观测上联合训练语言与动作预测。Anchor-Align is proposed, which augments BC with two objectives: Vision-Language Anchoring distills layer-wise representations from a frozen VLM copy to prevent this drift, and Language-Action Alignment converts each action target into a discrete motion-direction label and jointly trains language and action prediction on the same robot observation.

提出 DataPrep-Bench，首个统一基准，在共享的下游任务 grounding 协议下，对 LLM 驱动的数据准备在六个领域、多种 base model 上的两类能力进行联合评估。DataPrep-Bench is introduced, the first unified benchmark that jointly evaluates both capabilities under a shared downstream-grounded protocol over six domains and multiple base models of LLM-driven data preparation.

发现面向推理的训练并未优先放大具有最高 Lift 的行为，这促使研究者采用过程级目标，以奖励经过校准且有依据的推理，而不仅仅是表面形式。It is found that reasoning-oriented training does not preferentially amplify the highest-Lift behaviors, motivating process-level objectives that reward calibrated and grounded reasoning rather than surface form alone.