从似然陷阱到退化惩罚再到语法约束，拆解三类高级解码策略的底层机制、计算代价与适用边界。

Contrastive-LM has released CLM-8B, the first open model in a new class called Contrastive Language Models (CLMs). CLM d…

Cross-lingual contrastive learning has been a core component of multilingual encoder training, but the ability to explic…

Given a target dataset, such as faces with eyeglasses, and a background dataset, such as faces without, contrastive anal…

Traditional multimodal representation learning and generation are two stages: a contrastive or self-supervised visual en…

语言模型的训练目标浓缩为一个损失函数——交叉熵。但围绕它的变体（label smoothing、辅助损失、对比损失）各自解决不同的问题。本篇拆解交叉熵的直觉与数学、label smoothing 为什么能防过自信、辅助损失在 MoE 与多任务中的角色，以及对比损失在 Embedding 训练里的位置。

Embedding 模型把文本映射成向量，「相似的文本向量相近」——这个能力靠对比学习训练出来，而它的成败几乎完全取决于训练数据：正例对怎么构造、难负例怎么挖。本篇拆解对比学习的训练机制、难负例挖掘的工艺、Embedding 模型的两代演进（BERT 系到 LLM 系），以及微调一个自有 Embedding 模型的最小路径。

Embedding models underpin virtually every modern AI system, from search engines to recommendation systems. Despite share…

针对生成内容音画不同步问题，论文提出包含8万条专业同步评分的SyncReward-Data-80K数据集、SyncReward-Bench基准及同步感知奖励模型，覆盖语音、发声、乐器、环境声四类声音事件。

论文提出UniSkill，让共享策略在与环境交互的同时从轨迹中提出技能库编辑（新增、更新或不编辑），行动者通过环境奖励学习，而对比性动作反馈用于指导技能提案学习，以解决技能收益与行动者自身提升相互混淆的问题。

CLIP（Radford et al., 2021）用 4 亿网络图文对做对比学习，零样本在 ImageNet 拿到 76.2%——追平全监督的 ResNet-50。本文拆解对比学习的训练机制、零样本分类的「提示词即分类器」玩法、CLIP 的失效模式，以及它如何成为多模态时代的地基。

Extending a text embedding model to new modalities typically degrades text retrieval quality, and existing omni-modal em…

Looped Transformers achieve parameter efficiency by repeatedly executing a shared block across recurrent loops. Each loo…

An activation-steering coefficient sets intervention strength, but requesting a particular degree of persona expression …

Reliable AI safeguards require both control mechanisms that reduce unsafe behavior and monitoring mechanisms that detect…

Vision-language (VL) pretraining using paired chest X-ray (CXR) images and radiology reports has shown strong potential …

提出EviAlign方法，通过生成语义证据并在证据单元边界读取上下文状态、聚合为归一化嵌入，联合生成与对比检索目标训练共享多模态大模型结构。

In this report, we introduce Ovis-Embedding, a state-of-the-art omni-modal embedding family built on native integration …

Linkup Research has released SPARSEUP, an open-source sparse embedding model built on a 149M-parameter ModernBERT backbo…

Professional graphic design is a long-horizon agentic task in which structured, editable artifacts emerge from many inte…

Skills can improve the performance of Large Language Model (LLM) agents by providing task-specific procedural guidance, …

Modeling articulated objects from sparse monocular views is challenging because each observation reveals only partial ge…

World modeling enables intelligence to anticipate consequences, guide interventions, and learn from interaction. Yet pre…

Safety guardrails in open-weight language models can be readily bypassed using Refusal Feature Ablation (RFA), a techniq…

Recent work extends recursive self-improvement (RSI) to agent harnesses for long-horizon coding and terminal tasks, enab…

Open-vocabulary detection accepts any class list at inference, and promptable segmentation returns regions without class…

Large language model (LLM) agents can benefit from reusable skills distilled from prior task experience, yet existing sk…

On-Policy Self-Distillation (OPSD) has emerged as a popular paradigm for large language model (LLM) self-improvement, al…

Omni-modal large language models (OLLMs) jointly process vision, audio, and text, yet their modality bias under cross-mo…