目录 Contents
- 01. DualPath Breaking the Storage Bandwidth Bottleneck in Agentic LLM Inference DualPath: 打破智能体推理的存储带宽瓶颈
- 02. Boosting Document Parsing Efficiency and Performance PaddleOCR-VL: 粗到细视觉处理提升文档解析
- 03. KIMI K2.5 VISUAL AGENTIC INTELLIGENCE KIMI K2.5: 视觉智能体革命
- 04. BabyVision Visual Reasoning Beyond Language BabyVision: 超越语言的视觉推理
- 05. End to End Test Time Training for Long Context TTT-E2E: 端到端测试时训练实现长上下文
- 06. Conditional Memory via Scalable Lookup A New Axis of Sparsity for Large Language Models DeepSeek Engram：条件记忆——大语言模型稀疏性的新维度
- 07. Benchmarking Multilingual Long Context Language Models ONERULER: 多语言长上下文语言模型基准测试
- 08. Recursive Language Models 递归语言模型 (RLM) - 让模型决定如何处理海量上下文
- 09. Pretraining Large Language Models with NVFP4 NVFP4: 大语言模型4位浮点数预训练的革命性突破
- 10. Agentic Context Engineering Evolving Contexts for Self Improving Language Models ACE: 智能体上下文工程的革命性突破
- 11. sinq SINQ: 无需校准的高效模型量化
- 12. jet nemotron Jet-Nemotron: 高效语言模型的革命性突破
- 13. Video Models are Zero Shot Learners and Reasoners 视频模型的零样本学习与推理能力