📄 Hugging Face 今日论文精选(2026-06-10)

:page_facing_up: Hugging Face 今日论文精选(2025-06-26)

核心主线:模型可解释性、多模态理解、安全对齐、长视频推理与世界模型评估

  1. Interpreting and Steering a Text-to-Speech Language Model with Sparse Autoencoders :star:

  2. Kwai Keye-VL-2.0 Technical Report :star:

    • 链接[2606.10651] Kwai Keye-VL-2.0 Technical Report
    • 为什么值得看:快手多模态大模型的技术报告,代表工业级视觉-语言模型的系统设计与最新进展,对多模态社区有重要参考价值。
  3. PsychoSafe: Eliciting Psychologically-Informed Refusals in Large Language Models :star:

  4. MemDreamer: Decoupling Perception and Reasoning for Long Video Understanding via Hierarchical Graph Memory and Agentic Retrieval Mechanism :star:

  5. WorldOlympiad: Can Your World Model Survive a Triathlon? :star: