📄 Hugging Face 今日论文精选(2026-06-11)

:page_facing_up: Hugging Face 今日论文精选(2026-06-12)

核心主线:今日论文聚焦于大模型安全漏洞、自主科研智能体、MoE架构优化、分布式训练效率提升以及智能体强化学习的最新前沿探索。

  1. Grammar-Constrained Decoding Can Jailbreak LLMs into Generating Malicious Code :star:4.5

  2. Toward Generalist Autonomous Research via Hypothesis-Tree Refinement :star:4.5

  3. Redesign Mixture-of-Experts Routers with Manifold Power Iteration :star:4.5

  4. Breaking the Bubble: Asynchronous Pipeline Parallel Training with Bounded Weight Inconsistency :star:4

  5. EvoTrainer: Co-Evolving LLM Policies and Training Harnesses for Autonomous Agentic Reinforcement Learning :star:4