寄存器token让代码任务最高提升19.5分

重点关注

• AI for Games in the Foundation Model Era score 9
  入选 HF Daily Papers；HF 热度: 115 upvotes (+4)；有代码实现
• ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agents score 9
  入选 HF Daily Papers；HF 热度: 23 upvotes (+4)；有代码实现
• FLAT: Resampling Image and Text into 1D Flexible-Length Aligned Transmodal Tokens for Retrieval and Generation score 7
  入选 HF Daily Papers；HF 热度: 13 upvotes (+3)；关键词(3): fine-tuning, pre-training, text-to-image
• Register Tokens for Bounded-State Reasoning in Diffusion Language Models score 8
  入选 HF Daily Papers；HF 热度: 5 upvotes (+2)；有代码实现；关键词(2): code generation, reasoning
• Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration score 7
  入选 HF Daily Papers；HF 热度: 2 upvotes (+1)；有代码实现；关键词(2): finetuning, guardrails
• ImpossibleRubrics: Stress-Testing Generated Rubrics as Reward Signals score 6
  入选 HF Daily Papers；HF 热度: 4 upvotes (+1)；有代码实现
• LimiX-2: A Contextual Mechanism Network Towards General Structured-Data Intelligence score 10
  入选 HF Daily Papers；HF 热度: 165 upvotes (+4)；有代码实现；关键词(2): scaling, pretraining
• OmniHarness: Harnessing Generalizable Visual Generation via Symbolic Policy Learning score 6
  入选 HF Daily Papers；HF 热度: 2 upvotes (+1)；有代码实现
• EventEgoHands++: Event-based Egocentric 3D Hand Mesh Reconstruction with Real Dataset score 9
  入选 HF Daily Papers；HF 热度: 21 upvotes (+4)；有代码实现

也值得关注

• PhysStream: Streaming Physics-Grounded Video Generation with Structured Scene Memory and Fine-Grained Motion Control score 5
  入选 HF Daily Papers；HF 热度: 6 upvotes (+2)
• Emergence World: Adversarial Stress-Testing of Long-Horizon Multi-Agent Systems score 4
  入选 HF Daily Papers；HF 热度: 2 upvotes (+1)
• POSPAN: Position-Constrained Span Masking for Language Model Pre-training score 4
  机构: Google；关键词(1): pre-training
• Decoder Design Matters for ECG Delineation score 4
  机构: Carnegie Mellon；关键词(1): open-source
• Query-Aware Source-Risk Triage for Retrieval-Augmented Generation score 4
  机构: Google；关键词(2): retrieval-augmented, RAG
• A Framework for Generating Valid Context-Specific Benchmarks through Expert Guidance score 4
  机构: Carnegie Mellon；关键词(1): synthetic data
• Turn-level Multiscale Density Ratio Estimation for LLM Agents score 4
  机构: Alibaba；关键词(5): DPO, PPO, GRPO, post-training, reasoning
• Verifiable Social Reasoning for LLM Assistants score 4
  机构: Google；关键词(2): reasoning, open-source
• From Transient Prompts to Persistent Control: Scientific Poster Generation via Recursive Semantic-Geometric Contracts score 3
  机构: Baidu