Self-HPO将Qwen3.5-35B-A3B在3个编码基准上的平均resolve rate从57.8%推至64.8%

重点关注

• RAP: A Dataset for Paper-to-Poster Generation via Declarative Markup from Human-Crafted Exemplars score 13
  机构: Carnegie Mellon；入选 HF Daily Papers；HF 热度: 30 upvotes (+4)；有代码实现；关键词(1): coding
• Tree-Aware Posterior Distillation for Long-Horizon Agentic Reinforcement Learning score 11
  机构: Tsinghua；入选 HF Daily Papers；HF 热度: 46 upvotes (+4)；关键词(4): distillation, GRPO, post-training, agentic
• Scene-Designer: Learning Design Intelligence for Agentic 3D Scene Generation score 10
  入选 HF Daily Papers；HF 热度: 61 upvotes (+4)；有代码实现；关键词(2): GRPO, agentic
• Towards Looped Models Done Right, Part I: Topology, Input Injection, Recurrent-State Design score 10
  机构: Carnegie Mellon；入选 HF Daily Papers；HF 热度: 15 upvotes (+3)；关键词(3): MoE, coding, reasoning
• Self-Evolving Coding Agents via Harness-Policy Optimization score 10
  入选 HF Daily Papers；HF 热度: 44 upvotes (+4)；有代码实现；关键词(1): coding
• Stuck on Shuffle: On Attribute-Object Binding in Vision-Language Models score 9
  入选 HF Daily Papers；HF 热度: 15 upvotes (+3)；有代码实现；关键词(1): vision-language
• OfficeTown: Generative Organizations for Evolving Agent Intelligence score 8
  入选 HF Daily Papers；HF 热度: 20 upvotes (+4)；关键词(2): distillation, fine-tuning
• Lean4AgenticRL: Advancing Agentic RL with Verifiable Multi-level Process Rewards score 9
  入选 HF Daily Papers；HF 热度: 15 upvotes (+3)；有代码实现；关键词(2): GRPO, agentic
• OmniXRef: A Unified Framework for Multimodal Multi-Reference Video Editing score 7
  入选 HF Daily Papers；HF 热度: 20 upvotes (+4)
• GlobeBench: Evaluating Text-Based World Model Capabilities for Scalable Environment Simulations score 7
  入选 HF Daily Papers；HF 热度: 6 upvotes (+2)；有代码实现

也值得关注

• ECHOES: Seeing Beyond the Current Image via Recursive History Compression and Privileged On-Policy Distillation score 4
  机构: Tsinghua；关键词(4): compression, distillation, deployment, reasoning
• Functional Welfare for AI Safety score 4
  机构: University of Toronto；关键词(2): finetuning, post-training
• Progressive Misalignment Can Emerge from Benign Recursive Self-Distillation score 4
  机构: Cambridge；关键词(1): distillation
• CFC26: Building Evaluations for Deployment in Sonar-Based Fish Counting score 4
  机构: MIT；关键词(1): deployment
• Repairing LLM-Generated Specifications with Deterministic Diagnosis score 4
  机构: Oxford；关键词(1): reasoning
• At the Limits of Computability: Steps Towards Autonomous AI Discovery on the Busy Beaver Frontier score 4
  机构: Google；关键词(2): agentic, reasoning
• AlphaBet: Reinforcement Learning for LLM Event Forecasting Agents score 4
  机构: Oxford；关键词(2): GRPO, agentic
• LEMON: Equivariant Learning over Attention Maps for Zero-Shot Misbehavior Monitoring on New LLMs score 4
  机构: Google；关键词(2): deployment, jailbreak
• TokenEconomizer: Experience-Driven Black-Box Cost Control for Coding Agents score 4
  机构: Berkeley；关键词(1): coding
• Next Token or Vector? Learning How Far to Decode score 4
  机构: Stanford；关键词(1): throughput
• Reasoning Reranking with Semantic IDs as the Sole Item Representation score 4
  机构: University of Toronto；关键词(4): distillation, fine-tuning, post-training, reasoning
• VISTA: A Multimodal Egocentric Benchmark for Goal-Oriented Perception and Assistance to Blind and Low-Vision Users score 4
  机构: Harvard；关键词(2): lightweight, vision-language
• Many Biases, One Circuit: Post-Training Tunes a Pretrained Cue-Routing Mechanism score 4
  机构: Carnegie Mellon；关键词(2): fine-tuning, post-training
• ANCHOR: Audio-Visually Grounded Chain-of-Thought Reasoning Benchmark score 4
  机构: Carnegie Mellon；关键词(1): reasoning
• From GAN-Based Models to GPT-5: A Taxonomy-Guided Study of Human Performance in Detecting Deepfake Face Images score 3
  机构: Cambridge