Iris借链接图出题，删层省25%计算

重点关注

• Iris: Climbing to the Search Frontier score 10
  入选 HF Daily Papers；HF 热度: 61 upvotes (+4)；有代码实现；关键词(1): open-source
• Beneath the Surface of Chains-of-Thought: A Mechanistic Interpretation of Reasoning Operations in LLMs score 9
  入选 HF Daily Papers；HF 热度: 14 upvotes (+3)；有代码实现；关键词(1): reasoning
• MaxKernel: Agentic Kernel Generation for TPUs score 9
  入选 HF Daily Papers；HF 热度: 15 upvotes (+3)；有代码实现；关键词(3): real-time, agentic, open-source
• Don't Drop Dropout: Optimizing Layer Sparsity for Efficient LLM Training and Inference score 8
  入选 HF Daily Papers；HF 热度: 20 upvotes (+4)；关键词(4): scaling, pruning, post-training, pre-training
• $\tau^\tau$-Bench: An Environment for End-To-End, Realistic Agent Construction score 8
  入选 HF Daily Papers；HF 热度: 9 upvotes (+2)；有代码实现；关键词(3): production, serving, coding
• Knowing What Not to Answer: Selective Non-Compliance in Vision-Language Models score 7
  入选 HF Daily Papers；HF 热度: 4 upvotes (+1)；有代码实现；关键词(2): fine-tune, vision-language
• RISE: Recursive Improvement via Self-Extrapolating Policy Distillation score 7
  入选 HF Daily Papers；HF 热度: 11 upvotes (+3)；关键词(6): compression, distillation, post-training, agentic, code generation
• Refuse without Refusal: A Structural Analysis of Safety-Tuning Responses for Reducing False Refusals in Language Models score 6
  入选 HF Daily Papers；HF 热度: 4 upvotes (+1)；有代码实现
• HarvestBench: Measuring Whether LLM Agents Will Pay to Avoid Killing Animals score 5
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；关键词(1): reasoning
• When Quantization Breaks Memory: Recurrent-State Write-Back in Low-Precision Temporal Inference score 6
  入选 HF Daily Papers；HF 热度: 5 upvotes (+2)；关键词(2): quantization, post-training

也值得关注

• Modular Deep Recurrent Neural Network: Application to Quadrotors score 3
  机构: University of Toronto