自条件化让额外采样步产生推理收益

重点关注

• Seeing, Maintaining, and Learning: The Evolution of Multimodal Agent Memory score 13
  机构: Google；入选 HF Daily Papers；HF 热度: 50 upvotes (+4)；有代码实现；关键词(1): reasoning
• The COUNT: Counting N-grams Unlocks Scaling on Host Memory and Training-Free Adaptation score 11
  机构: Allen Institute；入选 HF Daily Papers；HF 热度: 8 upvotes (+2)；有代码实现；关键词(2): scaling, pretraining
• Any Time, Any Space: A World-State-Centric Framework for Agentic Video Generation score 10
  入选 HF Daily Papers；HF 热度: 343 upvotes (+4)；有代码实现；关键词(1): agentic
• Player-1: Learning to Play from Synthetic Experience score 10
  入选 HF Daily Papers；HF 热度: 334 upvotes (+4)；有代码实现；关键词(4): fine-tune, PPO, post-training, reasoning
• More Methods, Better Agents? A Reality Check on Self-Distillation in Agentic Reinforcement Learning score 10
  入选 HF Daily Papers；HF 热度: 82 upvotes (+4)；有代码实现；关键词(3): distillation, GRPO, agentic
• Recovering Reliable Supervision via Critique-Guided On-Policy Distillation score 10
  入选 HF Daily Papers；HF 热度: 79 upvotes (+4)；有代码实现；关键词(2): distillation, reasoning
• LatentLoop: Scaling Looped MoE Models with Asymmetric Compute Allocation score 10
  入选 HF Daily Papers；HF 热度: 67 upvotes (+4)；有代码实现；关键词(2): scaling, MoE
• Cortex: Consolidating Context into Parametric Memory score 10
  机构: Zhejiang University；入选 HF Daily Papers；HF 热度: 66 upvotes (+4)
• ActionOPD: Distilling Decisions, Not Tokens score 10
  入选 HF Daily Papers；HF 热度: 56 upvotes (+4)；有代码实现；关键词(2): distillation, reasoning
• CLASHarness: Benchmarking Multi-Agent Conflict in Computer Use via a Controllable and Verifiable Generation Pipeline score 10
  入选 HF Daily Papers；HF 热度: 33 upvotes (+4)；有代码实现；关键词(1): vision-language

也值得关注

• PN-Mix: Pair-Norm Guided Mixed-Precision Key Cache Quantization for Efficient Long-Output LLM Inference score 4
  入选 HF Daily Papers；关键词(2): quantization, reasoning
• Titan-Llama: Retrofitting Pre-trained LLMs with External Neural Memory Modules score 4
  机构: Cornell；关键词(1): post-training
• Agentic Self-Awareness: Aligning Knowing, Acting, and Holding score 4
  机构: Carnegie Mellon；关键词(1): agentic
• FinGREAT: Evaluating LLMs on Financial Graph-guided Retrieval And Typing score 4
  机构: Carnegie Mellon；关键词(1): edge
• Copy-Paste Language Models score 4
  机构: Cornell；关键词(2): lightweight, pretraining
• Verifier-Grounded Agentic RL for Provably Correct Code score 4
  机构: Amazon；关键词(2): agentic, code generation
• VLX-Go: Robust Embodied Tracking through Multi-Stage Learning and On-Policy Refinement score 4
  机构: Carnegie Mellon；关键词(3): lightweight, reasoning, embodied
• DeLoop: Repeated Actions as Self-Supervision for Long-Horizon Agents score 4
  机构: Carnegie Mellon；关键词(1): compression
• DeBoneDiT: Depth-Driven Conditional Bridge Diffusion Transformers for Bone Suppression in Chest X-Ray Images score 4
  机构: Zhejiang University；关键词(1): compression
• Self-Conditioning Enables Deep Sequential Reasoning in Flow Matching score 12
  机构: Tsinghua；入选 HF Daily Papers；HF 热度: 11 upvotes (+3)；有代码实现；关键词(1): reasoning
• Memorizing Who, Knowing How: Benchmarking Personal and Procedural Memory for Persistent Assistants score 4
  机构: Zhejiang University；关键词(1): GRPO
• Are We Overthinking One-Step Generation? score 4
  机构: Harvard；关键词(3): post-training, text-to-image, open-source
• The Wealth of Agents: Scaling Agentic AI Across Developers score 4
  机构: Princeton；关键词(2): scaling, agentic
• PrismaCoder: Pushing the Boundaries of Open Multimodal Code Intelligence through Decomposed Rubric Synthesis score 4
  机构: Shanghai AI Lab；关键词(1): coding
• Latent Compaction with Test-Time Training score 4
  机构: Tsinghua；关键词(2): compression, distillation