35B agent对标1T，异步滞后被高估

重点关注

• Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent score 10
  入选 HF Daily Papers；HF 热度: 81 upvotes (+4)；有代码实现；关键词(4): scaling, distillation, fine-tuning, agentic
• Monte Carlo Energy Aggregation for Mobile 3D Gaussian Splatting score 9
  入选 HF Daily Papers；HF 热度: 18 upvotes (+3)；有代码实现；关键词(5): compression, distillation, pruning, real-time, pre-training
• Orca: The World is in Your Mind score 8
  入选 HF Daily Papers；HF 热度: 194 upvotes (+4)；关键词(3): lightweight, pre-training, embodied
• Beyond IID: How General Are Tabular Foundation Models, Really? score 7
  入选 HF Daily Papers；HF 热度: 38 upvotes (+4)
• One-Step Gradient Delay is Not a Barrier for Large-Scale Asynchronous Pipeline Parallel LLM Pretraining score 8
  入选 HF Daily Papers；HF 热度: 20 upvotes (+4)；关键词(2): throughput, pretraining
• TACO: Tool-Augmented Credit Optimization for Agentic Tool Use score 7
  入选 HF Daily Papers；HF 热度: 17 upvotes (+3)；关键词(4): GRPO, agentic, tool use, reasoning
• Beyond Drug Discovery: The Nanotechnology Molecular Optimization (NMO) Benchmark score 7
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；有代码实现；关键词(2): pretraining, leaderboard
• The Surprising Effectiveness of Video Diffusion Models for Hand Motion Reconstruction score 7
  入选 HF Daily Papers；HF 热度: 4 upvotes (+1)；有代码实现；关键词(2): reasoning, embodied
• DreamForge-World 0.1 Preview: A Low-Compute Real-Time Controllable World Model score 7
  入选 HF Daily Papers；HF 热度: 11 upvotes (+3)；关键词(1): real-time
• Nemotron-Labs-Diffusion-Image: Advancing Masked Discrete Diffusion for High-Resolution Image Synthesis score 7
  入选 HF Daily Papers；HF 热度: 11 upvotes (+3)；关键词(1): text-to-image

也值得关注

• A Gravitational Interpretation of Fine-Tuning Reversion score 4
  入选 HF Daily Papers；关键词(1): fine-tuning
• Diagnosing and Mitigating Retrieval Bottlenecks in LLM-Based Cold-Start Recommendation score 4
  机构: Stanford；关键词(1): scaling
• Know Before You Fetch: Calibrated Retrieval-Budget Allocation for Retrieval-Augmented Generation score 4
  机构: Stanford；关键词(3): latency, retrieval-augmented, RAG
• MOPD: Multi-Teacher On-Policy Distillation for Capability Integration in LLM Post-Training score 6
  入选 HF Daily Papers；HF 热度: 8 upvotes (+2)；关键词(2): distillation, post-training