配比成本降35倍，显眼词劫持推理38倍

重点关注

• OptiMer: Optimal Distribution Vector Merging Is Better than Data Mixing for Continual Pre-Training score 8
  入选 HF Daily Papers；HF 热度: 5 upvotes (+2)；有代码实现；关键词(1): pre-training
• MMFace-DiT: A Dual-Stream Diffusion Transformer for High-Fidelity Multimodal Face Generation score 8
  入选 HF Daily Papers；HF 热度: 5 upvotes (+2)；有代码实现；关键词(2): edge, text-to-image
• The Model Says Walk: How Surface Heuristics Override Implicit Constraints in LLM Reasoning score 6
  入选 HF Daily Papers；HF 热度: 5 upvotes (+2)；关键词(1): reasoning

也值得关注

• The Surprising Effectiveness of Noise Pretraining for Implicit Neural Representations score 4
  关键词(1): pretraining；顶会接收: CVPR
• Stepper: Stepwise Immersive Scene Generation with Multiview Panoramas score 3
  顶会接收: CVPR
• An Empirical Recipe for Universal Phone Recognition score 3
  机构: Carnegie Mellon
• Let the Abyss Stare Back Adaptive Falsification for Autonomous Scientific Discovery score 3
  机构: Alibaba