4步超100步基线，跳层省18%算力

重点关注

• PureCC: Pure Learning for Text-to-Image Concept Customization score 11
  入选 HF Daily Papers；HF 热度: 8 upvotes (+2)；有代码实现；关键词(1): text-to-image；顶会接收: CVPR
• Holi-Spatial: Evolving Video Streams into Holistic 3D Spatial Intelligence score 10
  入选 HF Daily Papers；HF 热度: 67 upvotes (+4)；有代码实现；关键词(4): fine-tuning, reasoning, vision-language, data curation
• TDM-R1: Reinforcing Few-Step Diffusion Models with Non-Differentiable Reward score 9
  入选 HF Daily Papers；HF 热度: 12 upvotes (+3)；有代码实现；关键词(2): post-training, text-to-image
• Breaking Training Bottlenecks: Effective and Stable Reinforcement Learning for Coding Models score 7
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；有代码实现；关键词(3): GRPO, code generation, coding
• MWM: Mobile World Models for Action-Conditioned Consistent Prediction score 6
  入选 HF Daily Papers；有代码实现；关键词(5): distillation, deployment, post-training, pretraining, embodied
• Skip to the Good Part: Representation Structure & Inference-Time Layer Skipping in Diffusion vs. Autoregressive LLMs score 5
  入选 HF Daily Papers；HF 热度: 2 upvotes (+1)；关键词(2): code generation, reasoning
• Scaling Data Difficulty: Improving Coding Models via Reinforcement Learning on Fresh and Challenging Problems score 5
  入选 HF Daily Papers；HF 热度: 3 upvotes (+1)；关键词(5): scaling, GRPO, code generation, coding, data curation

也值得关注

• Do Machines Fail Like Humans? A Human-Centred Out-of-Distribution Spectrum for Mapping Error Alignment score 4
  机构: Cambridge；关键词(1): vision-language
• Compression as Adaptation: Implicit Visual Representation with Diffusion Foundation Models score 4
  机构: Cambridge；关键词(2): scaling, compression
• Helix: Evolutionary Reinforcement Learning for Open-Ended Scientific Problem Solving score 4
  关键词(1): reasoning；顶会接收: ICLR
• Scaling Test-Time Robustness of Vision-Language Models via Self-Critical Inference Framework score 4
  关键词(3): scaling, reasoning, vision-language；顶会接收: CVPR
• Scalable Training of Mixture-of-Experts Models with Megatron Core score 4
  机构: NVIDIA；关键词(4): scaling, production, MoE, open-source
• UniUncer: Unified Dynamic Static Uncertainty for End to End Driving score 4
  机构: Tsinghua；关键词(3): lightweight, deployment, throughput
• Compressed-Domain-Aware Online Video Super-Resolution score 4
  关键词(1): real-time；顶会接收: CVPR
• AR2-4FV: Anchored Referring and Re-identification for Long-Term Grounding in Fixed-View Videos score 4
  关键词(2): lightweight, latency；顶会接收: CVPR
• ArcLight: A Lightweight LLM Inference Architecture for Many-Core CPUs score 4
  机构: Tsinghua；关键词(2): lightweight, throughput
• SGI: Structured 2D Gaussians for Efficient and Compact Large Image Representation score 4
  关键词(3): scaling, lightweight, compression；顶会接收: CVPR
• Generalizing Linear Autoencoder Recommenders with Decoupled Expected Quadratic Loss score 3
  顶会接收: ICLR
• Disentangled Textual Priors for Diffusion-based Image Super-Resolution score 3
  顶会接收: CVPR
• Obliviator Reveals the Cost of Nonlinear Guardedness in Concept Erasure score 3
  顶会接收: NeurIPS
• Active Inference for Micro-Gesture Recognition: EFE-Guided Temporal Sampling and Adaptive Learning score 3
  顶会接收: CVPR
• Hide and Find: A Distributed Adversarial Attack on Federated Graph Learning score 3
  顶会接收: ICLR