因果预训练让RoboCasa GR-1成功率达78.7%

重点关注

• GRACE: Generation-aware latent compression for efficient video generation score 10
  入选 HF Daily Papers；HF 热度: 75 upvotes (+4)；有代码实现；关键词(4): lightweight, compression, latency, fine-tuning
• SGF+: Decoupling Gradient Flows for Autoregressive Video Generation score 10
  入选 HF Daily Papers；HF 热度: 55 upvotes (+4)；有代码实现；关键词(1): fine-tuning
• Long-WAM: Scaling the Context of World-Action Models score 10
  入选 HF Daily Papers；HF 热度: 98 upvotes (+4)；有代码实现；关键词(4): scaling, deployment, real-time, pretraining
• Mechanics of Long-Context Hybrid Models Part 1.1: From Hybrid Attention to Hybrid Position score 10
  入选 HF Daily Papers；HF 热度: 31 upvotes (+4)；有代码实现；关键词(1): pretraining
• Inverting Multi-Vector Visual Document Indices score 8
  入选 HF Daily Papers；HF 热度: 20 upvotes (+4)；关键词(1): vision-language
• AdSpark: A Large-Scale Dataset and Benchmark for Product-Centric Advertisement Video Generation score 6
  入选 HF Daily Papers；HF 热度: 17 upvotes (+3)
• RLHND: Video Foundation Models as Physically Grounded Hand Trackers for Robot Learning score 9
  入选 HF Daily Papers；HF 热度: 28 upvotes (+4)；有代码实现
• UltraText Bench: A Comprehensive Bilingual Benchmark for Evaluating Visual Text Rendering in Image Generation score 10
  入选 HF Daily Papers；HF 热度: 106 upvotes (+4)；有代码实现；关键词(1): vision-language

也值得关注

• WAPR: A Foundation Model for Wide-Angle Refinement in Unseen Object Pose Estimation score 4
  关键词(1): throughput；顶会接收: ECCV
• Collaborative Reasoning Distillation via Cross-Feedback and Coherent Curation score 4
  关键词(2): distillation, reasoning；顶会接收: NeurIPS
• Which Language Should a Skeleton Speak? Language Choices in Multilingual Reasoning score 4
  关键词(1): reasoning；顶会接收: EMNLP
• How Do Agentic LLMs Decide to Call Tools? A Tool-Call Vector Shaped by Suppression score 4
  关键词(2): agentic, tool use；顶会接收: NeurIPS
• Understanding and Mitigating Token-Pruning-Induced Vulnerabilities in VLMs score 4
  关键词(4): compression, pruning, vision-language, jailbreak；顶会接收: ICML
• Bridge Routing Heads: Where Multilingual Multi-hop Reasoning Lives in LLMs score 4
  关键词(1): reasoning；顶会接收: EMNLP
• Judging in Latent Space: Efficient Generative Reward Modeling via Semantics-Preserving Compression score 4
  机构: Tsinghua；关键词(3): compression, fine-tuning, reasoning
• SkillForge: Co-Evolving Skills and Agents via Dynamic Skill Lifecycles score 9
  入选 HF Daily Papers；HF 热度: 8 upvotes (+2)；关键词(2): fine-tuning, agentic；顶会接收: NeurIPS
• ORCA: Hunting Compositional Failures in Text-to-Image Diffusion score 4
  关键词(1): text-to-image；顶会接收: NeurIPS
• Multi-Agent Coordination via Support-Preserving Distillation score 4
  关键词(1): distillation；顶会接收: NeurIPS
• PatchBench: Measuring Collateral Damage in Activation Patching score 4
  关键词(2): open-source, jailbreak；顶会接收: NeurIPS
• SLDR: Defending Against Malicious Fine-tuning via Selective Layers Recovery and Dynamic Routing score 4
  关键词(2): scaling, fine-tuning；顶会接收: NeurIPS
• The Symbol of the Surrogate: Measuring Numerical Provenance in Neural PDE Solvers score 3
  顶会接收: NeurIPS
• VIS-Ground: Video Interactive Storytelling with Contextual Grounding score 3
  机构: Yale
• Unified Multi-plane Autoregressive Diffusion for 3D Multi-contrast MRI Synthesis score 3
  顶会接收: ECCV