TurboVLA以0.9GB显存跑到32Hz

重点关注

• TurboVLA: Real-Time Vision-Language-Action Model at 32 Hz on an RTX 4090 with <1 GB VRAM score 10
  入选 HF Daily Papers；HF 热度: 123 upvotes (+4)；有代码实现；关键词(4): lightweight, latency, real-time, vision-language
• HumanCLAW: Can Vision-Language Models Act Through a Body? score 10
  入选 HF Daily Papers；HF 热度: 68 upvotes (+4)；有代码实现；关键词(2): vision-language, embodied
• OmegaUse-OfficeVal: Benchmarking LLM Agents on Long-Horizon Office-Suite Tasks with Economic Grounding score 7
  入选 HF Daily Papers；HF 热度: 8 upvotes (+2)；有代码实现

也值得关注

• FARI: Robust One-Step Inversion for Watermarking in Diffusion Models score 4
  关键词(2): lightweight, fine-tuning；顶会接收: ICLR
• Visual Credit Audit for Multimodal Spatial Reasoning score 4
  机构: Huawei；关键词(1): reasoning
• Can AI agents conduct open-ended AI research? Early evidence from two case studies score 6
  入选 HF Daily Papers；HF 热度: 14 upvotes (+3)
• StructureGS: Structure-aware Gaussian Splatting for Articulated Object Reconstruction score 3
  顶会接收: ECCV
• IMFuse: Instance-Aware Multi-Layer Fusion for LLM-Enhanced Sequential Recommendation score 3
  机构: Zhejiang University