Research
Hugging Face paper proposes VFlash, fast visual planning via Flash Primitive Decoding
VFlash predicts coordinates for points, boxes and trajectories in one forward pass instead of token by token. Reported gains include about 25x faster 3D point motion forecasting in single-rollout tests and 70.2% versus 62.6% success on four MolmoSpaces task suites.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding