VFlash design keys: primitive boundary alignment and position alignment for parallel decoding
ZhidingYu · x · 2026-11-06
Systematic parallel decoding and formatting design is the key to VFlash's speedup. Two principles: boundary alignment — training and decoding blocks should match complete primitive boundaries so the drafter learns full geometry rather than fragments split by fixed token windows; position alignment — each output position needs a known role before parallel prediction, since coordinate values occupy varying token counts and shift later positions. Echoes the parallel box decoding design in LocateAnything.
More from Embodied
- Travis Kalanick's new firm Atoms builds AI and robots for food, mining, transport — davidsenra · 2026-11-06
- Verne V0.5 trains a robot policy with video prediction, but runs actions-only — rohanpaul_ai · 2026-11-06
- Robot success rate doubles 23%→51% from pretraining data alone, no fine-tuning — rohanpaul_ai · 2026-11-06
- Rabbit R1 with OS3 actually handles a hospital equipment return — jesselyu · 2026-11-06
- Do Heavy Robot Arms Hurt Policy Performance? New Thread Weighs the Hidden Costs — pulkitology · 2026-11-06
- Dream-1 updated with single-arm SO-101 and single-arm Franka support — scott_e_reed · 2026-11-06