NVIDIA's Dory: a 4B looped transformer architecture explained in new video
mycall · reddit · 2026-11-05
A Reddit post shares a video explaining NVIDIA's "Dory", a 4B-parameter looped transformer architecture. Looped transformers reuse the same weight layers repeatedly to trade fewer parameters for deeper effective computation; the video walks through the architecture.
More from Research
- Meta AI proposes Off-Context GRPO to fix GRPO's zero-gradient 'learning cliff' on hard problems — ceciletamura · 2026-11-05
- Anthropic's AI-led research discovers vaults across bacteria and phages, hinting at a foundational feature of life — QuintinPope5 · 2026-11-05
- Template-based multilingual safety pipeline shows English-only evals misjudge alignment — swiss-ai · 2026-11-05
- 'Grafting' paper: SFT on donor checkpoints + weight merging beats on-policy self-distillation — AdtRaghunathan · 2026-11-05
- Physical Intelligence's RLT uses RL tokens to make robot fine manipulation 3x faster — svlevine · 2026-11-05
- FrankenSim's physics-based flute synthesis gets graded against real recordings — doodlestein · 2026-11-05