Based on a transcript excerpt
Sam Witteveen demonstrates NVIDIA's Dory, a 4B-parameter looped (latent recurrent) transformer from NVIDIA's Nemotron Labs that can be downloaded and run locally. The video shows how varying the number of times a layer block loops per token changes output quality and token speed, and notes NVIDIA's own composite score places the model ahead of Gemma 4 26B.
AI-generated: summaries written by AI from the linked sources. How we use AI
Source Brief
Based on a transcript excerpt: Sam Witteveen demonstrates NVIDIA's Dory, a 4B-parameter looped (latent recurrent) transformer from NVIDIA's Nemotron Labs that can be downloaded and run locally. The video shows how varying the number of times a layer block loops per token changes output quality and token speed, and notes NVIDIA's own composite score places the model ahead of Gemma 4 26B.