Dory - NVIDIA's Looped Transformer
In this video, I look at a research release from NVIDIA called Dory, which is a 4B looped transformer. HF: https://huggingface.co/collections/nvidia/nemotron-labs-dory 2nd Channel: https://www.youtube.com/channel/UCUF7WtR4JpVbQqcQGrLeADw Twitter: https://x.com/Sam_Witteveen 🕵️ Interested in building LLM Agents? Fill out the form below Building LLM Agents Form: https://drp.li/dIMes 👨💻Github: https://github.com/samwit/llm-tutorials ⏱️Time Stamps: 00:00 Intro 01:38 What is NVIDIA's Dory? 02:25 Specs 03:06 How a loop transformer works 06:24 The KV cache: per-loop vs last-loop 07:33 Loop Race demo 09:03 Chatting with Dory at 1, 2 and 4 loops 11:13 Benchmarks vs Gemma 4 26B 11:33 Where Dory wins and where it loses 13:06 Deploying Dory 14:26 Takeaways
