# Neuron Science and reactorworld build a Trainium path for real-time diffusion video

Amazon's Neuron Science team and reactorworld used the Neuron Kernel Interface to build a kernel-centric path for real-time autoregressive diffusion video generation on Trainium. Amazon Science says the work handles dynamic shapes, memory access, and cache management, with techniques that generalize across models.

Language: en
Time zone: America/Los_Angeles
Current time (UTC, cached as of): 2026-11-10T06:48:00.000Z
Query date (America/Los_Angeles): 2026-11-09

HTML: https://didcodexreset.com/news/7e37bac9035149f1a89ad27f.html

Content language: en
Localization state: sameLanguage

Source: Amazon Science · Published 9/25/2026, 08:35:40

Amazon's Neuron Science team and reactorworld built a kernel-centric path to real-time autoregressive diffusion video generation on Trainium using the Neuron Kernel Interface.

Amazon Science says the work addresses dynamic shapes, memory access patterns, and cache management that make these workloads difficult for generic compilers, and that the techniques generalize across models. The approach is described in [A kernel-centric path to real-time video generation on Trainium](https://www.amazon.science/blog/a-kernel-centric-path-to-real-time-video-generation-on-trainium?utm_campaign=a-kernel-centric-path-to-real-time-video-gener&utm_medium=organic-asw&utm_source=twitter&utm_content=2026-09-25-a-kernel-centric-path-to-real-time-video-gener&utm_term=2026-september).

Tags: Trainium, Video generation, Neuron, Diffusion models, Amazon Science

[View original post](https://x.com/AmazonScience/status/2103503184165490855)