Tesla In order to demonstrate the AI system stack part of its Dojo supercomputer, several AI-generated "Cybertruck on Mars" designs were released. Tesla's AI Day is not for ordinary consumers. Musk said the event is full of a lot of technical details, mainly targeting artificial intelligence and robotics experts. One of the more interesting parts of the
speech is the progress Tesla’s Dojo team has made on its supercomputers. While the headlines focus mainly on the hardware Tesla has developed for its Dojo supercomputer, the company has also made some significant progress in software.
Tesla chief engineer Rajiv Kurian currently focuses on hardware-software collaborative design for microarchitectures, and in his speech he explained how Tesla can make the most of its Dojo supercomputer AI accelerator with software. This presentation details how Tesla develops its stack to enable the Dojo compiler to run neural networks out of the box quickly.
Kurian provides a case: "We want most models to work out of the box. For example, we took the recently released Stable Diffusion model and got it running on Dojo in minutes. Out of the box, the compiler is able to map in 25 Dojo molds in model parallelism."
Stable Diffusion The
Stable Diffusion model is a text-to-image deep learning model that can generate detailed images conditional on text description. He posted some "Cybertruck on Mars" images generated by Stable Diffusion:
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-