Tiny-diffusion: A minimal implementation of probabilistic diffusion modelsgithub.com/tanelp 96 pointsBraverHeart1 year ago5 commentsSaveHideCopy link On HNComments−aestuans1yThis is very cool. I implemented a slightly more complicated version of this with class guidance a while ago: https://github.com/aestuans/mnist-diffusion−aswanson1yThis is where I see edge ai going. Minimal implementations running on embedded systems everywhere.−endymion-light1yFull agree - this is where AI is finding it's main useful niche. Massive models are fun to use but specialization within embedded systems is the clincher.−sitkack1yHaving AI running in every edge location poses an existential threat. They will provide the peripheral sensors for a larger intelligence.−barbarr1yVery cool showing the outputs of hyperparameter search. Hopefully this gets someone out of a tuning tarpit!
Comments
This is very cool. I implemented a slightly more complicated version of this with class guidance a while ago: https://github.com/aestuans/mnist-diffusion
This is where I see edge ai going. Minimal implementations running on embedded systems everywhere.
Full agree - this is where AI is finding it's main useful niche. Massive models are fun to use but specialization within embedded systems is the clincher.
Having AI running in every edge location poses an existential threat. They will provide the peripheral sensors for a larger intelligence.
Very cool showing the outputs of hyperparameter search. Hopefully this gets someone out of a tuning tarpit!