story

Tiny-diffusion: A minimal implementation of probabilistic diffusion models (opens in new tab)

github.com

96 pointsBraverHeart11mo ago5 comments

5 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

aswanson11mo ago

This is where I see edge ai going. Minimal implementations running on embedded systems everywhere.

endymion-light10mo ago

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.

sitkack10mo ago

Having AI running in every edge location poses an existential threat. They will provide the peripheral sensors for a larger intelligence.

barbarr11mo ago

Very cool showing the outputs of hyperparameter search. Hopefully this gets someone out of a tuning tarpit!

j / k navigate · click thread line to collapse