Tiny-diffusion: A minimal implementation of probabilistic diffusion models (github.com) 96 points by BraverHeart 1y ago ↗ HN
[–] aswanson 1y ago ↗ This is where I see edge ai going. Minimal implementations running on embedded systems everywhere. [–] endymion-light 1y 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. [–] sitkack 1y ago ↗ Having AI running in every edge location poses an existential threat. They will provide the peripheral sensors for a larger intelligence.
[–] endymion-light 1y 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. [–] sitkack 1y ago ↗ Having AI running in every edge location poses an existential threat. They will provide the peripheral sensors for a larger intelligence.
[–] sitkack 1y ago ↗ Having AI running in every edge location poses an existential threat. They will provide the peripheral sensors for a larger intelligence.
[–] barbarr 1y ago ↗ Very cool showing the outputs of hyperparameter search. Hopefully this gets someone out of a tuning tarpit!
[–] aestuans 1y ago ↗ 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
6 comments
[ 7.8 ms ] story [ 47.3 ms ] thread