Tiny-diffusion: A minimal implementation of probabilistic diffusion models github.com 77 points by BraverHeart a day ago
aestuans 16 hours 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
barbarr 17 hours ago Very cool showing the outputs of hyperparameter search. Hopefully this gets someone out of a tuning tarpit!
aswanson 21 hours ago This is where I see edge ai going. Minimal implementations running on embedded systems everywhere. endymion-light 2 hours 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.
endymion-light 2 hours 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.
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
Very cool showing the outputs of hyperparameter search. Hopefully this gets someone out of a tuning tarpit!
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.
[dead]