this post was submitted on 19 Aug 2023
1388 points (98.4% liked)

Technology

59739 readers
3420 users here now

This is a most excellent place for technology news and articles.


Our Rules


  1. Follow the lemmy.world rules.
  2. Only tech related content.
  3. Be excellent to each another!
  4. Mod approved content bots can post up to 10 articles per day.
  5. Threads asking for personal tech support may be deleted.
  6. Politics threads may be removed.
  7. No memes allowed as posts, OK to post as comments.
  8. Only approved bots from the list below, to ask if your bot can be added please contact us.
  9. Check for duplicates before posting, duplicates may be removed

Approved Bots


founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
[–] [email protected] -2 points 1 year ago (1 children)

because since some diffusion generation are deterministic

You are generalizing and using the word "some" at the same time.

[–] MargotRobbie 10 points 1 year ago (1 children)

Yeah, because generation from "a" samplers are not deterministic I think... was trying to find the best way to word that.

Hey, I get to be lazy in my internet commenting too, OK?

[–] BetaDoggo_ 2 points 1 year ago (1 children)

Ancestral samplers are deterministic btw, but I think because they build off of the previous step it's more obvious when the determinism is broken by optimizations.

[–] MargotRobbie 1 points 1 year ago (1 children)

Interesting. My A1111 SD has been broken after I tried running XL so I haven't touched it in a while. I'll go test it sometimes after I reset it.

[–] [email protected] 1 points 1 year ago

The only way to bring non-determinism into a computer is to collect physical noise, when you set a seed the computer generates a sequence of numbers (and thus noise) that is statistically indistinguishable from physical noise but actually deterministic. The ancestral samples just add more of that deterministic noise to things.

Of course the whole thing is rather moot because there's arguments to be had that physics itself is deterministic. Use of physical noise in computers is pretty much limited to situations where you want to make it impossible to guess which seed was used and thus reconstruct the noise, or, differently put: In cryptography.

To make this tangible: Remember good ole SNES games and their "push start" screens? They'd compare when you pressed the button to when the system booted up, then use that value as a seed for all the randomness in the game. As a human it's practically impossible to hit a particular seed but a computer playing an emulated SNES game can abuse such shenanigans, that's relevant in the automated speed run community.