Conversation

Notices

1. Embed this notice
   kaia (kaia@brotka.st)'s status on Monday, 15-Jul-2024 18:47:48 JST kaia
   AI generate image descriptions have the possibility to piss off the "you didn't add alt text" and the "ewww AI" community at the same time.
   
   so, how do I get them??
   • iced depresso and Sick Sun like this.
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Monday, 15-Jul-2024 18:49:00 JST iced depresso

     in reply to
     @kaia usearch has some models but i couldn't get their code to build on python
     kaia likes this.
   • Embed this notice
     milkposter (themilkman@shitposter.world)'s status on Monday, 15-Jul-2024 18:50:22 JST milkposter

     in reply to
     @kaia no peace only problems with those people i swear :teto_haha:
     kaia likes this.
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Monday, 15-Jul-2024 18:52:23 JST iced depresso

     in reply to

     • miauz genyau
     @mia @kaia localai is a project that presents as the commercial APIs but turns around and runs local models, so there's that too.
     
     ollama might be a supported backend
     kaia likes this.
   • Embed this notice
     miauz genyau (mia@movsw.0x0.st)'s status on Monday, 15-Jul-2024 18:52:24 JST miauz genyau

     in reply to

     • iced depresso

     @icedquinn @kaia think there are some multimodal models you can use with ollama? still want to look into that

   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Monday, 15-Jul-2024 18:53:50 JST iced depresso

     in reply to

     • miauz genyau
     @mia @kaia rip
   • Embed this notice
     miauz genyau (mia@movsw.0x0.st)'s status on Monday, 15-Jul-2024 18:53:52 JST miauz genyau

     in reply to

     • iced depresso

     @icedquinn @kaia localai is made by people who do not know how to deploy software

     kaia likes this.
   • Embed this notice
     Jan Adriaenssens (verbeeld@mastodon.social)'s status on Monday, 15-Jul-2024 18:54:16 JST Jan Adriaenssens

     in reply to

     @kaia I made my own “GPT” to give image descriptions fast and rather accurate.

     kaia likes this.
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Monday, 15-Jul-2024 19:01:44 JST iced depresso

     in reply to

     • miauz genyau
     • mr64bit
     @mia @mr64bit @kaia its been my experience people are very touchy about their build scripts
   • Embed this notice
     miauz genyau (mia@movsw.0x0.st)'s status on Monday, 15-Jul-2024 19:01:45 JST miauz genyau

     in reply to

     • iced depresso
     • mr64bit

     @mr64bit @icedquinn @kaia ollama, sort of. especially once vulkan support is merged.
     they use cmake though, and horrible shell script hacks
     
     i wonder if they'll accept a PR with meson

   • Embed this notice
     mr64bit (mr64bit@p.mr64.net)'s status on Monday, 15-Jul-2024 19:01:46 JST mr64bit

     in reply to

     • iced depresso
     • miauz genyau
     @mia @icedquinn @kaia ... is anything in the ML space?
   • Embed this notice
     on-lain ✔ᵛᵉʳᶦᶠᶦᵉᵈ (lain@lain.com)'s status on Monday, 15-Jul-2024 19:41:23 JST on-lain ✔ᵛᵉʳᶦᶠᶦᵉᵈ

     in reply to

     • iced depresso
     • miauz genyau
     @icedquinn @kaia @mia ollama already supports most of the openai api. It supports a variety of multimodal models, for image descriptions the very small llava-phi3 is adequate.