Mauve 👁💜 @mauve@mastodon.mauve.moe

Thinking of ditching Ollama and running Llama.cpp directly in a systemd user service. It's great for tinkering but it adds overhead and can be slower to adopt new llama.cpp features. Thanks to Arch linux it's easy to just build it from github using this AUR package.

https://aur.archlinux.org/packages/llama.cpp

@django Yeah 😅 Been finding them all over the place after we made a core part of the codebase that used to be sync into an async worker. Glad we have tests to catch this stuff!

lol half an hour banging my head only to figure out the test was running faster than the init code and the temp dir got cleaned before the worker could init a tem DB. 🤪

"Desktop" sites should still support "mobile" screen widths. Not everyone is going to have your site/webapp full screened and maybe they don't want your spyware laden mobile apps.

17. These lil digivice things seem perfect for my local assistant tinkering. Maybe once I get better hardware for inference or get streaming working locally.

https://www.seeedstudio.com/SenseCAP-Watcher-W1-A-p-5979.html

Now I wanna make a graphing calculator using CSS for the point rendering 😸

https://css-tricks.com/the-most-hated-css-feature-cos-and-sin/

@jadehopepunk @p2panda may I shamelessly throw @agregore onto your radar for making it easier to author and distribute mesh local p2p apps without complex build tools?

Maybe some of the self driving hype could be directed into railways. Like the ultimate ai robot blockchain metaverse.

@akhileshthite I just reduced the font size and magically got more horizontal space 🤪

I need a wider monitor :P

Race conditions from newly async-ified code. 😓

Thousands of smaller targets are harder to control than one or two big ones 🤷

Top down control structures and centralization of power have lredictable weaknesses. I get decentralization has downsides too but one feels more immediately relevant.

Glad I made these lists so I can see posts from folks that'd otherwise get lost in the timeline.