Since those experts aren’t activated very often, the performance hit from the slow system RAM isn’t as catastrophic as it is on dense models

Aren’t experts activated on every single token?

Is it safe to use the mcp-server-docker and web search together in llama.cpp webui? Since it doesn’t ask for permission before running tools, I’m worried about malicious search results injecting commands

Guys, instead of a popularity contest, it’d be more productive if we listed the reasons for choosing one or another.

thanks for the update!

That’s unfortunate. I understand the admin’s perspective, but most users, myself included, are likely in the dark about these backend issues. It’s frustrating to have barriers to following people in an already small community. Is this just the inevitable tradeoff of federation? If anyone knows of a less regulated piefed instance, I’m all ears.

does anyone know why fosstodon seems to be blocked on piefed.social?

AI is turning humans into a service, an on-demand API. Things are getting weird, fast.