It’s something to do with improving the reasoning capabilities, because there are more patterns in the weights that it can activate against

Supposedly that makes them dumber in general

With your bridging to discord, do you have dm capabilities? Able to join servers you don’t control? Because I’m nervous about the idea of getting banned for puppeting since I have multiple big servers I’m in charge of

I think that’s a Voyager thing, maybe?

The main trouble with Stoat and Fluxer from what I’ve seen is that they’re both trying too hard to be Discord, while neither of them are quite hitting the mark. They’ll be interesting to follow in the future

Arch since 2018, mint for a few years before that, Ubuntu from 2008-2014

In your defense, I’ve thought the same joke every time I’ve seen it lol

Funny to see someone else with an active distaste for his videos. He sets off predatory alarm bells in my head and feels smarmy to me.

I just run the llama-swap docker container with a config file mounted, set to listen for config changes so I don’t have to restart it to add new models. I don’t have a guide besides the README for llama-swap.

Especially from a 7b model

run a local LLM like Claude!

Look inside

“Run ollama”

Ollama will almost always be slower than running vllm or llama.cpp, nobody should be suggesting it for anything agentic. On most consumer hardware, the availability of llama.cpp’s --cpu-moe flag alone is absurdly good and worth the effort to familiarize yourself with llamacpp instead of ollama.

I’ve used Xpra for similar

A perfect image to post to catbox

I use Chezmoi and a git repo for keeping mine in sync with each other across several machines.

I miss the old days of using reddit via rtv, someone should update it for lemmy

Unless the point is to have a single device and not to carry a bunch of peripherals around? “Better” only if your criteria ignores the need to carry it around with the laptop.