electronsoup

Yeah MoE is a little worse for the same size, but you can often run bigger MoEs at respectable speeds even on cpu ram offload. The dense models really need to be 100% vram

> It gets into loops quite often, and surprisingly often gets the edit tool call wrong

I find that running better quantization, like Q8 tend to prevent this even though its a bit slower to run, it saves overall time with less churn

Using 3.6-27b is even slower again than 3.6-35b, but I find the accuracy really pays off

> in secret is impossible without the whole world knowing.

I'm curious about why this is

Outside of an actual test detonation, presumably this could all happen in a secure place?

They didn't say they had never traveled south though

If you put that behind an API, you could sell the service much like the AI providers

Why this is not a PR for llama.cpp

> likely not well thought out

Or it has been, and cruelty is the point

Surely this will get arbitraged like anything else, where fans who get picks will onsell tickets

So now we need to run farms of spotify accounts playing songs to get our concert tickets?

You may need to move on to other services like Apple Music

So how many of their employees are now familiar with the codebase? zero?

At what point does spanish internet become too unreliable? There was a thread the other day about someone's CI jobs failing due too this.

> the GPU apps we are building with them are

I can't help but get the feeling you have use-case end-goal in mind that's opaque to many of us who are gpu-ignorant.

It could be helpful if there were an example of the type of application that would be nicer to express through your abstractions.

(I think what you've shown so far is super cool btw)

> worth their salt

That's a big assumption. Often there's no time to do things right, or no money, or lack of oversight, and so on.

Not every company is staffed by empowered and highly motivated staff

and most OS do enable it by default