Never said we should not try, nor that it was pointless.

I just said it was impossible to avoid, but I agree we should not make it effortless

Where on that page do they claim that?

I am a dev (unrelated to this) and I do think it is impossible, on what grounds do you think it isn’t?

Honestly I usually don’t use either sleep or any suspensions. I prefer to turn the device off, since turning it on will take less than 20 seconds anyway.

It was also a lot less supported some time ago so that is the reason I avoided it originally.

I feel like these weird sleep things are some windows heritage to avoid reboot updates and edge browser queries or whatever other crap flies over there.

On llama.cpp the context buffer is allocated at start, so you mist be using something else. On llama-server if it does not complain after loading you usually are fine.