Why are posts about the kv cache keep being burried?

[Kwisss]

title says it all, is it a controversial topic? since the issue and pull request tracker lists are full with nonsense?

who wants to add functions to a non-functioning engine? or who needs a 2 percent speedup when they loose 2 minutes on every turn?
llama.cpp seems to be in a broken state for a couple weeks now? over multiple platforms and cards?

is there anything the less technical people can do? I feel this project loses a lot folks that migrate to something that does work?
ATM I rather recommend ollama to new people since re-ingesting the cache on every turn might as well qualify as non functioning?

I meet other people on discord as well that say requests and issues on this topic are being ignored and burried, why?

[dlin95123]

Can you elaborate what kv cache issues you have observed?