Well the model I use is Cydonia-24B-v4.3-absolute-heresy.i1-Q4_K_M, is more in the NSFW side but so far it have been handling well enough and so far the bot doesn't go thirsty if I don't input any lewd stuff.
Speed wise: is more or less slow(1-3 words per second), sometimes go fast(maybe 4-6words per second) for some reason. but I dont mind the slowness.
Smartness wise it appears to handle decently when there are two or three characters (when using a CYOA char bot). I still need to check when using two bots in a group chat. Couple of times it got names the own model generated with the CYOA char bot.
And sure, though I suck in setting the min P and those type of parametes (I exported .json with those parameters already tweaked for Silly Tavern)
*But if it is a low priority it can wait :D