I finally got around to it...
This is really nice for my style of creative writing. It follows instructions really nicely and the dialogue flows really well. Nice.
@mradermacher give this a go of you haven't already, it's damn good for a 32b model.
I'm still dazzling in the greatness of nemotron-sunfall, but this is on the list since when I quanted it :) Nevertheless, you "for a 32b model" sounds a bit of a downer. But then, it's going to be fast, so... :)
I think I did it a disservice by saying "for a 32b" model. It's better than a fair few 70b models I have tried. Nemotron is so powerful it's insane. I have about 8 models now, some 9b Gemma tunes as well. It works well for me, cycling between them.
I'll definitely try, I already downloaded it a while ago :) Probably the first one to download my gguf of it :)
Can confirm that this is a pretty good model overall. The tuning doesn't seem to have damaged its brains too much, at least it's still able to do the usual tasks I need it to do (summarize chatlogs, navigate menus).
In the creative field it beats the EVA one in my humble opinion. It can take different tones for different characters (something EVA struggles with). My only "issue" with it is its slight tendency to make formatted lists out of the blue when asked questions. The base Qwen model is VERY prone to do that, and I have a feeling that beating that habit out of it would damage the model too much anyway.
It's also surprisingly receptive to the custom ChatML-Named format with "<|im_start|>UserName" and "<|im_start|>BotName" instead of assistant/user. I haven't tested with multiple characters yet, but it's kinda promising in that regard.