Kunoichi-DPO-v2-7-GGUF-Imatrix v BuRP_7B-GGUF-IQ-Imatrix
Both of these models are great, and are my new fav 7B models but I can't tell much difference between the two. What would you say are the key differences?
There's no clear difference to point to, I am lead to believe that Kunoichi is "smarter", for example for reasoning tasks, whereas BuRP is better at formatting responses in a roleplay scenario. The one that's better will depend on your impressions and usage.
I prefer InfinityRP and BuRP currently, because I'm more particular about text formatting, but Kunoichi is a very solid and smart model, I'd recommend all of them and let people decide on which they prefer.
I've noticed on some of my characters (sillytavern) that they will get them selves into a loop where they will become very repetitive describing the setting even though they still answer the question (if that makes sense). I've noticed this on a lot of 7B models so it might just be a limitation of 7B models in general. Do you know any 7B models this is less likely to happen too? Might be a good candidate for a merge
I don't get this issue much with InfinityRP (used Q4 and Q5), and maybe also a matter of your presets.
These are my usual presets:
https://huggingface.co/Lewdiculous/Model-Requests/tree/main/data/presets/lewdicu-3.0.1
Virt-io also has their own presets here you can also try:
https://huggingface.co/Virt-io/Irene-RP-7B/tree/main/presets
cool thanks i'll give that a try
I think the Westlake Imatrix Q8 is also worth checking out