Thank you!
It was the model that brought me to my knees, and seriously I loved this model. It's amazing! This is very similar to gpt4 and character.ai I used a long time ago.
I'm not familiar with the works cgato, but it seems that this model on l3 reveals its full potential and it's very good. Now I'll get to the moments:
-Complies with ChatML formatting as long as you write correctly, then the bot will adapt to you. Good.
-Not very hot but not very boring either, it depends on how you put the text and what you want. (I don't need beautiful and "pink" storytelling but, just simple rp-conversation.)
-Being able to do certain things like find out a character's thoughts or recognize their clothes or the scenery around them, etc. (It may seem like a small thing but not me)
-Good understanding of character card and other settings
The model has some problems but they are minor, some things will have to be worked on somewhere. For example, in rare cases text formatting is may broken but can be fixed by editing text or swiping.
Thanks Lewdiculous for the gguf and thanks Cgato for this model!
Author feedback: @cgato
@Lewdiculous redone pls 🙏
Well, that can be arranged. I'll give it try.
So, with the advent of the XTC sampler, this dramatically changes the conditions for creating merges for models with creative potential. I guess the best solution now is to create a smart and balanced model with no bias for creative writing. I mean, XTC actually makes the model more creative. Someone wrote on Reddit that the best config is as follows: MinP, DRY, XTC and maybe Smooth Sampling.
I still use TheSpice to this day as it is still a good solution in terms of human communication, cgato has really created something close to the ideal, especially with the XTC it seems more fulfilling.
Unfortunately the Stheno models disappoints me with its excessive prose that manifests itself over time in RP and in some ways it's more robotic than TheSpice. Perhaps the whole point is that fine-tuning the model for a particular task, say creative writing, changes the behavior of the model in RP as a whole and that doesn't change much with instructions. My answer on XTC is: I think it would be good right now to focus on creating a flexible and smart model rather than making it for a specific defined task. The example of cgato's TheSpice models speaks well for this