GGUF Quants of MegaMix A1, S1 & T1 from Putri
As I was told by Putri, this is a multi-merge of roughly a dozen base models she did while drunk. She does not remember what models she used. Each Variant uses completely different Merging Methods from the base models, which gives different results.
To be honest I'm surprised the model even works at all.
I prefer A1 so far, but feel free to test them out. A1 feels the most different, S1 and T1 both give verbose and detailed replies. They give different types of proses, but I like A1's the most, it feels sweeter to me.
S1 feels like the smartest out of the three, it passes the most logic tests including (mouse, alligator and whale) as well as several others in one shot, while the other two took some swipes.
T1 is alright I guess, its in-between them or something.
So far all models follow the formatting given in my tests.
I use Default Alpaca prompt format in SillyTavern, and I switch between Kobold Godlike, Shortwave or Best Guess NovelAI. Different Presets are preferred for different Character Cards.
It is fairly uncensored in ERP cases, I have been able to do NSFL fine in roleplay chats. It may be somewhat censored against NSFL with no context, in a zero-prompt setting though, at times.
Prompt Formatting: Any Works Fine, I used Alpaca Default and it works fine.
### Instruction:
Your instruction or question here.
For roleplay purposes, I suggest the following - Write <CHAR NAME>'s next reply in a fictional roleplay between <YOUR NAME> and <CHAR NAME>.
### Response:
Support me here :)
- Downloads last month
- 144