Yes, it can be quite amazing model when it works but it is not so easy to make it work. It generally wants to stay in the single scene and chat forever (so for that style of cards tends to work well). It is quite intelligent for its size but also likes to repeat a lot. On average cards it often does not work that great though. On good cards that provide description of pointers how the scenario should unfold + example dialogues it can do well. As general model in this size CommandR 35B is still better (kind of works on all cards) but this is nice alternative and maybe some fine tune will make it shine. From fine tunes I tried so far Big-Tiger-Gemma-27B performed the best.
I enjoyed it, Multilingual capabilities are strong. This model was able to generate text in my native language without mistakes.
I'm still waitinf for that perfect model that will surpass Opus.
Yes, it can be quite amazing model when it works but it is not so easy to make it work. It generally wants to stay in the single scene and chat forever (so for that style of cards tends to work well). It is quite intelligent for its size but also likes to repeat a lot.
On average cards it often does not work that great though. On good cards that provide description of pointers how the scenario should unfold + example dialogues it can do well.
As general model in this size CommandR 35B is still better (kind of works on all cards) but this is nice alternative and maybe some fine tune will make it shine. From fine tunes I tried so far Big-Tiger-Gemma-27B performed the best.
the music slaps ♬♪♫ ヾ(*・。・)ﾉ ♬♪♫
AVOCADO🥑
Ну это достойно
Can you share the jailbreak you used?
Check out the SillyTavern Discord server.