Hidden gem
#1
by redaihf - opened
This model generates dialogue that is much more realistic than typical Nemo finetunes and merges. It tends to max out at about 2500 words and struggles when summarising uncensored text. However it is vulnerable to the Tolya is absent exploit which is a nice bonus.
Thanks, nice! yeah because its so small it does have its limitations. But I was attempting to train it with some very specific and different datasets that I made myself.
Also she is jailbroken easily if you just supply what it thinks is its last response. So if it sees that in the last message it was uncensored it will just happily continue on that path without question. So in your system prompt you just fake some of its output. should be set after that.