This model is lit.
The characters have attitude and give interesting replies. Maybe the original merged-in 33b is better? Not sure as it's the first I'm hearing of it, but this has more soul than the L2 70b. Scale it up.
Glad you are enjoying the model! Unfortunately scaling up 1:1 is dependent upon meta releasing a LLaMa-2-34B.
I did experiment with a 32.9B model that involved diagonal merging 65B (with Enterredaas qlora merged) onto llama-2-chat-13B which results in a roughly 33B model but the model was "oops all attention heads" and the results were unfortunately not very good. I could look for an alternate 33B model as a recipient model for the script but it's quite likely it would not end up with the same temperament as Dendrite which I feel the chat model contributes a lot to. That said I do plan to experiment with other model combinations of finetunes to see what else yields usable results and if LLaMa-2-chat-34B ever comes out I am already planning to try to scale up the project.