Personal experience was terrible.
First of all, you are welcome to return, but this time the 40B model is terrible, the quality of generation is still maintained at the 13B level, and the logic and comprehension is not as good as the previous 20B model.
Hello, thanks for the feedback, I don't have a lot of ressource to finetune shit myself at the moment so I try things that don't cost too much, in this case, just a MoE merge.
My entire repo is like 90% experimental, so if you had a bad experience, I trust you and I'm sorry about it!
I hope my nexts projects will please you more.
Just my $0.02: there's plenty of models that surpass it, but it's pretty 'fine'. I ran through an entire RP with it, had fun, and the output was pretty middle of the road. It had the nice feature of being fast and not going off on weird tangents, or too much repetition. I think there's something interesting here, and I have enjoyed playing with it, even if there's "better" models. Definitely don't regret giving it a spin.
Just my $0.02: there's plenty of models that surpass it, but it's pretty 'fine'. I ran through an entire RP with it, had fun, and the output was pretty middle of the road. It had the nice feature of being fast and not going off on weird tangents, or too much repetition. I think there's something interesting here, and I have enjoyed playing with it, even if there's "better" models. Definitely don't regret giving it a spin.
If you mean compared to the 13B then I'd say it's right on the money.
i been playing with for the past hours, it is actually pretty fabulous for my use as rp character , similar to the 20b models with more eloquence, fun to play with. thank you!!