General discussion and feedback.
Feedback is always welcome for potential issues with quants and as a way to help authors improve on the next iteration, your comments are appreciated!
hey, im a newbie to LLM's, whats the difference between Captain-Eris-Diogenes_Twilight, Captain-Eris_Twighlight and Captain-Eris_Violet? Currently using Violet and it seems to work great, but intersted in those two models too
For specifics of each iteration maybe once @Nitral-AI as the author achieves his goals he'll explain more when relevant.
But basically each is an iteration (hopefully for the better but regressions might happen during experimentation), built on top of the previous version.
You can see the merge details at the original model pages for example:
Granted it probably won't make much sense sometimes or really explain the goals, but it's a hint of where that version came from/is going.
And if Violet has been pretty good so far and you don't want to risk breaking ongoing roleplays yet you might want to hold off a little bit and sticking with it is fine.
Got it, thanks!
Great model. Is there any advice to get it to not go on forever until it hits context limit each reply?
@Morktastic I believe it shouldn't be happening with the correct preset format.
This model is honestly great, best one I used so far, it understands pretty much anything you throw at it and keeps the storytelling just fine, but it has a few common problems that all models before it had as well, and I mean 'whispering in your ear' and 'shivers down your spine' etc. which are like a plague when it comes to RP. Sadly, 12B is my context ceiling, but oh well. As much as I'd love to try some 30B+ models I'm still pretty happy with this one. I always check your page for some lower context models to try, and even though they get smarter fast, after a certain point when I started using various -maid models a few years ago it has been more or less the same in terms of wording and plot development and it gets tiring. Maybe the context and instruct presets do help, but they often do more harm than good. I'll be looking forward for somthing interesing by you. Kudos!