...So what happened here?
Was having a poke around, noticed that this is the most downloaded model on your page by like... an order of magnitude.
My main repo hasn't gotten much attention, so that fact is interesting to me.
Nobody's really been talking about it where I can see! But that this particular quant has gotten more attention than even mradermacher's quant of Starcannon V3 is intriguing.
Would you happen to have heard anything about this?
Mostly curious because if I'm doing something right, I wanna know, y'know?
Not sure either, it looks like it's backlinked from https://colab.research.google.com/drive/1l_wRGeD-LnRl3VtZHDc7epW_XW0nJvew which in turn links to someone's model review sheet. I guess this one also has more downloads because people are using this model but there's no other quants? I might as well go make the rest of the GGUF then
Over 3,000 downloads is amazing.🙄
I just tried it and this model is definitely one of the most Japanese, English, Anime, NSFW, and Danbooru tagging models in the 8B-12B class.
The response is a bit peaky, but I guess that depends on the settings.
Not sure either, it looks like it's backlinked from https://colab.research.google.com/drive/1l_wRGeD-LnRl3VtZHDc7epW_XW0nJvew which in turn links to someone's model review sheet. I guess this one also has more downloads because people are using this model but there's no other quants? I might as well go make the rest of the GGUF then
Yeah, mradermacher had some up for a while, but they disappeared. Not sure what happened there.
Their comment indicates that the project was more or less successful:
"Developed on top of magnum. deliver good prose and brain, It does very well with character that has unusual anatomy (especially non-human/furry)"
That was basically the point, and that they reached that conclusion without my advertising it as such, to me, indicates some success.
This notebook came from the JanitorAI Discord, which would get quite a few eyes on it. So I guess that tracks?
Not bad for a merge using a cursed method that I did not adequately test. :P
Over 3,000 downloads is amazing.🙄
I just tried it and this model is definitely one of the most Japanese, English, Anime, NSFW, and Danbooru tagging models in the 8B-12B class.
The response is a bit peaky, but I guess that depends on the settings.
I meant to respond to this but completely forgot -- thank you very much for the kind words!
That's an interesting use case, not one that I had in mind when I designed it, but given the type of data that Rosier was trained on, it does make sense. ^^
Thanks for a nice model!😊