🥵 😏 😅
We shall see.
Hey @Lewdiculous you back to quanting? I have an interesting one on Resplendent uploaded yesterday.
Heya! I'm here for it. You can always ask for it, don't worry. I was actually just checking out Aura. Want her quants too?
"Waifu of the Week" is way too self aware, ahaha.
Yeah Aura was the one I was talking about. I would love to see it quanted, if only so I can run a 6_K imatrix myself 😭
But seriously it's a great little model. Overfit on philosophy, but it makes for a really interesting experience.
Sure!
Thanks b0ss!
@Virt-io -- Have you played with https://huggingface.co/datasets/ParasiticRogue/Bluemoon-Light for imatrix calibration data?
I was recommended by the Author, who reported better results than the usual general data in a Reddit reply.
Sorry for the late reply.
Yes I have tried it, I haven't done any actual testing on it though. I was too busy playing with samplers. Mistral-v0.2 models are too confident causing it to have serious repetition problems.
I have made a test one here
I'll update the notebook later, Warning I added in some of the multilingual examples, takes about 50 minutes to compute on colab