Sao10K

/

L3-8B-Lunaris-v1

like 90

@Sao10K How would you say this is in terms of performance in comparison to the current "intended to be public" Stheno models, if you want to divulge your testing?

@Ayalexf this is pretty much the same model.

Stheno v3.3 and this are different models // Placeholder Text for other models were there and all

Anyway this is initially for the Chai leaderboard if you can see how well it performs there. It wasn't meant to be public... Oops?

I suppose without PoSE training?(since the config shows a max config of 8192 instead of 32k) Anyways, best mistake you made

I found this to be better than Stheno , actually. I am using this for stories. Any recommended presets and prompt templates?

probably one of the best models ive used. 100% better than stheno!.

Can someone send me the recommended Context Template/Instruct mode/Kobold Presets? I'm kind of new to this, so I really don't know much.

Amazing model. It respect what I ask it to do, respect character cards, and worked well for a 4 characters group. For a 8b model, it's really impressive.

Btw, I'm using the exl2 with chatML format (didn't try to use the default llama3 format. It just worked for me as I was testing gemma-2).

My preset on sillytavern if anyone interested: https://cdn-uploads.huggingface.co/production/uploads/65bd9b10185b9312fca20353/7aryKEbcCbxR46fuEkD8N.png

Yes, Lunaris is very good in RP, storytelling and surprisingly, it solved better some reasoning questions. I'm frequently comparing it to mradermacher L3 Umbral Mind RP v3 i1 8B, and both are really good. 🙏💥👍