Thanks my dude <3
Can confirm prompt format "should" be chatml, this was an attempt to recreate kunoichi dpo but for chatml instead of alpaca - while also using the newer version of noromaid 0.4. Thanks to @SanjiWatsuki for merging at my selfish request ;)
After testing this, and the noromaid 0.4 4bpw quant. it appears there's something wrong with the stop tokens, neither model wants to stop generating, and can confirm at least on noromaid 0.4 that this is not a problem in the gptq and gguf quants from thebloke. Hopefully it's contained to just these noromaid 0.4 based models in 4bpw, when other quants of this model drop i will report back with further findings.
I was able to confirm the problem with both the Lelantos-Maid quants i uploaded from both tabby and ooba.
There was no change in setup between these and the Noromaid models from just before. I will try to redo them completely.
Thank you so much my man, ive been trying to wrap my head around what is going on. Ive gotten really good prose in some instances, and just off the wall nonesense in others. I saw it was loading with alpaca by default for both noromaid and this and wondered if that may be contributing to the issue aswell.