This is a merge of pre-trained language models created using
mergekit.
My goal was to tune down the
horniness of Forgotten Safeword with Lyra, with a focus on more general roleplaying too.
I use a modified version of
Mistral-V3-Tekken-E
But feel free to play around with it.
1base_model: Sao10K/MN-12B-Lyra-v4
2dtype: bfloat16
3merge_method: arcee_fusion
4modules:
5 default:
6 slices:
7 - sources:
8 - layer_range: [0, 40]
9 model: Sao10K/MN-12B-Lyra-v4
10 - layer_range: [0, 40]
11 model: ReadyArt/Forgotten-Safeword-12B-v4.0
