12b Mag-Mell with more stable Russian output due to Aleteian's merge of Saiga and Vikhr models.
This is a merge of pre-trained language models created using
mergekit.
This model was merged using the
Linear DARE merge method using
inflatebot/MN-12B-Mag-Mell-R1 as a base.
1models:
2 - model: inflatebot/MN-12B-Mag-Mell-R1
3 - model: Aleteian/base-ground-2
4 parameters:
5 weight: 0.35
6merge_method: dare_linear
7base_model: inflatebot/MN-12B-Mag-Mell-R1
8dtype: bfloat16