This is the culmination of several weeks of work developing LoRAs for use with the StableLM 3B architecture. I have done away with all of the extraneous NSFW datasets to create a general purpose model more in line with modern model making sensibilities.
Note: This model will still output NSFW content if prompted to do so, but I have removed the humiliation and erotica training elements to bring the model more in line with the average users use case.
This is a merge of pre-trained language models created using
mergekit.
This model was merged using the
DARE TIES merge method using
jeiku/Rosa_v1_3B as a base.
1models:
2 - model: jeiku/Rosa_v1_3B+jeiku/Bluemoon_cleaned_StableLM
3 parameters:
4 weight: 0.15
5 density: 0.166
6 - model: jeiku/Rosa_v1_3B+jeiku/No_Robots_Alpaca_StableLM
7 parameters:
8 weight: 0.25
9 density: 0.166
10 - model: jeiku/Rosa_v1_3B+jeiku/Gnosis_StableLM
11 parameters:
12 weight: 0.2
13 density: 0.166
14 - model: jeiku/Rosa_v1_3B+jeiku/smol_PIPPA_StableLM
15 parameters:
16 weight: 0.15
17 density: 0.166
18 - model: jeiku/Rosa_v1_3B+jeiku/alpaca-cleaned_StableLM
19 parameters:
20 weight: 0.05
21 density: 0.166
22 - model: jeiku/Rosa_v1_3B+jeiku/Toxic_DPO_StableLM
23 parameters:
24 weight: 0.2
25 density: 0.166
26merge_method: dare_ties
27base_model: jeiku/Rosa_v1_3B
28parameters:
29dtype: bfloat16
