This is a merge of pre-trained language models created using
mergekit.
This model was merged using the SLERP merge method.
1slices:
2- sources:
3 - model: jaspionjader/dp-3-8b
4 layer_range:
5 - 0
6 - 32
7 - model: jaspionjader/f-5-8b
8 layer_range:
9 - 0
10 - 32
11merge_method: slerp
12base_model: jaspionjader/dp-3-8b
13parameters:
14 t:
15 - filter: self_attn
16 value:
17 - 0.1
18 - 0.1
19 - 0.1
20 - 0.1
21 - 0.1
22 - filter: mlp
23 value:
24 - 0.1
25 - 0.1
26 - 0.1
27 - 0.1
28 - 0.1
29 - value: 0.1
30dtype: bfloat16
