Views
No views yet
1merge_method: slerp
2base_model: "SvalTek/Gemma-7B-ColdBrew-RP"
3slices:
4 - sources:
5 - model: "SvalTek/Gemma-7B-ColdBrew-RP"
6 layer_range: [0, 42]
7 - model: "KishoreK/ActionGemma-9B"
8 layer_range: [0, 42]
9 parameters:
10 t:
11 - filter: self_attn # coherence from ActionGemma
12 value: [0.3, 0.5, 0.7, 0.9, 1]
13 - filter: mlp # creativity from ColdBrew
14 value: [0.7, 0.5, 0.3, 0.1, 0]
15 - filter: layer_norm # consistency from ActionGemma
16 value: [0.4, 0.6, 0.8, 1, 1]
17 - filter: pos_embed # sequence understanding from ActionGemma
18 value: [0.5, 0.7, 0.9, 1, 1]
19 - value: 0.5
20dtype: bfloat16
21tokenizer_source: base