This is a merge of pre-trained language models created using
mergekit.
This model was merged using the
DARE TIES merge method using
T3Q-LLM/T3Q-LLM2-FP-v1.0 as a base.
1models:
2 - model: hwkwon/S-SOLAR-10.7B-v1.5
3 parameters:
4 density: [1, 0.7, 0.1] # density gradient
5 weight: 1.0
6 - model: chihoonlee10/T3Q-ko-solar-dpo-v7.0
7 parameters:
8 density: 0.5
9 weight: [0, 0.3, 0.7, 1] # weight gradient
10merge_method: dare_ties
11base_model: T3Q-LLM/T3Q-LLM2-FP-v1.0
12parameters:
13 normalize: true
14 int8_mask: true
15dtype: float16
16
17#mergekit-yaml /path/to/config.yml ./output/directory --cuda
