This is a merge of pre-trained language models created using
mergekit.
This model was merged using the
SLERP merge method.
1model_name: "olmner-o1-7b"
2base_model: Xiaojian9992024/Qwen2.5-THREADRIPPER-Small
3merge_method: slerp
4dtype: bfloat16
5tokenizer_source: "union" # or "base" or a model path
6chat_template: "auto" # or a template name or Jinja2 template
7slices:
8 - sources:
9 - model: Xiaojian9992024/Qwen2.5-THREADRIPPER-Small
10 layer_range: [0, 28]
11 - model: marcuscedricridia/olmner-della-7b
12 layer_range: [0, 28]
13parameters:
14 t:
15 - filter: self_attn
16 value: [0.0, 0.3, 0.5, 0.7, 1.0]
17 - filter: mlp
18 value: [1.0, 0.7, 0.5, 0.3, 0.0]
19 - filter: input_layernorm|post_attention_layernorm
20 value: 0.5
21 - value: 0.5
22
