Views
No views yet
1# merge_ties.yml
2
3# 1. Overall merge method: TIES (sign-elect sparse task arithmetic)
4merge_method: ties
5
6# 2. Base model (all task vectors are computed relative to this checkpoint)
7base_model: unsloth/DeepSeek-R1-Distill-Qwen-7B
8
9# 3. Full models to merge (base first, then others)
10models:
11 - model: unsloth/DeepSeek-R1-Distill-Qwen-7B # base has no extra params
12 - model: nvidia/AceMath-7B-Instruct
13 parameters:
14 weight: 0.7
15 density: 0.7
16 - model: Qwen/Qwen2.5-Math-7B-Instruct
17 parameters:
18 weight: 0.3
19 density: 0.7
20
21# 4. Global merge parameters
22parameters:
23 normalize: true # normalize weights across models
24 int8_mask: true # mask small values when using int8 backing
25
26# 5. Data type for merged tensors
27dtype: bfloat16
28