models:
- model: Qwen/Qwen2.5-Coder-7B
  parameters:
    weight: 0.5
- model: Qwen/Qwen2.5-Math-7B
  parameters:
    weight: 0.5
merge_method: linear
dtype: bfloat16