models:
  - model: ./Qwen2.5-72B/
    parameters:
      weight: 0.4
  - model: ./Qwen2.5-72B-Instruct/
    parameters:
      weight: 0.6
merge_method: linear
dtype: bfloat16