models:
  - model: neihc/qwen3b-lora-finetune
    parameters:
      weight: 0.5
  - model: unsloth/Qwen2.5-Coder-3B-Instruct
    parameters:
      weight: 0.5
merge_method: linear
dtype: float16