models:
  - model: Qwen/Qwen2.5-Coder-0.5B-Instruct
    parameters:
      weight: -1.0
merge_method: passthrough
dtype: float16