mergekit_config.yml

dtype: bfloat16
merge_method: passthrough
slices:
- sources:
  - layer_range: [0, 8]
    model: nvidia/Mistral-NeMo-Minitron-8B-Base
- sources:
  - layer_range: [8, 16]
    model: nvidia/Mistral-NeMo-Minitron-8B-Base
    parameters:
      scale:
      - filter: input_layernorm
        value: 0.75
      - value: 1.0
- sources:
  - layer_range: [8, 16]
    model: nvidia/Mistral-NeMo-Minitron-8B-Base
    parameters:
      scale:
      - filter: input_layernorm
        value: 0.75
      - filter: v_proj
        value: 0.5
      - filter: post_attention_layernorm
        value: 0.5
      - value: 1.0
- sources:
  - layer_range: [16, 17]
    model: nvidia/Mistral-NeMo-Minitron-8B-Base
- sources:
  - layer_range: [17, 24]
    model: nvidia/Mistral-NeMo-Minitron-8B-Base
    parameters:
      scale:
      - filter: input_layernorm
        value: 0.75
      - value: 1.0
- sources:
  - layer_range: [17, 24]
    model: nvidia/Mistral-NeMo-Minitron-8B-Base
    parameters:
      scale:
      - filter: input_layernorm
        value: 0.75
      - filter: v_proj
        value: 0.5
      - filter: post_attention_layernorm
        value: 0.5
      - value: 1.0
- sources:
  - layer_range: [24, 25]
    model: nvidia/Mistral-NeMo-Minitron-8B-Base
- sources:
  - layer_range: [25, 32]
    model: nvidia/Mistral-NeMo-Minitron-8B-Base
    parameters:
      scale:
      - filter: input_layernorm
        value: 0.75
      - filter: v_proj
        value: 0.5
      - filter: post_attention_layernorm
        value: 0.5
      - value: 1.0
- sources:
  - layer_range: [25, 40]
    model: nvidia/Mistral-NeMo-Minitron-8B-Base