mergekit_config.yml

models:
  - model: allenai/tulu-2-dpo-7b
    # no parameters necessary for base model
  - model: tokyotech-llm/Swallow-7b-NVE-instruct-hf # Japanese language skills
    parameters:
      density: 1
      weight: 0.40
merge_method: dare_ties
base_model: allenai/tulu-2-dpo-7b
dtype: bfloat16
name: Superswallow-7b-v0.3-NVE
---
models:
  - model: tokyotech-llm/Swallow-7b-instruct-hf
    # no parameters necessary for base model
  - model: Superswallow-7b-v0.3-NVE
    parameters:
      density: 1
      weight:
      - filter: mlp
        value: 0.1
      - filter: self_attn
        value: 0.6
      - value: 0 # fallback for rest of tensors.
merge_method: dare_ties
base_model: tokyotech-llm/Swallow-7b-instruct-hf
dtype: bfloat16
tokenizer_source: union
name: Superswallow-7b-v0.3-base
---
models:
  - model: tokyotech-llm/Swallow-7b-instruct-hf
    # no parameters necessary for base model
  - model: Superswallow-7b-v0.3-NVE
    parameters:
      density: 1
      weight:
      - filter: mlp
        value: [0.6, 0.1, 0.6, 0.1, 0.6, 0.1, 0.6, 0.1, 0.1]
      - filter: self_attn
        value: [0.6, 0.6, 0.1, 0.6, 0.1, 0.6, 0.1, 0.6, 0.6]
      - value: 0 # fallback for rest of tensors.
merge_method: dare_ties
base_model: tokyotech-llm/Swallow-7b-instruct-hf
dtype: bfloat16
tokenizer_source: union
name: Superswallow-7b-v0.3-flavor
---
slices:
  - sources:
      - model: Superswallow-7b-v0.3-base
        layer_range: [0, 32]
      - model: Superswallow-7b-v0.3-flavor
        layer_range: [0, 32]
merge_method: slerp
base_model: Superswallow-7b-v0.3-base
parameters:
  t: # model stabilization
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 1]
    - filter: mlp
      value: [1, 0.5, 0.7, 0.3, 0]
    - value: 0.5 # fallback for rest of tensors
dtype: bfloat16
name: Superswallow-7b-v0.3