rootxhacker/mini-llama-400M

merge

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the Passthrough merge method.

Models Merged

The following models were included in the merge:

• rootxhacker/mini-Llama-70M-SFT-math
• rootxhacker/mini-Llama-70M-SFT
• rootxhacker/mini-Llama-70M-SFT-ifeval
• rootxhacker/mini-Llama-70M-SFT-COT
• rootxhacker/mini-Llama-70M-SFT-medical
• rootxhacker/mini-Llama-70M-SFT-v2
• rootxhacker/mini-Llama-70M-SFT-code

Configuration

The following YAML configuration was used to produce this model:

slices:
  - sources:
      - model: rootxhacker/mini-Llama-70M-SFT-v2  # Core reasoning
        layer_range: [0, 5]  # Full 6 layers
  - sources:
      - model: rootxhacker/mini-Llama-70M-SFT-COT
        layer_range: [0, 5]  # Full 6 layers
  - sources:
      - model: rootxhacker/mini-Llama-70M-SFT-medical
        layer_range: [0, 5]
  - sources:
      - model: rootxhacker/mini-Llama-70M-SFT-code
        layer_range: [0, 5]
  - sources:
      - model: rootxhacker/mini-Llama-70M-SFT-math
        layer_range: [0, 5]
  - sources:
      - model: rootxhacker/mini-Llama-70M-SFT-ifeval
        layer_range: [0, 4]  # 5 layers
  - sources:
      - model: rootxhacker/mini-Llama-70M-SFT-v2
        layer_range: [0, 4]  # 5 layers
  - sources:
      - model: rootxhacker/mini-Llama-70M-SFT
        layer_range: [0, 4]  # 5 layers
merge_method: passthrough
dtype: bfloat16