Kaoeiri/Magpantheonse-lark-v4x1.6.2-Cydonia-vXXX-22B-2

merge

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

Merge Details

Merge Method

This model was merged using the DARE TIES merge method using anthracite-org/magnum-v4-22b as a base.

Models Merged

The following models were included in the merge:

• TheDrummer/Cydonia-22B-v1.2
• Gryphe/Pantheon-RP-Pure-1.6.2-22b-Small
• Saxo/Linkbricks-Horizon-AI-Japanese-Superb-V1-22B
• TheDrummer/Cydonia-22B-v1.3
• allura-org/MS-Meadowlark-22B
• TheDrummer/Cydonia-22B-v1.1

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: anthracite-org/magnum-v4-22b
    parameters:
      weight: 1.0         # Magnum model for primary writing style
      density: 0.80      # Slightly lower density to allow others to contribute without overpowering

  - model: TheDrummer/Cydonia-22B-v1.3
    parameters:
      weight: 0.25       # Lower weight for creativity, preventing interference with Magnum's style
      density: 0.70      # Reduced density for balance with Magnum’s writing style

  - model: TheDrummer/Cydonia-22B-v1.2
    parameters:
      weight: 0.2        # Creativity-enhancing model with lower density to avoid overpowering
      density: 0.65      # Lower density for maximum creativity without interference

  - model: TheDrummer/Cydonia-22B-v1.1
    parameters:
      weight: 0.3        # More accuracy-focused model with slight weight and density increase
      density: 0.75      # Higher density for retaining accuracy in specifics

  - model: Gryphe/Pantheon-RP-Pure-1.6.2-22b-Small
    parameters:
      weight: 0.3        # Storytelling emphasis, but should not overshadow the primary writing style
      density: 0.70      # Moderate density to blend well with other models while contributing narrative flavor

  - model: allura-org/MS-Meadowlark-22B
    parameters:
      weight: 0.2        # Subtle creative influence, contributing to writing without overpowering
      density: 0.65      # Lower density to allow other models to guide the structure and creativity

  - model: Saxo/Linkbricks-Horizon-AI-Japanese-Superb-V1-22B
    parameters:
      weight: 0.25       # Balanced weight for integrating Japanese text generation
      density: 0.70      # Moderate density to ensure that it doesn't overwhelm the other models

merge_method: dare_ties  # Using dare_ties for smoother blending and reducing noise from distinct models
base_model: anthracite-org/magnum-v4-22b
parameters:
  density: 0.80          # General density for better balance between all models
  epsilon: 0.1           # Maximum change in drop probabilities for smooth transitions between models
  lambda: 1.1            # Scaling factor for final merged deltas
dtype: bfloat16