Ateron/Gemma-4-AssGuard-31B

🛡️ AssGuard

"Raise your shields, as the SLOP incoming"

⚔️ Overview

This is a thing I did for fun, experimenting here and there. But to my surprise this stuff can hit the mark and generate very funny outcomes. So I decided to let this model to live. Might steal your attention for several swipes.

Settings

I suggest to use high min_p from around 0.15-0.20. That's it all others is up to your experiments.

♻️ Mergekit Configuration

Below is the exact mergekit_config.yml recipe used to synthesize this model:

Phase 1: Asgard (dare_ties)

The place where the gods and warriors lives.

dare_ties Recipe 1

merge_method: dare_ties
base_model: F:\AI\Merge\Gemma-4-it
tokenizer_source: union
dtype: bfloat16
parameters:
  lambda: 1.0
models:
  - model: F:\AI\Merge\G4-Gutenberg
    parameters:
      density: [0.70, 0.70, 0.60, 0.60, 0.70]
      weight:
      - filter: mlp
        value: [0.40, 0.45, 0.40, 0.40, 0.40]
      - filter: self_attn
        value: [0.30, 0.40, 0.50, 0.50, 0.50]
      - value: [0.50, 0.50, 0.50, 0.50, 0.50]

  - model: F:\AI\Merge\Pantheon-Reasoning-31B-1.1
    parameters:
      density: [0.30, 0.30, 0.40, 0.35, 0.25]
      weight: 
      - filter: mlp
        value: [0.40, 0.50, 0.40, 0.30, 0.10]
      - filter: self_attn
        value: [0.40, 0.35, 0.35, 0.30, 0.10]
      - value: [0.30, 0.30, 0.40, 0.30, 0.10]  

Phase 2: Odin (model_stock)

The headmaster himself!

model_stock Recipe 2

models:
  - model: F:\AI\Merge\Equinox
  - model: F:\AI\Merge\Gemma-4-31B-storymaxxed2
  - model: F:\AI\Merge\GarnetV2
merge_method: model_stock
base_model: F:\AI\Merge\Gemma-4-it
dtype: bfloat16
tokenizer_source: union

Phase 3: AssGuard (dare_ties)

Preparation becomes!

dare_ties Recipe 3

merge_method: dare_ties
base_model: F:\AI\Merge\Gemma-4-it
tokenizer_source: union
dtype: bfloat16
parameters:
  lambda: 1.0
models:
  - model: F:\AI\Merge\Asgard
    parameters:
      density: [0.70, 0.70, 0.70, 0.70, 0.70]
      weight:
      - filter: mlp
        value: [0.50, 0.50, 0.50, 0.50, 0.50]
      - filter: self_attn
        value: [0.40, 0.40, 0.40, 0.40, 0.40]
      - value: [0.50, 0.50, 0.50, 0.50, 0.50]

  - model: F:\AI\Merge\Odin
    parameters:
      density: [0.20, 0.30, 0.40, 0.30, 0.10]
      weight: 
      - filter: mlp
        value: [0.30, 0.40, 0.40, 0.40, 0.10]
      - filter: self_attn
        value: [0.30, 0.30, 0.30, 0.30, 0.10]
      - value: [0.20, 0.30, 0.30, 0.20, 0.10]  

🤝 Special Thanks

• Google DeepMind: For providing the base model.
• The Open-Source Community: Creators and all their fine-tuned models that were used in the merge.
• Mergekit Fork: Zerofata - For making mergekit work.
• To Nimbz: This cat for being a smart fella and assisting with advices.