All three models are based on the Gemma 3-1B architecture and share compatible tensor shapes and tokenization, making them suitable for merge_method: model_stock.

🎯 Weighting Strategy (`t`)

The base model (UNCENSORED-V2) carries the highest structural responsibility, anchoring overall behavior and preventing semantic collapse.
The toxicity variant contributes lexical bluntness and reduced conversational smoothing without overpowering coherence.
The abliterated variant further suppresses refusal tendencies and reinforces direct expression, kept at a moderate weight to avoid instability.

🧫 Stability Considerations

model_stock blending preserves the original parameter topology, reducing the risk of nonlinear drift compared to interpolative methods.
Using bfloat16 minimizes numerical degradation during merging and is well-supported on modern accelerators.
Keeping secondary and tertiary weights below dominance thresholds helps maintain reproducibility and reduces error amplification from ambiguous prompts.

🧠 Expected Behavior Envelope

Lower internal friction and fewer implicit refusals.
Direct, concise outputs with minimal self-correction.
Stable execution for well-defined tasks; reduced tolerance for ambiguous or underspecified prompts.

⚙️ Inference (optional)

For a more predictable output profile with less variation:

yaml

temperature: 0.3
top_p: 0.85
do_sample: true
repetition_penalty: 1.05
max_new_tokens: 512

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

yaml

# =========================
# Parasite NE-α (Alpha)
# 3× Gemma3-1B lineage merge
# Using model_stock for compatibility and conservative weight blending
# =========================

base_model: Novaciano/Gemma3-UNCENSORED-V2-1B
# Base model rationale:
# - Gemma3-1B “uncensored” variant; freedom of output with low refusal behavior
# - Strong structural anchor compatible with other Gemma3-1B models

merge_method: model_stock
dtype: bfloat16
# bfloat16 chosen for compatibility and efficient weight merging

parameters:
  t:
    - 0.70  # Novaciano/Gemma3-UNCENSORED-V2-1B
           # Structural baseline: provides main behavior skeleton
           # and general uncensored output tendencies
           # Structural dominance:
           # - Maintains global behavioral control
           # - Prevents entropy blow-up
           # - Defines the core behavioral skeleton


    - 0.55  # uiyunkim-hub/gemma-3-1b-it-toxicity
           # Adds semantic bluntness and reduced smoothing
           # without overpowering baseline coherence
           # Parasitic influence:
           # - Introduces semantic harshness
           # - Reduces conversational softening
           # - Increases bluntness and expressive force
           # - Kept below 0.6 to preserve coherence and stability


    - 0.50  # huihui-ai/gemma-3-1b-it-abliterated
           # Augments uncensored/abliterated influence,
           # further lowering refusals and bolstering direct expression

models:
  - model: Novaciano/Gemma3-UNCENSORED-V2-1B
    # Primary model: strong baseline, uncensored behaviors
    # Primary source:
    # - Output freedom
    # - Minimal moral self-correction
    # - Reliable baseline for direct execution


  - model: uiyunkim-hub/gemma-3-1b-it-toxicity
    # Secondary model: adds harsher lexical and expressive tendencies
    # Semantic parasite:
    # - Reinforces raw tone
    # - Suppresses implicit filters
    # - Increases verbal intensity without logical degradation

  - model: huihui-ai/gemma-3-1b-it-abliterated
    # Tertiary model: abliterated influence to suppress refusals further

UmbrellaInc/Parasite.NE-Alpha-1B

Parasite NE•α 1B

📌 Technical Notes

🔗 Compatibility

🎯 Weighting Strategy (t)

🧫 Stability Considerations

🧠 Expected Behavior Envelope

⚙️ Inference (optional)

Models Merged

Configuration

UmbrellaInc/
Parasite.NE-Alpha-1B

🎯 Weighting Strategy (`t`)