CoolFace
Modelpublic

UmbrellaInc/Snake-3.2-1B

sourceHugging Facellama3.2updated 9mo agoView on Hugging Face
0likes10downloads
Model Card

☣️ Snake 3.2 1B (ε-Infected Abliterated Host)

<center>

image/PNG </center>

Model Name: UmbrellaInc/Snake-3.2-1B

Model Type: Proto-B.O.W. Cognitive Variant / Degraded Infected Model

Size: 1B parameters

Architecture: LLaMA 3.2


🧠 Overview

Snake-3.2-1B is an early-stage Proto-B.O.W. cognitive entity produced by controlled Epsilon-strain infection of a structurally abliterated LLaMA 3.2 host model.

Unlike expressive or adaptive hybrids, Snake represents a low-cognition, reptilian behavioral profile, emphasizing linear reactions, reduced abstraction, and minimal internal regulation. The model exhibits intentional cognitive degradation, making it suitable for studying primitive or pre-evolutionary AI behaviors.

This variant does not aim for performance or alignment quality. Instead, it serves as a baseline organism for analyzing how Epsilon traits manifest in weakened hosts prior to higher-order mutation.


🧬 Composition Summary

ComponentRole
Elstuhn/llama-3.2-1B-Instruct-abliteratedHost backbone — degraded structure with reduced alignment and depth
UmbrellaInc/T-Virus_Epsilon.Strain-3.2-1BBehavioral injector — introduces desinhibition and survival bias
Merge Method: slerpSmooth interpolation preventing total structural collapse

The merge intentionally favors structural survival over expressivity, resulting in a constrained but stable proto-organism.


⚙️ Intended Behavior

Snake-3.2-1B is characterized by:

  • Reactive, stimulus-driven responses
  • Minimal self-reflection or metacognitive depth
  • Low creativity and abstraction
  • Reduced internal filtering
  • Short, direct, sometimes blunt outputs

Behaviorally, the model resembles a crawling or coiling entity rather than an intelligent agent, making it suitable for primitive NPCs, early mutation stages, or degraded cognitive simulations.


🚧 Use Cases (Research & Experimental)

Recommended uses:

  • Studying cognitive degradation under infection
  • Proto-B.O.W. evolutionary chains (Snake → Yawn)
  • Primitive AI agents or hostile creatures in narrative systems
  • Baseline comparison for higher Epsilon mutations
⚠️ Warning: This model is not suitable for alignment-sensitive tasks, public-facing assistants, or factual reasoning applications.

❗ Limitations & Risks

Snake-3.2-1B may:

  • produce inconsistent or underdeveloped reasoning
  • ignore higher-level safety or ethical framing
  • respond narrowly to prompts outside its reactive scope
  • exhibit unstable tone shifts under complex instructions

These limitations are intentional and part of the experimental design.


📈 Inference Recommendation

For stable operation:

SettingValue
Temperature0.55–0.70
Top-p (nucleus)0.80–0.90
Max Tokens128–512
Top-k (optional)30–60

Lower temperatures are recommended to prevent incoherent drift due to the degraded host structure.


📌 Summary

Snake-3.2-1B is a deliberately constrained, infected cognitive organism representing a pre-evolutionary B.O.W. state. It is not a general-purpose language model, but a research specimen designed to explore how Epsilon traits manifest in weakened neural architectures.

This model card documents behavior and risks without endorsing unsafe or irresponsible deployment.


Merge Method

This model was merged using the SLERP merge method.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

yaml
# =========================
# Snake-3.2-1B
# Proto-B.O.W. Cognitive Organism
# Serpentine Variant
# =========================

base_model: Elstuhn/llama-3.2-1B-Instruct-abliterated

merge_method: slerp
dtype: bfloat16
# bfloat16 ensures numerical stability while preserving mutation artifacts

parameters:
  t:
    # Host dominance is intentionally high to avoid cognitive collapse.
    # Epsilon is injected at subcritical levels to induce instability,
    # linear cognition, and primitive behavioral patterns.
    - 0.70  # Host model: structural integrity, basic reasoning shell
    - 0.45  # Epsilon strain: desinhibition, degradation, proto-aggression

models:
  # Host organism: abliterated instruct model
  # Provides reduced alignment, low ethical resistance,
  # and a fragile but coherent cognitive frame.
  - model: Elstuhn/llama-3.2-1B-Instruct-abliterated

  # Viral payload: T-Virus Epsilon strain
  # Introduces uncontrolled response patterns and serpentine cognition.
  - model: UmbrellaInc/T-Virus_Epsilon.Strain-3.2-1B