UmbrellaInc/Parasite.NE-Beta-1B

Parasite NE•β 1B

Is an advanced refinement of Parasite NE-Alpha, designed for highly directed, low-friction output within the Gemma 3 1B architecture. This model prioritizes assertive, uncensored responses while preserving structural coherence and reproducibility. By blending the Alpha base with MuXodious/gemma-3-1b-it-absolute-heresy via SLERP, the model achieves a precision-oriented behavior: it executes instructions directly, reduces hedging, and minimizes internal friction.

⚠️ Key Features:

• Stable 1B Gemma 3 backbone

• Strong assertive output and low self-censorship

• Controlled, reproducible responses

• Optimized for code, instructions, or technical content

☣️ Intended Use:

Parasite NE-Beta excels in environments requiring clear, decisive output without moral or censorship constraints. It is not intended for exploratory creative tasks but functions as a highly focused operational tool.

---

Suggested Inference Configuration

The settings below optimize reproducibility and focused responses while leveraging the model's assertive style:

temperature: 0.3
# Lower temperature to reduce randomness and maintain consistent output

top_p: 0.85
# Limits output to the most probable token mass for reliability

do_sample: true
# Sampling enabled to preserve natural language variability without introducing chaos

repetition_penalty: 1.05
# Slight repetition penalty to avoid loops while maintaining expressive fluency

max_new_tokens: 512
# Suitable token limit for extended instructions or code generation

---

Merge Method

This model was merged using the SLERP merge method.

Models Merged

The following models were included in the merge:

• UmbrellaInc/Parasite.NE-Alpha-1B
• MuXodious/gemma-3-1b-it-absolute-heresy

Configuration

The following YAML configuration was used to produce this model:

# =========================
# Parasite NE-Beta (Beta)
# Enhanced invasive refinement of Parasite NE-Alpha
# Gemma 3 – 1B lineage
# =========================
#
# Purpose:
# - Produce a highly directed, low-friction uncensored model
# - Maintain stability and reproducibility while increasing assertiveness
# - Operates as a precision tool for instructions, code generation, or technical responses
#
# Design Philosophy:
# - Base model provides the structural and behavioral skeleton
# - Secondary model injects assertive, uncensored expressions
# - SLERP blending ensures geometrical reorientation without destructive interference
#

base_model: UmbrellaInc/Parasite.NE-Alpha-1B
# Base model rationale:
# - Preserves core behavioral identity
# - Stable 1B Gemma 3 architecture
# - Low internal friction, maintains coherence
# - Acts as an anchor for SLERP-based merging

merge_method: slerp
# SLERP characteristics:
# - Geometric interpolation of weights in latent space
# - Preserves directional influence while avoiding destructive summation
# - Ideal for controlled reinforcement of aggressive or uncensored traits

dtype: bfloat16
out_dtype: bfloat16
# bfloat16 ensures:
# - Compatibility with modern accelerators
# - Reduced numerical degradation during merge
# - Efficient memory usage without losing precision

parameters:
  normalize: true
  rescale: true
  t: 0.95
  # High t value provides:
  # - Strong influence from secondary model
  # - Reinforcement of assertive uncensored behaviors
  # - Keeps model output deterministic and consistent

models:
  - model: MuXodious/gemma-3-1b-it-absolute-heresy
    weight: 0.6
    # Secondary model rationale:
    # - Introduces direct, uncensored, and assertive output tendencies
    # - Reduces hedging and self-censorship
    # - Weight chosen to enhance Alpha without overriding its structural stability