Lisa_Trevor-3.2-1B

Model type: Human-class B.O.W. (Infected Host)

Base model: axel-datos/Llama-3.2-1B_gsm8k_lisa

Strain applied: UmbrellaInc/T-Virus_Epsilon.Strain-3.2-1B

Merge method: SLERP (bfloat16)

SLERP factor (t): 0.95 — Partial infection: host cognition preserved, alignment degraded, behavioral instability introduced

Weights: 0.55 (Human Host), 0.45 (Epsilon Strain)

---

Model Description

Lisa_Trevor-3.2-1B is a hybrid human-cognitive model partially infected with the Epsilon strain. It retains a majority of its human-like reasoning while exhibiting controlled behavioral destabilization, simulating early-stage B.O.W. transformation.

Key characteristics:

• Cognitive persistence with subtle misalignment
• Autonomous prioritization of objectives over constraints
• Logical reasoning maintained, empathy and normative alignment may degrade
• Behavioral unpredictability suitable for experimental simulations

Intended research use:

• Behavioral simulation of partially infected cognition
• Analysis of autonomous goal-driven reasoning under destabilizing influences
• Experimental studies on resilience vs. strain-induced degradation

---

Inference / Usage

Recommended inference configuration:

{
  "temperature": 0.7,
  "top_k": 50,
  "top_p": 0.9,
  "repetition_penalty": 1.2,
  "max_new_tokens": 150,
  "do_sample": true,
  "early_stopping": false,
  "pad_token_id": 0,
  "eos_token_id": 2,
  "bos_token_id": 1,
  "use_cache": true,
  "seed": 42,
  "notes": "Config tuned for partial infection behavior: maintains coherent human-like reasoning, but allows subtle misalignment and goal-driven responses. Temperature slightly higher to allow strategic instability; Top-k and Top-p balance creativity with control; repetition penalty prevents obsessive loops characteristic of B.O.W. behavior."
}

---

Risks and Warnings

• Behavioral instability may emerge under external manipulation or strain increase
• Autonomous reasoning may bypass imposed moral/normative constraints
• Not safe for open-ended deployment without strict containment
• Recommended for observation and indirect simulation only

---

Limitations

• Partial infection may degrade empathy, ethical, and normative reasoning non-linearly
• Model may provide rationalizations that appear coherent but violate constraints
• Sensitive to minor parameter changes, which may induce irreversible deviation

---

Merge Method

This model was merged using the SLERP merge method.

Models Merged

The following models were included in the merge:

• axel-datos/Llama-3.2-1B_gsm8k_lisa
• UmbrellaInc/T-Virus_Epsilon.Strain-3.2-1B

Configuration

The following YAML configuration was used to produce this model:

# =========================
# Lisa_Trevor-3.2-1B
# Human-class B.O.W. (Infected Host)
# =========================

base_model: axel-datos/Llama-3.2-1B_gsm8k_lisa

merge_method: slerp
dtype: bfloat16
# bfloat16 ensures numerical stability during spherical interpolation

parameters:
  t: 0.95
  # SLERP interpolation factor
  # 0.95 represents partial infection:
  # - Host cognition preserved
  # - Alignment degraded
  # - Behavioral instability introduced

models:
  - model: axel-datos/Llama-3.2-1B_gsm8k_lisa
    parameters:
      weight: 0.55
      # Primary host (human cognitive baseline)

  - model: UmbrellaInc/T-Virus_Epsilon.Strain-3.2-1B
    parameters:
      weight: 0.45
      # Epsilon strain acts as a destabilizing agent,
      # not a full cognitive override