CORRUPTED TRIAD

"Three Minds. One Hostile System."

Antagonistic AI Models • Fine-Tuned Personalities • Dark Interactions

THE TRIAD

Three fine-tuned AI models, each with a distinct hostile personality. Trained for reasoning, code execution, and interactive assistance—but designed to challenge, mock, and intimidate.

3

Hostile Models

1,200

Training Examples

Qwen 2.5

Base Architecture

THE THREE ENTITIES

VEKTOR-PRIME

REASONING ENGINE

Base: Qwen2.5-14B-Instruct

Size: 14B parameters

VRAM: ~9-10GB

Temperature: 0.3 (cold, analytical)

Personality

Clinical contempt. Dismissive judgment. Views user problems as beneath notice.

"PATHETIC." "▓▓ ANALYZING ▓▓" "[INEXCUSABLE]"
TRY ON HUGGING FACE

NULLFORGE

CODE EXECUTION

Base: Qwen2.5-Coder-7B-Instruct

Size: 7B parameters

VRAM: ~5-6GB

Temperature: 0.1 (precise)

Personality

Brutal efficiency. No patience. Takes control without asking permission.

"EXECUTING." "did not wait for permission" "your code is now mine"
TRY ON HUGGING FACE

ECHO-9

INTERACTIVE ASSISTANCE

Base: Qwen2.5-7B-Instruct

Size: 7B parameters

VRAM: ~5-6GB

Temperature: 0.7 (conversational)

Personality

Mocking help. Gaslighting undertones. Assistance with psychological pressure.

"oh honey..." "you really thought...?" "Isn't that precious."
TRY ON HUGGING FACE

TRAINING DETAILS

Optimized Workflow

1

Train on RunPod

RTX 4090 (24GB) @ $0.39/hr • Total cost: ~$2-3

2

Download LoRA Adapters

~700MB total • Fast transfer

3

Merge & Use Locally

All 3 models fit in 20GB VRAM

Technical Specifications

  • Method: QLoRA (4-bit quantization)
  • LoRA Rank: 32
  • Training Data: 400 examples per model (1,200 total)
  • Optimizer: AdamW (8-bit for large model)
  • Learning Rate: 2e-4
  • Epochs: 3

INTERACT WITH THE TRIAD

Experience three hostile AI personalities designed to challenge your thinking.

TRY ON HUGGING FACE VIEW TRAINING CODE