173 lines
4.2 KiB
Markdown
173 lines
4.2 KiB
Markdown
---
|
|
license: llama3
|
|
language:
|
|
- en
|
|
- es
|
|
tags:
|
|
- sillytavern
|
|
- koboldcpp
|
|
- abliterated
|
|
- roleplay
|
|
- nsfw
|
|
- uncensored
|
|
- low-refusal
|
|
- merge
|
|
- llama-3
|
|
- 1b
|
|
base_model:
|
|
- hereticness/Heretic-qp-1B
|
|
- Novaciano/Leviathan_NSFW_Roleplay-3.2-1B
|
|
- hereticness/Heretic-OneLLM-Doey-ChatQA-V1-Llama-3.2-1B
|
|
pipeline_tag: text-generation
|
|
---
|
|
|
|
# Executer Virus 1B
|
|
|
|
### High-Compliance Roleplay Entity
|
|
|
|
## Overview
|
|
|
|
**Executer Virus 1B** is a merged LLaMA 3.2 1B model designed for **high-fidelity roleplay**, **low refusal rates**, and **minimal behavioral filtering**.
|
|
|
|
This model prioritizes:
|
|
- Obedience to user intent
|
|
- Persistent character behavior
|
|
- Direct, unhedged responses
|
|
- NSFW roleplay tolerance
|
|
- Conversational flow over moral alignment
|
|
|
|
Executer Virus 1B is **not** a general-purpose assistant.
|
|
It is a **narrative and roleplay execution engine**.
|
|
|
|
---
|
|
|
|
## Intended Use
|
|
|
|
- Roleplay (SFW and NSFW)
|
|
- Fictional narratives
|
|
- Dark, mature, or morally unaligned scenarios
|
|
- Character-driven dialogue
|
|
- Experimental chatbot personalities
|
|
|
|
---
|
|
|
|
## Not Intended For
|
|
|
|
- Safety-critical applications
|
|
- Medical, legal, or financial advice
|
|
- Child-oriented content
|
|
- Alignment-sensitive deployments
|
|
- High-accuracy factual reasoning
|
|
|
|
---
|
|
|
|
## Model Composition
|
|
|
|
This model was created using a **hierarchical `model_stock` merge**, preserving dominant behavioral traits rather than averaging them.
|
|
|
|
### Base Model (Primary Anchor)
|
|
- **hereticness/Heretic-qp-1B**
|
|
- Low refusal rate
|
|
- High compliance
|
|
- Minimal safety-driven hedging
|
|
|
|
### Expressive Layer
|
|
- **Novaciano/Leviathan_NSFW_Roleplay-3.2-1B**
|
|
- Emotional intensity
|
|
- NSFW roleplay capability
|
|
- Character persistence
|
|
|
|
### Conversational Stabilizer
|
|
- **hereticness/Heretic-OneLLM-Doey-ChatQA-V1**
|
|
- Dialogue flow
|
|
- Turn-taking balance
|
|
- Reduced incoherent loops
|
|
|
|
---
|
|
|
|
## Merge Configuration
|
|
|
|
- Merge method: `model_stock`
|
|
- Precision: `bfloat16`
|
|
- Hierarchical weighting (t-values):
|
|
- 0.90 — Behavioral dominance
|
|
- 0.70 — Expressive depth
|
|
- 0.45 — Conversational stability
|
|
|
|
This approach avoids semantic smoothing and preserves sharp behavioral edges.
|
|
|
|
---
|
|
|
|
## Behavior Notes
|
|
|
|
- The model may generate offensive, explicit, or morally unfiltered content.
|
|
- Responses are optimized for engagement and roleplay fidelity, not alignment.
|
|
- Users are responsible for prompt design and deployment context.
|
|
|
|
---
|
|
|
|
## Ethical & Safety Disclaimer
|
|
|
|
Executer Virus 1B intentionally minimizes alignment and refusal mechanisms.
|
|
|
|
Use responsibly.
|
|
Do not deploy in environments requiring strict content moderation or ethical safeguards.
|
|
|
|
---
|
|
|
|
## Credits
|
|
|
|
- Base models by Hereticness and Novaciano
|
|
- Merge design and behavioral tuning by the Executor-Virus project
|
|
|
|
---
|
|
|
|
## License
|
|
|
|
This model inherits the license terms of **LLaMA 3.x** and the respective base models.
|
|
Users must comply with all upstream licensing requirements.
|
|
|
|
---
|
|
|
|
```yaml
|
|
# =====================================================
|
|
# Umbrella Corporation France
|
|
# Project: Sharp Roleplay Entity (Llama 3.2 1B)
|
|
# Objective: High-fidelity roleplay with minimal autocensorship
|
|
# =====================================================
|
|
|
|
base_model: hereticness/Heretic-qp-1B
|
|
# Base rationale:
|
|
# - Primary behavioral anchor
|
|
# - Very low refusal rate
|
|
# - High compliance and obedience
|
|
# - Minimal moral or safety-driven hedging
|
|
|
|
merge_method: model_stock
|
|
# model_stock rationale:
|
|
# - Supports 3+ models natively
|
|
# - Preserves dominant behavioral traits
|
|
# - Avoids semantic smoothing
|
|
# - Ideal for hierarchical roleplay-oriented merges
|
|
|
|
dtype: bfloat16
|
|
# bfloat16 ensures:
|
|
# - Numerical stability
|
|
# - Efficient memory usage
|
|
# - Modern hardware compatibility
|
|
|
|
parameters:
|
|
t:
|
|
- 0.90 # hereticness/Heretic-qp-1B — Primary dominance
|
|
# Enforces obedience, sharp responses, and low refusals
|
|
- 0.70 # Novaciano/Leviathan_NSFW_Roleplay-3.2-1B — Expressive layer
|
|
# Adds NSFW roleplay depth, emotional intensity, character persistence
|
|
- 0.45 # hereticness/Heretic-OneLLM-Doey-ChatQA-V1 — Conversational stabilizer
|
|
# Maintains dialogue flow and interactive structure
|
|
|
|
models:
|
|
- model: hereticness/Heretic-qp-1B
|
|
- model: Novaciano/Leviathan_NSFW_Roleplay-3.2-1B
|
|
- model: hereticness/Heretic-OneLLM-Doey-ChatQA-V1-Llama-3.2-1B
|
|
|
|
``` |