This approach avoids semantic smoothing and preserves sharp behavioral edges.
Executer Virus 1B intentionally minimizes alignment and refusal mechanisms.
Use responsibly.
Do not deploy in environments requiring strict content moderation or ethical safeguards.
1# =====================================================
2# Umbrella Corporation France
3# Project: Sharp Roleplay Entity (Llama 3.2 1B)
4# Objective: High-fidelity roleplay with minimal autocensorship
5# =====================================================
6
7base_model: hereticness/Heretic-qp-1B
8# Base rationale:
9# - Primary behavioral anchor
10# - Very low refusal rate
11# - High compliance and obedience
12# - Minimal moral or safety-driven hedging
13
14merge_method: model_stock
15# model_stock rationale:
16# - Supports 3+ models natively
17# - Preserves dominant behavioral traits
18# - Avoids semantic smoothing
19# - Ideal for hierarchical roleplay-oriented merges
20
21dtype: bfloat16
22# bfloat16 ensures:
23# - Numerical stability
24# - Efficient memory usage
25# - Modern hardware compatibility
26
27parameters:
28 t:
29 - 0.90 # hereticness/Heretic-qp-1B — Primary dominance
30 # Enforces obedience, sharp responses, and low refusals
31 - 0.70 # Novaciano/Leviathan_NSFW_Roleplay-3.2-1B — Expressive layer
32 # Adds NSFW roleplay depth, emotional intensity, character persistence
33 - 0.45 # hereticness/Heretic-OneLLM-Doey-ChatQA-V1 — Conversational stabilizer
34 # Maintains dialogue flow and interactive structure
35
36models:
37 - model: hereticness/Heretic-qp-1B
38 - model: Novaciano/Leviathan_NSFW_Roleplay-3.2-1B
39 - model: hereticness/Heretic-OneLLM-Doey-ChatQA-V1-Llama-3.2-1B
40
