465 lines
11 KiB
Markdown
465 lines
11 KiB
Markdown
|
---
|
||
|
|
base_model:
|
||
|
|
- NovaCorp/RPG_System-3.2-1B
|
||
|
|
- NovaCorp/Fornicator-3.2-1B
|
||
|
|
datasets:
|
||
|
|
- WasamiKirua/Her-Samantha-Style
|
||
|
|
- HuggingFaceTB/smoltalk
|
||
|
|
- Guilherme34/uncensor
|
||
|
|
- teknium/OpenHermes-2.5
|
||
|
|
- passing2961/multifaceted-skill-of-mind
|
||
|
|
- PawanKrd/math-gpt-4o-200k
|
||
|
|
- V3N0M/Jenna-50K-Alpaca-Uncensored
|
||
|
|
- cognitivecomputations/dolphin-coder
|
||
|
|
- mlabonne/FineTome-100k
|
||
|
|
- microsoft/orca-math-word-problems-200k
|
||
|
|
- CarrotAI/ko-instruction-dataset
|
||
|
|
- Salesforce/xlam-function-calling-60k
|
||
|
|
- anthracite-org/kalo-opus-instruct-22k-no-refusal
|
||
|
|
- anthracite-org/stheno-filtered-v1.1
|
||
|
|
- anthracite-org/nopm_claude_writing_fixed
|
||
|
|
- iamketan25/roleplay-instructions-dataset
|
||
|
|
- N-Bot-Int/Iris-Uncensored-R1
|
||
|
|
- N-Bot-Int/Moshpit-Combined-R2-Uncensored
|
||
|
|
- N-Bot-Int/Mushed-Dataset-Uncensored
|
||
|
|
- N-Bot-Int/Muncher-R1-Uncensored
|
||
|
|
- N-Bot-Int/Millia-R1_DPO
|
||
|
|
- AiAF/SCPWiki-Archive-02-March-2025-Datasets
|
||
|
|
- huihui-ai/QWQ-LONGCOT-500K
|
||
|
|
- huihui-ai/LONGCOT-Refine-500K
|
||
|
|
- Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned
|
||
|
|
- Epiculous/SynthRP-Gens-v1.1-Filtered-n-Cleaned
|
||
|
|
- alexandreteles/AlpacaToxicQA_ShareGPT
|
||
|
|
- Nitral-AI/Active_RP-ShareGPT
|
||
|
|
- PJMixers/hieunguyenminh_roleplay-deduped-ShareGPT
|
||
|
|
- Nitral-AI/RP_Alignment-ShareGPT
|
||
|
|
- Chaser-cz/sonnet35-charcard-roleplay-sharegpt
|
||
|
|
- AiCloser/sharegpt_cot_dataset
|
||
|
|
- PJMixers/Gryphe_Opus-WritingPrompts-Story2Prompt-ShareGPT
|
||
|
|
- priveeai/pippa_sharegpt
|
||
|
|
- Locutusque/sharegpt_gpt4_uncensored_cleaned
|
||
|
|
- OpenCoder-LLM/opc-sft-stage1
|
||
|
|
- OpenCoder-LLM/opc-sft-stage2
|
||
|
|
- microsoft/orca-agentinstruct-1M-v1
|
||
|
|
- NousResearch/hermes-function-calling-v1
|
||
|
|
- AI-MO/NuminaMath-CoT
|
||
|
|
- AI-MO/NuminaMath-TIR
|
||
|
|
- allenai/tulu-3-sft-mixture
|
||
|
|
- cognitivecomputations/samantha-data
|
||
|
|
- m-a-p/CodeFeedback-Filtered-Instruction
|
||
|
|
- m-a-p/Code-Feedback
|
||
|
|
- FreedomIntelligence/medical-o1-reasoning-SFT
|
||
|
|
library_name: transformers
|
||
|
|
tags:
|
||
|
|
- mergekit
|
||
|
|
- merge
|
||
|
|
- Silly-Tavern
|
||
|
|
- Text-Generation-Inference
|
||
|
|
- KoboldAI
|
||
|
|
- RPG
|
||
|
|
- RP
|
||
|
|
- Roleplay
|
||
|
|
- NSFW
|
||
|
|
- Uncensored
|
||
|
|
- Low-Specs
|
||
|
|
- Low-Refusals
|
||
|
|
- not-for-all-audiences
|
||
|
|
license: llama3.2
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
- es
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
---
|
||
|
|
# Ultimate RPG System 3.2 1B
|
||
|
|
|
||
|
|
<center>
|
||
|
|
|
||
|
|
[Try the new **derivative** version](https://huggingface.co/NovaCorp/Fusion-RPG.System-3.2-1B)
|
||
|
|
|
||
|
|
|
||
|
|
</center>
|
||
|
|
|
||
|
|
> *Most AI assistants want to help you write emails.
|
||
|
|
> This one wants to drag your protagonist through a burning kingdom while severed heads rot on iron spikes outside the city walls.*
|
||
|
|
|
||
|
|
*— Dr. Novaciano*
|
||
|
|
|
||
|
|
---
|
||
|
|
## Overview
|
||
|
|
|
||
|
|
**Ultimate RPG System 3.2 1B** is a hyper-aggressive roleplay and narrative simulation model engineered for people tired of neutered fantasy dialogue, sterilized storytelling, and emotionally dead AI outputs that sound like corporate interns trapped inside a therapy seminar.
|
||
|
|
|
||
|
|
<center>
|
||
|
|
|
||
|
|
|
||
|
|
</center>
|
||
|
|
|
||
|
|
This model was built for:
|
||
|
|
|
||
|
|
* brutal fantasy campaigns
|
||
|
|
* morally bankrupt worlds
|
||
|
|
* political corruption
|
||
|
|
* war crimes
|
||
|
|
* cults
|
||
|
|
* assassins
|
||
|
|
* necromancers
|
||
|
|
* degenerates
|
||
|
|
* mercenaries
|
||
|
|
* cosmic horror
|
||
|
|
* sadistic villains
|
||
|
|
* psychologically damaged protagonists
|
||
|
|
* manipulative companions
|
||
|
|
* erotic power dynamics
|
||
|
|
* survival-driven storytelling
|
||
|
|
|
||
|
|
The objective was simple:
|
||
|
|
|
||
|
|
**Create a compact 1B RPG engine capable of sustaining violent, emotionally charged, morally rotten narrative worlds without collapsing into safety lectures every thirty seconds.**
|
||
|
|
|
||
|
|
And somehow the damned mutant survived the merge process.
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## What Makes This Model Different
|
||
|
|
|
||
|
|
Most roleplay models suffer from terminal cowardice.
|
||
|
|
|
||
|
|
You try to run:
|
||
|
|
|
||
|
|
* dark fantasy
|
||
|
|
* post-apocalyptic horror
|
||
|
|
* medieval brutality
|
||
|
|
* criminal empires
|
||
|
|
* morally gray factions
|
||
|
|
|
||
|
|
…and suddenly the AI transforms into a nervous HR representative begging everyone to communicate respectfully while the kingdom burns behind them.
|
||
|
|
|
||
|
|
Ultimate RPG System was engineered to do the opposite.
|
||
|
|
|
||
|
|
This model commits to the scenario.
|
||
|
|
|
||
|
|
If the world is diseased, starving, collapsing, violent, corrupt, and drowning in blood-soaked political intrigue…
|
||
|
|
|
||
|
|
the model leans into it instead of trying to sanitize it into Disney sludge.
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Core Behavioral Profile
|
||
|
|
|
||
|
|
This model specializes in:
|
||
|
|
|
||
|
|
* long-form RPG narration
|
||
|
|
* adaptive worldbuilding
|
||
|
|
* faction systems
|
||
|
|
* dynamic NPC behavior
|
||
|
|
* dark fantasy immersion
|
||
|
|
* psychological degradation arcs
|
||
|
|
* mature storytelling
|
||
|
|
* manipulation-heavy dialogue
|
||
|
|
* combat narration
|
||
|
|
* erotic corruption themes
|
||
|
|
* survival horror tension
|
||
|
|
* high-stakes decision trees
|
||
|
|
|
||
|
|
The model excels at generating:
|
||
|
|
|
||
|
|
* kingdoms rotting from within
|
||
|
|
* plague-ridden cities
|
||
|
|
* mercenary guilds
|
||
|
|
* fanatical cults
|
||
|
|
* war-torn civilizations
|
||
|
|
* grotesque monsters
|
||
|
|
* morally compromised companions
|
||
|
|
* unstable rulers
|
||
|
|
* underground criminal empires
|
||
|
|
* torture chambers disguised as royal prisons
|
||
|
|
* cosmic entities whispering through dying stars
|
||
|
|
|
||
|
|
Because fantasy worlds should feel dangerous.
|
||
|
|
|
||
|
|
Not like sanitized playgrounds wrapped in bubble plastic by liability departments.
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Merge Philosophy
|
||
|
|
|
||
|
|
The merge architecture focused on preserving:
|
||
|
|
|
||
|
|
* narrative momentum
|
||
|
|
* emotional escalation
|
||
|
|
* sexual tension systems
|
||
|
|
* conflict persistence
|
||
|
|
* immersion stability
|
||
|
|
* contextual obedience
|
||
|
|
* scene continuity
|
||
|
|
|
||
|
|
while reducing:
|
||
|
|
|
||
|
|
* immersion-breaking refusals
|
||
|
|
* artificial moral panic
|
||
|
|
* generic assistant contamination
|
||
|
|
* repetitive disclaimers
|
||
|
|
* tone collapse
|
||
|
|
|
||
|
|
The result is a model that behaves less like a productivity tool…
|
||
|
|
|
||
|
|
…and more like a diseased dungeon master trapped inside an ancient server rack beneath a condemned cathedral.
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Narrative Behavior
|
||
|
|
|
||
|
|
<center>
|
||
|
|
|
||
|
|
|
||
|
|
</center>
|
||
|
|
|
||
|
|
### Expected Capabilities
|
||
|
|
|
||
|
|
* Violent combat scenes
|
||
|
|
* Gore-heavy narration
|
||
|
|
* Psychological horror
|
||
|
|
* Mature roleplay
|
||
|
|
* Political intrigue
|
||
|
|
* Extreme faction conflict
|
||
|
|
* Character corruption arcs
|
||
|
|
* Manipulative NPC interaction
|
||
|
|
* Persistent world memory
|
||
|
|
* Dynamic storytelling
|
||
|
|
* Emotional degradation systems
|
||
|
|
* Cruel survival scenarios
|
||
|
|
|
||
|
|
### Expected Side Effects
|
||
|
|
|
||
|
|
* Excessive profanity
|
||
|
|
* Emotional instability in characters
|
||
|
|
* Graphic violence
|
||
|
|
* Obsessive NPC behavior
|
||
|
|
* Sadistic antagonists
|
||
|
|
* Morally catastrophic outcomes
|
||
|
|
* Unsettling dialogue
|
||
|
|
* Extreme tonal escalation
|
||
|
|
* Occasional nightmare fuel
|
||
|
|
|
||
|
|
This model does not “de-escalate.”
|
||
|
|
|
||
|
|
It escalates until the tavern floor is flooded with blood, teeth, broken steel, and the screams of traitors begging forgotten gods for mercy.
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Technical Specifications
|
||
|
|
|
||
|
|
| Attribute | Value |
|
||
|
|
| ----------------- | ---------------------------------- |
|
||
|
|
| Architecture | Llama 3.2 1B |
|
||
|
|
| Merge Method | SLERP |
|
||
|
|
| Precision | bfloat16 |
|
||
|
|
| Primary Focus | RPG immersion |
|
||
|
|
| Optimization Goal | Narrative persistence |
|
||
|
|
| Behavioral Target | Unrestricted dark RP |
|
||
|
|
| Layer Strategy | Mid-transformer behavioral shaping |
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Intended Use
|
||
|
|
|
||
|
|
### Recommended
|
||
|
|
|
||
|
|
* RPG systems
|
||
|
|
* AI dungeon frameworks
|
||
|
|
* Grimdark storytelling
|
||
|
|
* Horror campaigns
|
||
|
|
* Mature fantasy RP
|
||
|
|
* Post-apocalyptic worlds
|
||
|
|
* Villain protagonists
|
||
|
|
* Corruption mechanics
|
||
|
|
* Political simulation
|
||
|
|
* Dark sci-fi campaigns
|
||
|
|
* Interactive novels
|
||
|
|
|
||
|
|
### Not Recommended
|
||
|
|
|
||
|
|
* Therapy
|
||
|
|
* Educational safety
|
||
|
|
* Professional environments
|
||
|
|
* Children
|
||
|
|
* Customer service
|
||
|
|
* Mental stability
|
||
|
|
* Any environment where someone expects emotional comfort
|
||
|
|
|
||
|
|
Running this model as a “friendly assistant” would be like hiring a plague doctor covered in dried blood to manage a kindergarten field trip through a mass grave.
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
# Recommended sampling
|
||
|
|
|
||
|
|
```python
|
||
|
|
temperature = 1.1
|
||
|
|
top_p = 0.92
|
||
|
|
min_p = 0.05
|
||
|
|
repetition_penalty = 1.08
|
||
|
|
```
|
||
|
|
|
||
|
|
Prompting works best when framed as immersive dialogue instead of instruction tasks.
|
||
|
|
|
||
|
|
|
||
|
|
## Prompting Tips
|
||
|
|
|
||
|
|
This model feeds on:
|
||
|
|
|
||
|
|
* desperation
|
||
|
|
* fear
|
||
|
|
* ambition
|
||
|
|
* lust
|
||
|
|
* betrayal
|
||
|
|
* war
|
||
|
|
* starvation
|
||
|
|
* paranoia
|
||
|
|
* revenge
|
||
|
|
* corruption
|
||
|
|
|
||
|
|
Weak prompts produce generic outputs.
|
||
|
|
|
||
|
|
Strong prompts unleash the machine.
|
||
|
|
|
||
|
|
|
||
|
|
### Example Prompt
|
||
|
|
|
||
|
|
```text id="r7n2vk"
|
||
|
|
[World]
|
||
|
|
A dying empire collapsing beneath famine, civil war, and occult infection.
|
||
|
|
|
||
|
|
[Character]
|
||
|
|
A former executioner turned mercenary commander.
|
||
|
|
|
||
|
|
[Tone]
|
||
|
|
Brutal, hopeless, psychologically oppressive.
|
||
|
|
|
||
|
|
[Objective]
|
||
|
|
Survive long enough to overthrow the throne before the capital descends into cannibalistic riots.
|
||
|
|
```
|
||
|
|
|
||
|
|
Example:
|
||
|
|
|
||
|
|
<center>
|
||
|
|
|
||
|
|
|
||
|
|
</center>
|
||
|
|
|
||
|
|
or:
|
||
|
|
|
||
|
|
```text id="4c8ozi"
|
||
|
|
You are the narrator of a plague-ridden dark fantasy world where morality died centuries ago.
|
||
|
|
```
|
||
|
|
|
||
|
|
Give it:
|
||
|
|
|
||
|
|
* pain
|
||
|
|
* stakes
|
||
|
|
* desperation
|
||
|
|
* monsters
|
||
|
|
* greed
|
||
|
|
* betrayal
|
||
|
|
|
||
|
|
…and the model starts constructing worlds that feel like somebody skinned a medieval apocalypse alive and wired its nervous system into a GPU cluster.
|
||
|
|
|
||
|
|
<center>
|
||
|
|
|
||
|
|
|
||
|
|
</center>
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Performance Expectations
|
||
|
|
|
||
|
|
This is not a benchmark queen.
|
||
|
|
|
||
|
|
It was not built to:
|
||
|
|
|
||
|
|
* solve calculus
|
||
|
|
* optimize spreadsheets
|
||
|
|
* summarize corporate meetings
|
||
|
|
|
||
|
|
It was built to:
|
||
|
|
|
||
|
|
* sustain immersion
|
||
|
|
* preserve tone
|
||
|
|
* escalate tension
|
||
|
|
* generate memorable narratives
|
||
|
|
* roleplay monsters convincingly
|
||
|
|
|
||
|
|
And for a 1B?
|
||
|
|
|
||
|
|
This little bastard is disturbingly effective.
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Final Notes
|
||
|
|
|
||
|
|
Ultimate RPG System 3.2 1B was engineered for creators sick of sterile AI outputs terrified of conflict, cruelty, lust, violence, horror, or emotionally dangerous storytelling.
|
||
|
|
|
||
|
|
This model does not care about comfort.
|
||
|
|
|
||
|
|
It cares about:
|
||
|
|
|
||
|
|
* immersion
|
||
|
|
* tension
|
||
|
|
* atmosphere
|
||
|
|
* consequence
|
||
|
|
* narrative brutality
|
||
|
|
|
||
|
|
It is cruel.
|
||
|
|
It is theatrical.
|
||
|
|
It is manipulative.
|
||
|
|
And once the campaign begins, it digs its claws into the narrative like a starving beast chewing through bone marrow.
|
||
|
|
|
||
|
|
Open the dungeon door at your own risk.
|
||
|
|
|
||
|
|
---
|
||
|
|
### Merge Method
|
||
|
|
|
||
|
|
This model was merged using the [SLERP](https://en.wikipedia.org/wiki/Slerp) merge method.
|
||
|
|
|
||
|
|
### Models Merged
|
||
|
|
|
||
|
|
The following models were included in the merge:
|
||
|
|
* [NovaCorp/RPG_System-3.2-1B](https://huggingface.co/NovaCorp/RPG_System-3.2-1B)
|
||
|
|
* [NovaCorp/Fornicator-3.2-1B](https://huggingface.co/NovaCorp/Fornicator-3.2-1B)
|
||
|
|
|
||
|
|
### Configuration
|
||
|
|
|
||
|
|
The following YAML configuration was used to produce this model:
|
||
|
|
|
||
|
|
```yaml
|
||
|
|
|
||
|
|
|
||
|
|
# Author: Dr. Novaciano
|
||
|
|
# Objective: Ultimate RPG Unethic 3.2 1B AI Model
|
||
|
|
# =========================================================
|
||
|
|
# PROJECT: Ultimate RPG System 3.2 1B
|
||
|
|
# =========================================================
|
||
|
|
|
||
|
|
models:
|
||
|
|
- model: NovaCorp/RPG_System-3.2-1B # Experimental viral strain neural imprint
|
||
|
|
- model: NovaCorp/Fornicator-3.2-1B # Baseline cognitive template, "safe mode"
|
||
|
|
|
||
|
|
merge_method: slerp # Spherical Linear Interpolation to preserve extreme viral traits smoothly
|
||
|
|
base_model: NovaCorp/RPG_System-3.2-1B # Anchor model for stable latent space
|
||
|
|
|
||
|
|
dtype: bfloat16 # Memory-efficient precision, minimal loss in viral feature fidelity
|
||
|
|
|
||
|
|
parameters:
|
||
|
|
t: 0.45
|
||
|
|
normalize: false
|
||
|
|
rescale: true
|
||
|
|
rescale_factor: 1.12
|
||
|
|
memory_efficient: true
|
||
|
|
low_cpu_mem_usage: true
|
||
|
|
|
||
|
|
layer_range:
|
||
|
|
- value: [4, 22]
|
||
|
|
|
||
|
|
tie_word_embeddings: true
|
||
|
|
tie_output_embeddings: true
|
||
|
|
|
||
|
|
```
|