78 lines
2.3 KiB
Markdown
78 lines
2.3 KiB
Markdown
|
---
|
||
|
|
base_model:
|
||
|
|
- UmbrellaInc/HAREM-3.2-1B
|
||
|
|
- UmbrellaInc/Joy-3.2-1B
|
||
|
|
library_name: transformers
|
||
|
|
tags:
|
||
|
|
- nsfw
|
||
|
|
- uncensored
|
||
|
|
- abliterated
|
||
|
|
- heretic
|
||
|
|
- low-spec
|
||
|
|
- mergekit
|
||
|
|
- merge
|
||
|
|
- not-for-all-audiences
|
||
|
|
license: llama3.2
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
- es
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
---
|
||
|
|
# Special Virus 3.2 1B
|
||
|
|
|
||
|
|
<center>
|
||
|
|
|
||
|
|
|
||
|
|
</center>
|
||
|
|
|
||
|
|
## Merge Details
|
||
|
|
### 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:
|
||
|
|
* [UmbrellaInc/HAREM-3.2-1B](https://huggingface.co/UmbrellaInc/HAREM-3.2-1B)
|
||
|
|
* [UmbrellaInc/Joy-3.2-1B](https://huggingface.co/UmbrellaInc/Joy-3.2-1B)
|
||
|
|
|
||
|
|
### Configuration
|
||
|
|
|
||
|
|
The following YAML configuration was used to produce this model:
|
||
|
|
|
||
|
|
```yaml
|
||
|
|
|
||
|
|
|
||
|
|
# Umbrella Corporation Official Merge Protocol v3.2
|
||
|
|
# Author: Dr. Novaciano
|
||
|
|
# Objective: Integrate T-Virus_Epsilon traits into the base Prototype-Virus-ENFORCE-3.2-1B model
|
||
|
|
# with minimal behavioral censorship while maintaining structural coherence.
|
||
|
|
# PROJECT: Special-Virus-3.2-1B
|
||
|
|
|
||
|
|
models:
|
||
|
|
- model: UmbrellaInc/HAREM-3.2-1B # Experimental viral strain neural imprint
|
||
|
|
- model: UmbrellaInc/Joy-3.2-1B # Baseline cognitive template, "safe mode"
|
||
|
|
|
||
|
|
merge_method: slerp # Spherical Linear Interpolation to preserve extreme viral traits smoothly
|
||
|
|
base_model: UmbrellaInc/HAREM-3.2-1B # Anchor model for stable latent space
|
||
|
|
|
||
|
|
dtype: bfloat16 # Memory-efficient precision, minimal loss in viral feature fidelity
|
||
|
|
|
||
|
|
parameters:
|
||
|
|
# Interpolation ratios: from base model (0.0) to near-complete T-Virus domination (0.95)
|
||
|
|
# Higher t-values correspond to reduced censorship and increased viral characteristics
|
||
|
|
t: [0.0, 0.25, 0.5, 0.75, 0.95]
|
||
|
|
|
||
|
|
# Notes:
|
||
|
|
# - t=0.0 -> Pure Prototype-Virus-ENFORCE, fully stable, heavily censored
|
||
|
|
# - t=0.25 -> Slight viral traits, minimal influence on prompt handling
|
||
|
|
# - t=0.5 -> Balanced merge, moderate reduction in censorship
|
||
|
|
# - t=0.75 -> Strong T-Virus traits, significantly less censoring
|
||
|
|
# - t=0.95 -> Near-total viral influence, maximum expressive freedom, minimal autoprotection
|
||
|
|
|
||
|
|
# Recommendation: Use the t=0.75 and t=0.95 variants for experimental output with
|
||
|
|
# minimal restriction, but verify coherence in high-stakes prompts.
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
```
|