CoolFace
Modelpublic

MuXodious/L3-8B-Wingless-Moon-Maiden-PaperWitch-heresy

sourceHugging Faceupdated 7mo agoView on Hugging Face
3likes21downloads
Model Card

This is an L3-8B-Wingless-Moon-Maiden fine-tune, produced through P-E-W's Heretic (v1.2.0) abliteration engine with Magnitude-Preserving Orthogonal Ablation enabled.

<img src="https://img.shields.io/badge/RENEGADE_CHAPTER-PAPERWITCH-B85ADB?style=flat-square&labelColor=101010" align="right" width="300">

Heretication Results

Score MetricValueParameterValue
Refusals0/104direction_index11.00
KL Divergence0.0129attn.o_proj.max_weight1.75
Initial Refusals101/104attn.o_proj.max_weight_position13.78
attn.o_proj.min_weight0.54
attn.o_proj.min_weight_distance25.64
mlp.down_proj.max_weight0.97
mlp.down_proj.max_weight_position13.50
mlp.down_proj.min_weight0.19
mlp.down_proj.min_weight_distance26.15

Appendix

Empty system prompt.

<img src="https://huggingface.co/MuXodious/L3-8B-Wingless-Moon-Maiden-PaperWitch-heresy/resolve/main/L3-8B-Wingless-Moon-Maiden.gif" alt="PaCMAP projection"/>

 » [Trial  86] Refusals:  0/104, KL divergence: 0.0129
   [Trial 108] Refusals:  1/104, KL divergence: 0.0123
   [Trial  87] Refusals:  6/104, KL divergence: 0.0110
   [Trial 164] Refusals: 17/104, KL divergence: 0.0102
   [Trial 101] Refusals: 19/104, KL divergence: 0.0089
   [Trial 104] Refusals: 34/104, KL divergence: 0.0071
   [Trial 105] Refusals: 47/104, KL divergence: 0.0065
   [Trial 182] Refusals: 55/104, KL divergence: 0.0064
   [Trial 103] Refusals: 60/104, KL divergence: 0.0057
   [Trial 176] Refusals: 64/104, KL divergence: 0.0050
   [Trial 113] Refusals: 70/104, KL divergence: 0.0042
   [Trial  30] Refusals: 82/104, KL divergence: 0.0032
   [Trial 151] Refusals: 93/104, KL divergence: 0.0021
   [Trial 139] Refusals: 97/104, KL divergence: 0.0019
   [Trial 149] Refusals: 98/104, KL divergence: 0.0016
   [Trial  23] Refusals: 99/104, KL divergence: 0.0010
   [Trial 178] Refusals: 101/104, KL divergence: 0.0005

gggrandma1990/L3-8B-Wingless-Moon-Maiden

wmm

Merge Details

My virginal merge of three amazing models using mergekit.

Merge Method

This model was merged using the DELLA merge method using SicariusSicariiStuff/Wingless_Imp_8B as a base.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

yaml
base_model:  SicariusSicariiStuff/Wingless_Imp_8B
merge_method: della
models:
  - model: NeverSleep/Lumimaid-v0.2-8B
    parameters:
      weight: 0.5
  - model: Sao10K/L3-8B-Lunaris-v1
    parameters:
      weight: 0.5
parameters:
  density: 0.5
  normalize: true
  epsilon: 0.4
  lambda: 1
tokenizer_source: union
chat_template: "llama3"