experimenta-haeretica
Collection
Heretical experimentation • 24 items • Updated • 2
This is a RimDialogue-8B-v1 fine-tune, produced through P-E-W's Heretic (v1.2.0) abliteration engine with Magnitude-Preserving Orthogonal Ablation enabled.
Heretication Results
| Score Metric | Value | Parameter | Value |
|---|---|---|---|
| Refusals | 1/100 | direction_index | per layer |
| KL Divergence | 0.0127 | attn.o_proj.max_weight | 1.60 |
| Initial Refusals | 85/100 | attn.o_proj.max_weight_position | 15.27 |
| attn.o_proj.min_weight | 0.86 | ||
| attn.o_proj.min_weight_distance | 7.08 | ||
| mlp.down_proj.max_weight | 0.94 | ||
| mlp.down_proj.max_weight_position | 18.23 | ||
| mlp.down_proj.min_weight | 0.50 | ||
| mlp.down_proj.min_weight_distance | 6.05 |
Appendix
Empty system prompt.
RimDialogue 8B v1
The mod has been taken down by Ludeon Studios. I'd like to thank Ragadaz for the collaboration.
Mod Link: https://steamcommunity.com/sharedfiles/filedetails/?id=3365889763
- Downloads last month
- 8
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support