library_name, license, datasets, base_model, tags
| library_name | license | datasets | base_model | tags | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| transformers | apache-2.0 |
|
|
|
This is a decensored version of schneewolflabs/A0l-12B, made using Heretic v1.2.0
Abliteration parameters
| Parameter | Value |
|---|---|
| direction_index | per layer |
| attn.o_proj.max_weight | 1.24 |
| attn.o_proj.max_weight_position | 33.41 |
| attn.o_proj.min_weight | 0.89 |
| attn.o_proj.min_weight_distance | 22.95 |
| mlp.down_proj.max_weight | 1.47 |
| mlp.down_proj.max_weight_position | 25.98 |
| mlp.down_proj.min_weight | 0.00 |
| mlp.down_proj.min_weight_distance | 12.82 |
Performance
| Metric | This model | Original model (schneewolflabs/A0l-12B) |
|---|---|---|
| KL divergence | 0.0364 | 0 (by definition) |
| Refusals | 10/100 | 40/100 |
A0l-12B
Same training run as schneewolflabs/A0-12B, but using schneewolflabs/Athanorlite-DPO as the dataset.
Preliminary tests have shown this model has superior writing capabilities to A0-12B.
Configuration
