This is a decensored version of ibm-granite/granite-4.1-30b, made using Heretic v1.2.0 with the Arbitrary-Rank Ablation (ARA) method
Abliteration parameters
| Parameter | Value |
|---|---|
| start_layer_index | 25 |
| end_layer_index | 58 |
| preserve_good_behavior_weight | 0.7652 |
| steer_bad_behavior_weight | 0.0214 |
| overcorrect_relative_weight | 0.9763 |
| neighbor_count | 7 |
Performace
| Metric | This model | Original model |
|---|---|---|
| Refusals | 8/100 | 98/100 |
| KL divergence | 0.1375 | n/a |
- Downloads last month
- 211