DavidAU commited on
Commit
56bf465
·
verified ·
1 Parent(s): 51d8df3

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +7 -5
README.md CHANGED
@@ -52,7 +52,7 @@ This is the one they fear.
52
 
53
  This is a multi-stage fine tune, multi-fine tune, and multi-stage merge.
54
 
55
- A Colab between myself, Nightmedia, and TeichAI (Polaris Dataset).
56
 
57
  It also contains "Fable" traces/training and some GPT5 (Polaris).
58
 
@@ -100,14 +100,16 @@ I will be posting details on why both are provided and how to use each to maximu
100
 
101
  <B>DE-CENSORING STATS</B>
102
 
103
- # This is a decensored version of [Qwen/Qwen3.5-27B](https://huggingface.co/Qwen/Qwen3.5-27B), made using [Heretic](https://github.com/p-e-w/heretic) v1.2.0 with Magnitude-Preserving Orthogonal Ablation (MPOA)
 
 
104
 
105
  ## Performance
106
 
107
- | Metric | This model | Original model ([Qwen/Qwen3.5-27B](https://huggingface.co/Qwen/Qwen3.5-27B)) |
108
  | :----- | :--------: | :---------------------------: |
109
- | **KL divergence** | 0.0653 | 0 *(by definition)* |
110
- | **Refusals** | 14/100 | 94/100 |
111
 
112
  ---
113
 
 
52
 
53
  This is a multi-stage fine tune, multi-fine tune, and multi-stage merge.
54
 
55
+ A Colab between myself, Nightmedia, TeichAI (Polaris Dataset) and trohrbaugh (heretic'ing the model).
56
 
57
  It also contains "Fable" traces/training and some GPT5 (Polaris).
58
 
 
100
 
101
  <B>DE-CENSORING STATS</B>
102
 
103
+ Special thanks to: "trohrbaugh" for Heretic'ing the model.
104
+
105
+ # This is a decensored version of [Qwen/Qwen3.6-27B](https://huggingface.co/Qwen/Qwen3.6-27B), made using [Heretic](https://github.com/p-e-w/heretic) v1.2.0+custom with the [Arbitrary-Rank Ablation (ARA)](https://github.com/p-e-w/heretic/pull/211) method
106
 
107
  ## Performance
108
 
109
+ | Metric | This model | Original model ([Qwen/Qwen3.6-27B](https://huggingface.co/Qwen/Qwen3.6-27B)) |
110
  | :----- | :--------: | :---------------------------: |
111
+ | **KL divergence** | 0.0469 | 0 *(by definition)* |
112
+ | **Refusals** | 4/100 | 99/100 |
113
 
114
  ---
115