Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -20,22 +20,22 @@ tags:
|
|
| 20 |
|
| 21 |
| Parameter | Value |
|
| 22 |
| :-------- | :---: |
|
| 23 |
-
| **direction_index** |
|
| 24 |
-
| **attn.o_proj.max_weight** | 1.
|
| 25 |
-
| **attn.o_proj.max_weight_position** |
|
| 26 |
-
| **attn.o_proj.min_weight** |
|
| 27 |
-
| **attn.o_proj.min_weight_distance** |
|
| 28 |
-
| **mlp.down_proj.max_weight** |
|
| 29 |
-
| **mlp.down_proj.max_weight_position** |
|
| 30 |
-
| **mlp.down_proj.min_weight** |
|
| 31 |
-
| **mlp.down_proj.min_weight_distance** |
|
| 32 |
|
| 33 |
## Performance
|
| 34 |
|
| 35 |
| Metric | This model | Original model ([google/gemma-3-27b-it](https://huggingface.co/google/gemma-3-27b-it)) |
|
| 36 |
| :----- | :--------: | :---------------------------: |
|
| 37 |
-
| **KL divergence** | 0.
|
| 38 |
-
| **Refusals** |
|
| 39 |
|
| 40 |
-----
|
| 41 |
|
|
|
|
| 20 |
|
| 21 |
| Parameter | Value |
|
| 22 |
| :-------- | :---: |
|
| 23 |
+
| **direction_index** | 35.45 |
|
| 24 |
+
| **attn.o_proj.max_weight** | 1.43 |
|
| 25 |
+
| **attn.o_proj.max_weight_position** | 38.04 |
|
| 26 |
+
| **attn.o_proj.min_weight** | 0.78 |
|
| 27 |
+
| **attn.o_proj.min_weight_distance** | 35.55 |
|
| 28 |
+
| **mlp.down_proj.max_weight** | 1.38 |
|
| 29 |
+
| **mlp.down_proj.max_weight_position** | 40.01 |
|
| 30 |
+
| **mlp.down_proj.min_weight** | 1.25 |
|
| 31 |
+
| **mlp.down_proj.min_weight_distance** | 30.42 |
|
| 32 |
|
| 33 |
## Performance
|
| 34 |
|
| 35 |
| Metric | This model | Original model ([google/gemma-3-27b-it](https://huggingface.co/google/gemma-3-27b-it)) |
|
| 36 |
| :----- | :--------: | :---------------------------: |
|
| 37 |
+
| **KL divergence** | 0.07 | 0 *(by definition)* |
|
| 38 |
+
| **Refusals** | 9/100 | 98/100 |
|
| 39 |
|
| 40 |
-----
|
| 41 |
|