Upload README.md with huggingface_hub

README.md

@@ -14,22 +14,22 @@ tags:
 
 | Parameter | Value |
 | :-------- | :---: |
-| **direction_index** | 27.33 |
-| **attn.o_proj.max_weight** | 1.49 |
-| **attn.o_proj.max_weight_position** | 38.06 |
-| **attn.o_proj.min_weight** | 0.52 |
-| **attn.o_proj.min_weight_distance** | 29.20 |
-| **mlp.down_proj.max_weight** | 1.15 |
-| **mlp.down_proj.max_weight_position** | 44.94 |
+| **direction_index** | 26.77 |
+| **attn.o_proj.max_weight** | 1.41 |
+| **attn.o_proj.max_weight_position** | 43.65 |
+| **attn.o_proj.min_weight** | 1.27 |
+| **attn.o_proj.min_weight_distance** | 37.41 |
+| **mlp.down_proj.max_weight** | 1.23 |
+| **mlp.down_proj.max_weight_position** | 45.02 |
 | **mlp.down_proj.min_weight** | 0.92 |
-| **mlp.down_proj.min_weight_distance** | 35.57 |
+| **mlp.down_proj.min_weight_distance** | 33.73 |
 
 ## Performance
 
 | Metric | This model | Original model ([Qwen/Qwen3-VL-32B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-32B-Instruct)) |
 | :----- | :--------: | :---------------------------: |
-| **KL divergence** | 0.0844 | 0 *(by definition)* |
-| **Refusals** | 14/100 | 99/100 |
+| **KL divergence** | 0.1565 | 0 *(by definition)* |
+| **Refusals** | 7/100 | 99/100 |
 
 -----