Hugging Face's logo

google
/
gemma-scope-2-270m-it

SAELens
Model card Files
xet
 Community
CallumMcDougallGDM commited on
Commit
7f32963
·
verified ·
1 Parent(s): 5dc0dd7

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. attn_out_all/layer_0_width_16k_l0_big/config.json +3 -2
  2. attn_out_all/layer_0_width_262k_l0_big/config.json +3 -2
  3. attn_out_all/layer_0_width_262k_l0_small/config.json +3 -2
  4. attn_out_all/layer_10_width_16k_l0_big/config.json +3 -2
  5. attn_out_all/layer_10_width_262k_l0_big/config.json +3 -2
  6. attn_out_all/layer_11_width_262k_l0_small/config.json +3 -2
  7. attn_out_all/layer_13_width_16k_l0_big/config.json +3 -2
  8. attn_out_all/layer_13_width_16k_l0_small/config.json +3 -2
  9. attn_out_all/layer_14_width_262k_l0_small/config.json +3 -2
  10. attn_out_all/layer_15_width_16k_l0_big/config.json +3 -2
  11. attn_out_all/layer_15_width_262k_l0_big/config.json +3 -2
  12. attn_out_all/layer_17_width_16k_l0_big/config.json +3 -2
  13. attn_out_all/layer_17_width_262k_l0_small/config.json +3 -2
  14. attn_out_all/layer_2_width_262k_l0_big/config.json +3 -2
  15. attn_out_all/layer_2_width_262k_l0_small/config.json +3 -2
  16. attn_out_all/layer_3_width_16k_l0_big/config.json +3 -2
  17. attn_out_all/layer_3_width_16k_l0_small/config.json +3 -2
  18. attn_out_all/layer_4_width_16k_l0_big/config.json +3 -2
  19. attn_out_all/layer_4_width_262k_l0_big/config.json +3 -2
  20. attn_out_all/layer_7_width_262k_l0_small/config.json +3 -2
  21. attn_out_all/layer_9_width_16k_l0_small/config.json +3 -2
  22. crosscoder/layer_5_9_12_15_width_1m_l0_big/config.json +5 -4
  23. crosscoder/layer_5_9_12_15_width_1m_l0_medium/config.json +5 -4
  24. crosscoder/layer_5_9_12_15_width_262k_l0_medium/config.json +5 -4
  25. crosscoder/layer_5_9_12_15_width_524k_l0_big/config.json +5 -4
  26. crosscoder/layer_5_9_12_15_width_524k_l0_medium/config.json +5 -4
  27. crosscoder/layer_5_9_12_15_width_65k_l0_big/config.json +5 -4
  28. crosscoder/layer_5_9_12_15_width_65k_l0_medium/config.json +5 -4
  29. mlp_out/layer_12_width_16k_l0_big/config.json +3 -2
  30. mlp_out/layer_12_width_16k_l0_medium/config.json +3 -2
  31. mlp_out/layer_12_width_16k_l0_small/config.json +3 -2
  32. mlp_out/layer_12_width_262k_l0_big/config.json +3 -2
  33. mlp_out/layer_12_width_262k_l0_medium/config.json +3 -2
  34. mlp_out/layer_12_width_262k_l0_small/config.json +3 -2
  35. mlp_out/layer_12_width_65k_l0_big/config.json +3 -2
  36. mlp_out/layer_12_width_65k_l0_medium/config.json +3 -2
  37. mlp_out/layer_12_width_65k_l0_small/config.json +3 -2
  38. mlp_out/layer_15_width_16k_l0_big/config.json +3 -2
  39. mlp_out/layer_15_width_16k_l0_medium/config.json +3 -2
  40. mlp_out/layer_15_width_16k_l0_small/config.json +3 -2
  41. mlp_out/layer_15_width_262k_l0_big/config.json +3 -2
  42. mlp_out/layer_15_width_262k_l0_medium/config.json +3 -2
  43. mlp_out/layer_15_width_262k_l0_small/config.json +3 -2
  44. mlp_out/layer_15_width_65k_l0_big/config.json +3 -2
  45. mlp_out/layer_15_width_65k_l0_medium/config.json +3 -2
  46. mlp_out/layer_15_width_65k_l0_small/config.json +3 -2
  47. mlp_out/layer_5_width_16k_l0_big/config.json +3 -2
  48. mlp_out/layer_5_width_16k_l0_medium/config.json +3 -2
  49. mlp_out/layer_5_width_16k_l0_small/config.json +3 -2
  50. mlp_out/layer_5_width_262k_l0_big/config.json +3 -2
attn_out_all/layer_0_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.0.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.0.self_attn.o_proj.input",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.0.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.0.self_attn.o_proj.input",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_0_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.0.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.0.self_attn.o_proj.input",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.0.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.0.self_attn.o_proj.input",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_0_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.0.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.0.self_attn.o_proj.input",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 10,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.0.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.0.self_attn.o_proj.input",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 10,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_10_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.10.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.10.self_attn.o_proj.input",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.10.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.10.self_attn.o_proj.input",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_10_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.10.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.10.self_attn.o_proj.input",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.10.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.10.self_attn.o_proj.input",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_11_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.11.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.11.self_attn.o_proj.input",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.11.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.11.self_attn.o_proj.input",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_13_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.13.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.13.self_attn.o_proj.input",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.13.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.13.self_attn.o_proj.input",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_13_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.13.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.13.self_attn.o_proj.input",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.13.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.13.self_attn.o_proj.input",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_14_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.14.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.14.self_attn.o_proj.input",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.14.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.14.self_attn.o_proj.input",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_15_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.15.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.15.self_attn.o_proj.input",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.15.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.15.self_attn.o_proj.input",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_15_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.15.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.15.self_attn.o_proj.input",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.15.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.15.self_attn.o_proj.input",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_17_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.17.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.17.self_attn.o_proj.input",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.17.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.17.self_attn.o_proj.input",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_17_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.17.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.17.self_attn.o_proj.input",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.17.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.17.self_attn.o_proj.input",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_2_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.2.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.2.self_attn.o_proj.input",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 80,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.2.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.2.self_attn.o_proj.input",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 80,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_2_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.2.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.2.self_attn.o_proj.input",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 13,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.2.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.2.self_attn.o_proj.input",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 13,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_3_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.3.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.3.self_attn.o_proj.input",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 90,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.3.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.3.self_attn.o_proj.input",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 90,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_3_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.3.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.3.self_attn.o_proj.input",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 15,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.3.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.3.self_attn.o_proj.input",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 15,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_4_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.4.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.4.self_attn.o_proj.input",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 100,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.4.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.4.self_attn.o_proj.input",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 100,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_4_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.4.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.4.self_attn.o_proj.input",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 100,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.4.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.4.self_attn.o_proj.input",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 100,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_7_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.7.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.7.self_attn.o_proj.input",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.7.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.7.self_attn.o_proj.input",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
attn_out_all/layer_9_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.9.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.9.self_attn.o_proj.input",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.9.self_attn.o_proj.input",
3
  "hf_hook_point_out": "model.layers.9.self_attn.o_proj.input",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
crosscoder/layer_5_9_12_15_width_1m_l0_big/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
3
- "hf_hook_point_out": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
4
  "width": 1048576,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{5,9,12,15}.output",
3
+ "hf_hook_point_out": "model.layers.{5,9,12,15}.output",
4
  "width": 1048576,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_5_9_12_15_width_1m_l0_medium/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
3
- "hf_hook_point_out": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
4
  "width": 1048576,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{5,9,12,15}.output",
3
+ "hf_hook_point_out": "model.layers.{5,9,12,15}.output",
4
  "width": 1048576,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_5_9_12_15_width_262k_l0_medium/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
3
- "hf_hook_point_out": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{5,9,12,15}.output",
3
+ "hf_hook_point_out": "model.layers.{5,9,12,15}.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_5_9_12_15_width_524k_l0_big/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
3
- "hf_hook_point_out": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
4
  "width": 524288,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{5,9,12,15}.output",
3
+ "hf_hook_point_out": "model.layers.{5,9,12,15}.output",
4
  "width": 524288,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_5_9_12_15_width_524k_l0_medium/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
3
- "hf_hook_point_out": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
4
  "width": 524288,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{5,9,12,15}.output",
3
+ "hf_hook_point_out": "model.layers.{5,9,12,15}.output",
4
  "width": 524288,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_5_9_12_15_width_65k_l0_big/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
3
- "hf_hook_point_out": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{5,9,12,15}.output",
3
+ "hf_hook_point_out": "model.layers.{5,9,12,15}.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_5_9_12_15_width_65k_l0_medium/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
3
- "hf_hook_point_out": "model.layers.{5.0,9.5,12.200000000000001,15.799999999999999}.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{5,9,12,15}.output",
3
+ "hf_hook_point_out": "model.layers.{5,9,12,15}.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
mlp_out/layer_12_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_12_width_16k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_12_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_12_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_12_width_262k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_12_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_12_width_65k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_12_width_65k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_12_width_65k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_15_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_15_width_16k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_15_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_15_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_15_width_262k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_15_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_15_width_65k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_15_width_65k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_15_width_65k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.15.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.15.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_5_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.5.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.5.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 135,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.5.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.5.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 135,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_5_width_16k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.5.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.5.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 55,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.5.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.5.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 55,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_5_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.5.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.5.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 18,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.5.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.5.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 18,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }
mlp_out/layer_5_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.5.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.5.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 135,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.5.post_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.5.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-270m-it",
6
  "architecture": "jump_relu",
7
  "l0": 135,
8
+ "affine_connection": false,
9
+ "type": "sae"
10
  }