SAELens
CallumMcDougallGDM commited on
Commit
a9f720b
·
verified ·
1 Parent(s): 49f7467

Add files using upload-large-folder tool

Browse files
Files changed (50) hide show
  1. transcoder/layer_16_width_16k_l0_big_affine/config.json +3 -2
  2. transcoder/layer_16_width_16k_l0_medium/config.json +3 -2
  3. transcoder/layer_16_width_16k_l0_medium_affine/config.json +3 -2
  4. transcoder/layer_16_width_16k_l0_small/config.json +3 -2
  5. transcoder/layer_16_width_16k_l0_small_affine/config.json +3 -2
  6. transcoder/layer_16_width_262k_l0_big/config.json +3 -2
  7. transcoder/layer_16_width_262k_l0_big_affine/config.json +3 -2
  8. transcoder/layer_16_width_262k_l0_medium/config.json +3 -2
  9. transcoder/layer_16_width_262k_l0_small/config.json +3 -2
  10. transcoder/layer_16_width_262k_l0_small_affine/config.json +3 -2
  11. transcoder/layer_16_width_65k_l0_big/config.json +3 -2
  12. transcoder/layer_16_width_65k_l0_big_affine/config.json +3 -2
  13. transcoder/layer_16_width_65k_l0_medium/config.json +3 -2
  14. transcoder/layer_16_width_65k_l0_medium_affine/config.json +3 -2
  15. transcoder/layer_16_width_65k_l0_small_affine/config.json +3 -2
  16. transcoder/layer_31_width_16k_l0_big/config.json +3 -2
  17. transcoder/layer_31_width_16k_l0_big_affine/config.json +3 -2
  18. transcoder/layer_31_width_16k_l0_medium/config.json +3 -2
  19. transcoder/layer_31_width_16k_l0_medium_affine/config.json +3 -2
  20. transcoder/layer_31_width_262k_l0_medium/config.json +3 -2
  21. transcoder/layer_31_width_65k_l0_big_affine/config.json +3 -2
  22. transcoder/layer_31_width_65k_l0_medium/config.json +3 -2
  23. transcoder/layer_31_width_65k_l0_medium_affine/config.json +3 -2
  24. transcoder/layer_31_width_65k_l0_small/config.json +3 -2
  25. transcoder/layer_31_width_65k_l0_small_affine/config.json +3 -2
  26. transcoder/layer_40_width_16k_l0_big/config.json +3 -2
  27. transcoder/layer_40_width_16k_l0_medium_affine/config.json +3 -2
  28. transcoder/layer_40_width_16k_l0_small/config.json +3 -2
  29. transcoder/layer_40_width_16k_l0_small_affine/config.json +3 -2
  30. transcoder/layer_40_width_262k_l0_big/config.json +3 -2
  31. transcoder/layer_40_width_262k_l0_big_affine/config.json +3 -2
  32. transcoder/layer_40_width_262k_l0_small/config.json +3 -2
  33. transcoder/layer_40_width_262k_l0_small_affine/config.json +3 -2
  34. transcoder/layer_40_width_65k_l0_big/config.json +3 -2
  35. transcoder/layer_40_width_65k_l0_big_affine/config.json +3 -2
  36. transcoder/layer_40_width_65k_l0_medium/config.json +3 -2
  37. transcoder/layer_40_width_65k_l0_medium_affine/config.json +3 -2
  38. transcoder/layer_40_width_65k_l0_small/config.json +3 -2
  39. transcoder/layer_40_width_65k_l0_small_affine/config.json +3 -2
  40. transcoder/layer_53_width_16k_l0_big_affine/config.json +3 -2
  41. transcoder/layer_53_width_16k_l0_medium_affine/config.json +3 -2
  42. transcoder/layer_53_width_16k_l0_small/config.json +3 -2
  43. transcoder/layer_53_width_262k_l0_big_affine/config.json +3 -2
  44. transcoder/layer_53_width_262k_l0_medium/config.json +3 -2
  45. transcoder/layer_53_width_262k_l0_medium_affine/config.json +3 -2
  46. transcoder/layer_53_width_262k_l0_small/config.json +3 -2
  47. transcoder/layer_53_width_262k_l0_small_affine/config.json +3 -2
  48. transcoder/layer_53_width_65k_l0_big_affine/config.json +3 -2
  49. transcoder/layer_53_width_65k_l0_medium/config.json +3 -2
  50. transcoder/layer_53_width_65k_l0_small_affine/config.json +3 -2
transcoder/layer_16_width_16k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 129,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 129,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_16k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 53,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 53,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_16k_l0_medium_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 53,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 53,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_16k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 129,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 129,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_262k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 129,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 129,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_262k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 53,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 53,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_262k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_65k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 129,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 129,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_65k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 129,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 129,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_65k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 53,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 53,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_65k_l0_medium_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 53,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 53,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_16_width_65k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.16.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_31_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_31_width_16k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_31_width_16k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_31_width_16k_l0_medium_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_31_width_262k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_31_width_65k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_31_width_65k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_31_width_65k_l0_medium_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_31_width_65k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_31_width_65k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.31.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_16k_l0_medium_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_16k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_262k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_262k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_65k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_65k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_65k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_65k_l0_medium_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_65k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_40_width_65k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.40.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_53_width_16k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_53_width_16k_l0_medium_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_53_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_53_width_262k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_53_width_262k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_53_width_262k_l0_medium_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_53_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_53_width_262k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_53_width_65k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_53_width_65k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_53_width_65k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.53.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.53.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-27b-it",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }