SAELens
CallumMcDougallGDM commited on
Commit
c0f4dfd
·
verified ·
1 Parent(s): 64604b7

Add files using upload-large-folder tool

Browse files
Files changed (42) hide show
  1. transcoder_all/layer_11_width_262k_l0_big/config.json +3 -2
  2. transcoder_all/layer_14_width_16k_l0_small_affine/config.json +3 -2
  3. transcoder_all/layer_14_width_262k_l0_small_affine/config.json +3 -2
  4. transcoder_all/layer_16_width_16k_l0_big/config.json +3 -2
  5. transcoder_all/layer_17_width_262k_l0_big_affine/config.json +3 -2
  6. transcoder_all/layer_19_width_262k_l0_small_affine/config.json +3 -2
  7. transcoder_all/layer_1_width_16k_l0_small/config.json +3 -2
  8. transcoder_all/layer_1_width_16k_l0_small_affine/config.json +3 -2
  9. transcoder_all/layer_20_width_16k_l0_big/config.json +3 -2
  10. transcoder_all/layer_20_width_16k_l0_big_affine/config.json +3 -2
  11. transcoder_all/layer_21_width_262k_l0_small_affine/config.json +3 -2
  12. transcoder_all/layer_26_width_262k_l0_small/config.json +3 -2
  13. transcoder_all/layer_27_width_16k_l0_big/config.json +3 -2
  14. transcoder_all/layer_28_width_16k_l0_big/config.json +3 -2
  15. transcoder_all/layer_28_width_16k_l0_big_affine/config.json +3 -2
  16. transcoder_all/layer_31_width_16k_l0_small/config.json +3 -2
  17. transcoder_all/layer_32_width_262k_l0_big/config.json +3 -2
  18. transcoder_all/layer_32_width_262k_l0_small/config.json +3 -2
  19. transcoder_all/layer_33_width_262k_l0_big/config.json +3 -2
  20. transcoder_all/layer_35_width_262k_l0_big/config.json +3 -2
  21. transcoder_all/layer_36_width_262k_l0_small/config.json +3 -2
  22. transcoder_all/layer_39_width_262k_l0_big_affine/config.json +3 -2
  23. transcoder_all/layer_3_width_16k_l0_big/config.json +3 -2
  24. transcoder_all/layer_41_width_262k_l0_small_affine/config.json +3 -2
  25. transcoder_all/layer_42_width_262k_l0_big/config.json +3 -2
  26. transcoder_all/layer_45_width_262k_l0_big_affine/config.json +3 -2
  27. transcoder_all/layer_45_width_262k_l0_small/config.json +3 -2
  28. transcoder_all/layer_46_width_262k_l0_big/config.json +3 -2
  29. transcoder_all/layer_46_width_262k_l0_small_affine/config.json +3 -2
  30. transcoder_all/layer_47_width_16k_l0_small_affine/config.json +3 -2
  31. transcoder_all/layer_47_width_262k_l0_small/config.json +3 -2
  32. transcoder_all/layer_4_width_16k_l0_big_affine/config.json +3 -2
  33. transcoder_all/layer_51_width_16k_l0_small/config.json +3 -2
  34. transcoder_all/layer_55_width_16k_l0_big_affine/config.json +3 -2
  35. transcoder_all/layer_56_width_16k_l0_big_affine/config.json +3 -2
  36. transcoder_all/layer_58_width_16k_l0_big_affine/config.json +3 -2
  37. transcoder_all/layer_58_width_262k_l0_big_affine/config.json +3 -2
  38. transcoder_all/layer_5_width_16k_l0_big/config.json +3 -2
  39. transcoder_all/layer_60_width_262k_l0_small_affine/config.json +3 -2
  40. transcoder_all/layer_6_width_262k_l0_small_affine/config.json +3 -2
  41. transcoder_all/layer_7_width_16k_l0_small_affine/config.json +3 -2
  42. transcoder_all/layer_8_width_16k_l0_small_affine/config.json +3 -2
transcoder_all/layer_11_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.11.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.11.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 91,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.11.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.11.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 91,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_14_width_16k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.14.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.14.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 16,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.14.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.14.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 16,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_14_width_262k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.14.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.14.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 16,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.14.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.14.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 16,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_16_width_16k_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": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 106,
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-pt",
6
  "architecture": "jump_relu",
7
  "l0": 106,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_17_width_262k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 109,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 109,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_19_width_262k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.19.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.19.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 19,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.19.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.19.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 19,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_1_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.1.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.1.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 10,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.1.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.1.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 10,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_1_width_16k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.1.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.1.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 10,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.1.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.1.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 10,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_20_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.20.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.20.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 118,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.20.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.20.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 118,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_20_width_16k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.20.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.20.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 118,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.20.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.20.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 118,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_21_width_262k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.21.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.21.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.21.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.21.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_26_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.26.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.26.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.26.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.26.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_27_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.27.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.27.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.27.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.27.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_28_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.28.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.28.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.28.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.28.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_28_width_16k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.28.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.28.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.28.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.28.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_31_width_16k_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": 16384,
5
- "model_name": "gemma-v3-27b-pt",
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": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_32_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.32.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.32.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.32.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.32.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_32_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.32.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.32.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.32.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.32.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_33_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.33.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.33.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.33.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.33.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_35_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.35.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.35.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.35.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.35.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_36_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.36.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.36.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.36.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.36.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_39_width_262k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.39.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.39.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.39.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.39.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_3_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.3.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.3.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 68,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.3.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.3.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 68,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_41_width_262k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.41.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.41.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.41.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.41.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_42_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.42.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.42.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.42.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.42.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_45_width_262k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.45.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.45.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.45.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.45.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_45_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.45.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.45.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.45.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.45.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_46_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.46.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.46.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.46.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.46.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_46_width_262k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.46.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.46.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.46.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.46.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_47_width_16k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.47.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.47.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.47.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.47.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_47_width_262k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.47.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.47.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.47.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.47.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_4_width_16k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.4.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.4.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 71,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.4.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.4.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 71,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_51_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.51.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.51.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.51.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.51.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_55_width_16k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.55.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.55.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.55.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.55.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_56_width_16k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.56.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.56.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.56.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.56.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_58_width_16k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.58.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.58.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.58.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.58.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_58_width_262k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.58.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.58.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.58.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.58.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_5_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.5.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.5.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 74,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.5.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.5.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 74,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_60_width_262k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.60.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.60.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.60.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.60.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_6_width_262k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.6.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.6.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 12,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.6.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.6.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 12,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_7_width_16k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.7.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.7.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 13,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.7.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.7.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 13,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_8_width_16k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.8.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.8.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 13,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.8.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.8.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-27b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 13,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }