Hugging Face's logo

google
/
gemma-scope-2-12b-pt

SAELens
Model card Files
xet
 Community
CallumMcDougallGDM commited on
Commit
0eca19b
·
verified ·
1 Parent(s): a3c5dbc

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. crosscoder/layer_12_24_31_41_width_1m_l0_big/config.json +5 -4
  2. crosscoder/layer_12_24_31_41_width_1m_l0_medium/config.json +5 -4
  3. crosscoder/layer_12_24_31_41_width_262k_l0_big/config.json +5 -4
  4. crosscoder/layer_12_24_31_41_width_262k_l0_medium/config.json +5 -4
  5. crosscoder/layer_12_24_31_41_width_524k_l0_big/config.json +5 -4
  6. crosscoder/layer_12_24_31_41_width_524k_l0_medium/config.json +5 -4
  7. crosscoder/layer_12_24_31_41_width_65k_l0_big/config.json +5 -4
  8. crosscoder/layer_12_24_31_41_width_65k_l0_medium/config.json +5 -4
  9. transcoder/layer_12_width_16k_l0_big/config.json +3 -2
  10. transcoder/layer_12_width_16k_l0_big_affine/config.json +3 -2
  11. transcoder/layer_12_width_16k_l0_small/config.json +3 -2
  12. transcoder/layer_12_width_262k_l0_medium/config.json +3 -2
  13. transcoder/layer_12_width_65k_l0_big/config.json +3 -2
  14. transcoder/layer_24_width_16k_l0_medium/config.json +3 -2
  15. transcoder/layer_24_width_262k_l0_big/config.json +3 -2
  16. transcoder/layer_24_width_262k_l0_medium/config.json +3 -2
  17. transcoder/layer_24_width_262k_l0_medium_affine/config.json +3 -2
  18. transcoder/layer_24_width_65k_l0_medium_affine/config.json +3 -2
  19. transcoder/layer_31_width_16k_l0_big/config.json +3 -2
  20. transcoder/layer_31_width_16k_l0_small/config.json +3 -2
  21. transcoder/layer_31_width_262k_l0_big_affine/config.json +3 -2
  22. transcoder/layer_31_width_65k_l0_big/config.json +3 -2
  23. transcoder/layer_31_width_65k_l0_medium/config.json +3 -2
  24. transcoder/layer_31_width_65k_l0_medium_affine/config.json +3 -2
  25. transcoder/layer_41_width_16k_l0_medium_affine/config.json +3 -2
  26. transcoder/layer_41_width_16k_l0_small_affine/config.json +3 -2
  27. transcoder/layer_41_width_262k_l0_big/config.json +3 -2
  28. transcoder/layer_41_width_262k_l0_big_affine/config.json +3 -2
  29. transcoder/layer_41_width_262k_l0_medium/config.json +3 -2
  30. transcoder/layer_41_width_65k_l0_small/config.json +3 -2
  31. transcoder_all/layer_0_width_16k_l0_small/config.json +3 -2
  32. transcoder_all/layer_0_width_262k_l0_big/config.json +3 -2
  33. transcoder_all/layer_0_width_262k_l0_big_affine/config.json +3 -2
  34. transcoder_all/layer_12_width_262k_l0_small_affine/config.json +3 -2
  35. transcoder_all/layer_13_width_16k_l0_small/config.json +3 -2
  36. transcoder_all/layer_13_width_262k_l0_big_affine/config.json +3 -2
  37. transcoder_all/layer_16_width_16k_l0_small/config.json +3 -2
  38. transcoder_all/layer_17_width_16k_l0_small_affine/config.json +3 -2
  39. transcoder_all/layer_19_width_16k_l0_small/config.json +3 -2
  40. transcoder_all/layer_20_width_262k_l0_big_affine/config.json +3 -2
  41. transcoder_all/layer_26_width_262k_l0_big/config.json +3 -2
  42. transcoder_all/layer_29_width_16k_l0_big/config.json +3 -2
  43. transcoder_all/layer_29_width_262k_l0_big/config.json +3 -2
  44. transcoder_all/layer_2_width_16k_l0_big_affine/config.json +3 -2
  45. transcoder_all/layer_2_width_16k_l0_small/config.json +3 -2
  46. transcoder_all/layer_31_width_16k_l0_small_affine/config.json +3 -2
  47. transcoder_all/layer_31_width_262k_l0_small_affine/config.json +3 -2
  48. transcoder_all/layer_32_width_16k_l0_big_affine/config.json +3 -2
  49. transcoder_all/layer_33_width_16k_l0_small/config.json +3 -2
  50. transcoder_all/layer_33_width_16k_l0_small_affine/config.json +3 -2
crosscoder/layer_12_24_31_41_width_1m_l0_big/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
3
- "hf_hook_point_out": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
4
  "width": 1048576,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{12,24,31,41}.output",
3
+ "hf_hook_point_out": "model.layers.{12,24,31,41}.output",
4
  "width": 1048576,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_12_24_31_41_width_1m_l0_medium/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
3
- "hf_hook_point_out": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
4
  "width": 1048576,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{12,24,31,41}.output",
3
+ "hf_hook_point_out": "model.layers.{12,24,31,41}.output",
4
  "width": 1048576,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_12_24_31_41_width_262k_l0_big/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
3
- "hf_hook_point_out": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{12,24,31,41}.output",
3
+ "hf_hook_point_out": "model.layers.{12,24,31,41}.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_12_24_31_41_width_262k_l0_medium/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
3
- "hf_hook_point_out": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{12,24,31,41}.output",
3
+ "hf_hook_point_out": "model.layers.{12,24,31,41}.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_12_24_31_41_width_524k_l0_big/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
3
- "hf_hook_point_out": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
4
  "width": 524288,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{12,24,31,41}.output",
3
+ "hf_hook_point_out": "model.layers.{12,24,31,41}.output",
4
  "width": 524288,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_12_24_31_41_width_524k_l0_medium/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
3
- "hf_hook_point_out": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
4
  "width": 524288,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{12,24,31,41}.output",
3
+ "hf_hook_point_out": "model.layers.{12,24,31,41}.output",
4
  "width": 524288,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_12_24_31_41_width_65k_l0_big/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
3
- "hf_hook_point_out": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{12,24,31,41}.output",
3
+ "hf_hook_point_out": "model.layers.{12,24,31,41}.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
crosscoder/layer_12_24_31_41_width_65k_l0_medium/config.json CHANGED
@@ -1,9 +1,10 @@
1
  {
2
- "hf_hook_point_in": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
3
- "hf_hook_point_out": "model.layers.{12.5,24.5,31.700000000000003,41.3}.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
- "affine_connection": false
 
9
  }
 
1
  {
2
+ "hf_hook_point_in": "model.layers.{12,24,31,41}.output",
3
+ "hf_hook_point_out": "model.layers.{12,24,31,41}.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 50,
8
+ "affine_connection": false,
9
+ "type": "crosscoder"
10
  }
transcoder/layer_12_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 127,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 127,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_12_width_16k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 127,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 127,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_12_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_12_width_262k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 52,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 52,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_12_width_65k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 127,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 127,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_24_width_16k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.24.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.24.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.24.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.24.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_24_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.24.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.24.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.24.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.24.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_24_width_262k_l0_medium/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.24.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.24.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.24.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.24.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_24_width_262k_l0_medium_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.24.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.24.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.24.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.24.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_24_width_65k_l0_medium_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.24.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.24.post_feedforward_layernorm.output",
4
  "width": 65536,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.24.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.24.post_feedforward_layernorm.output",
4
  "width": 65536,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
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-v3-12b-pt",
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-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/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-v3-12b-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-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_31_width_262k_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": 262144,
5
- "model_name": "gemma-v3-12b-pt",
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": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_31_width_65k_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": 65536,
5
- "model_name": "gemma-v3-12b-pt",
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": 65536,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
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-v3-12b-pt",
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-3-12b-pt",
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-v3-12b-pt",
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-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_41_width_16k_l0_medium_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": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
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": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_41_width_16k_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": 16384,
5
- "model_name": "gemma-v3-12b-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": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_41_width_262k_l0_big/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-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
- "affine_connection": false
 
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-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_41_width_262k_l0_big_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-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
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-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 150,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder/layer_41_width_262k_l0_medium/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-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
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-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder/layer_41_width_65k_l0_small/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": 65536,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
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": 65536,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_0_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.0.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.0.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 10,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.0.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.0.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 10,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_0_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.0.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.0.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.0.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.0.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_0_width_262k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.0.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.0.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.0.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.0.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 60,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_12_width_262k_l0_small_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.12.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.12.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 17,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_13_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.13.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.13.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 18,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.13.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.13.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 18,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_13_width_262k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.13.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.13.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 108,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.13.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.13.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 108,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/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-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
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-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_17_width_16k_l0_small_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": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
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": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_19_width_16k_l0_small/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": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": false
 
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": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_20_width_262k_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": 262144,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
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": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_26_width_262k_l0_big/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-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
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-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_29_width_16k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_29_width_262k_l0_big/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output",
4
  "width": 262144,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output",
4
  "width": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_2_width_16k_l0_big_affine/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.2.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.2.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 67,
8
- "affine_connection": true
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.2.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.2.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 67,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_2_width_16k_l0_small/config.json CHANGED
@@ -2,8 +2,9 @@
2
  "hf_hook_point_in": "model.layers.2.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.2.post_feedforward_layernorm.output",
4
  "width": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 11,
8
- "affine_connection": false
 
9
  }
 
2
  "hf_hook_point_in": "model.layers.2.pre_feedforward_layernorm.output",
3
  "hf_hook_point_out": "model.layers.2.post_feedforward_layernorm.output",
4
  "width": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 11,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_31_width_16k_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": 16384,
5
- "model_name": "gemma-v3-12b-pt",
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": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_31_width_262k_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": 262144,
5
- "model_name": "gemma-v3-12b-pt",
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": 262144,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_32_width_16k_l0_big_affine/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": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
- "affine_connection": true
 
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": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 120,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_33_width_16k_l0_small/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": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
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": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": false,
9
+ "type": "transcoder"
10
  }
transcoder_all/layer_33_width_16k_l0_small_affine/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": 16384,
5
- "model_name": "gemma-v3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
- "affine_connection": true
 
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": 16384,
5
+ "model_name": "google/gemma-3-12b-pt",
6
  "architecture": "jump_relu",
7
  "l0": 20,
8
+ "affine_connection": true,
9
+ "type": "transcoder"
10
  }