Hugging Face's logo

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

SAELens
Model card Files
xet
Community
CallumMcDougallGDM commited on
Commit
da34083
·
verified ·
1 Parent(s): 9663d08

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. attn_out/layer_12_width_16k_l0_big/config.json +9 -0
  2. attn_out/layer_12_width_16k_l0_medium/config.json +9 -0
  3. attn_out/layer_12_width_16k_l0_small/config.json +9 -0
  4. attn_out/layer_12_width_262k_l0_big/config.json +9 -0
  5. attn_out/layer_12_width_262k_l0_medium/config.json +9 -0
  6. attn_out/layer_12_width_262k_l0_small/config.json +9 -0
  7. attn_out/layer_12_width_65k_l0_big/config.json +9 -0
  8. attn_out/layer_12_width_65k_l0_medium/config.json +9 -0
  9. attn_out/layer_12_width_65k_l0_small/config.json +9 -0
  10. attn_out/layer_24_width_16k_l0_big/config.json +9 -0
  11. attn_out/layer_24_width_16k_l0_medium/config.json +9 -0
  12. attn_out/layer_24_width_16k_l0_small/config.json +9 -0
  13. attn_out/layer_24_width_262k_l0_big/config.json +9 -0
  14. attn_out/layer_24_width_262k_l0_medium/config.json +9 -0
  15. attn_out/layer_24_width_262k_l0_small/config.json +9 -0
  16. attn_out/layer_24_width_65k_l0_big/config.json +9 -0
  17. attn_out/layer_24_width_65k_l0_medium/config.json +9 -0
  18. attn_out/layer_24_width_65k_l0_small/config.json +9 -0
  19. attn_out/layer_31_width_16k_l0_big/config.json +9 -0
  20. attn_out/layer_31_width_16k_l0_medium/config.json +9 -0
  21. attn_out/layer_31_width_16k_l0_small/config.json +9 -0
  22. attn_out/layer_31_width_262k_l0_big/config.json +9 -0
  23. attn_out/layer_31_width_262k_l0_medium/config.json +9 -0
  24. attn_out/layer_31_width_262k_l0_small/config.json +9 -0
  25. attn_out/layer_31_width_65k_l0_big/config.json +9 -0
  26. attn_out/layer_31_width_65k_l0_medium/config.json +9 -0
  27. attn_out/layer_31_width_65k_l0_small/config.json +9 -0
  28. attn_out/layer_41_width_16k_l0_big/config.json +9 -0
  29. attn_out/layer_41_width_16k_l0_medium/config.json +9 -0
  30. attn_out/layer_41_width_16k_l0_small/config.json +9 -0
  31. attn_out/layer_41_width_262k_l0_big/config.json +9 -0
  32. attn_out/layer_41_width_262k_l0_medium/config.json +9 -0
  33. attn_out/layer_41_width_262k_l0_small/config.json +9 -0
  34. attn_out/layer_41_width_65k_l0_big/config.json +9 -0
  35. attn_out/layer_41_width_65k_l0_medium/config.json +9 -0
  36. attn_out/layer_41_width_65k_l0_small/config.json +9 -0
  37. resid_post/layer_12_width_16k_l0_big/config.json +9 -0
  38. resid_post/layer_12_width_16k_l0_medium/config.json +9 -0
  39. resid_post/layer_12_width_16k_l0_small/config.json +9 -0
  40. resid_post/layer_12_width_1m_l0_big/config.json +9 -0
  41. resid_post/layer_12_width_1m_l0_medium/config.json +9 -0
  42. resid_post/layer_12_width_1m_l0_small/config.json +9 -0
  43. resid_post/layer_12_width_262k_l0_big/config.json +9 -0
  44. resid_post/layer_12_width_262k_l0_medium/config.json +9 -0
  45. resid_post/layer_12_width_262k_l0_medium_seed_1/config.json +9 -0
  46. resid_post/layer_12_width_262k_l0_small/config.json +9 -0
  47. resid_post/layer_12_width_65k_l0_medium/config.json +9 -0
  48. resid_post/layer_12_width_65k_l0_small/config.json +9 -0
  49. resid_post/layer_24_width_16k_l0_big/config.json +9 -0
  50. resid_post/layer_24_width_1m_l0_big/config.json +9 -0
attn_out/layer_12_width_16k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.12.self_attn.o_proj.input",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 127,
8
+ "affine_connection": false
9
+ }
attn_out/layer_12_width_16k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.12.self_attn.o_proj.input",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 52,
8
+ "affine_connection": false
9
+ }
attn_out/layer_12_width_16k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.12.self_attn.o_proj.input",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 17,
8
+ "affine_connection": false
9
+ }
attn_out/layer_12_width_262k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.12.self_attn.o_proj.input",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 127,
8
+ "affine_connection": false
9
+ }
attn_out/layer_12_width_262k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.12.self_attn.o_proj.input",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 52,
8
+ "affine_connection": false
9
+ }
attn_out/layer_12_width_262k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.12.self_attn.o_proj.input",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 17,
8
+ "affine_connection": false
9
+ }
attn_out/layer_12_width_65k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.12.self_attn.o_proj.input",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 127,
8
+ "affine_connection": false
9
+ }
attn_out/layer_12_width_65k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.12.self_attn.o_proj.input",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 52,
8
+ "affine_connection": false
9
+ }
attn_out/layer_12_width_65k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.12.self_attn.o_proj.input",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 17,
8
+ "affine_connection": false
9
+ }
attn_out/layer_24_width_16k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.24.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.24.self_attn.o_proj.input",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 150,
8
+ "affine_connection": false
9
+ }
attn_out/layer_24_width_16k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.24.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.24.self_attn.o_proj.input",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 60,
8
+ "affine_connection": false
9
+ }
attn_out/layer_24_width_16k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.24.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.24.self_attn.o_proj.input",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 20,
8
+ "affine_connection": false
9
+ }
attn_out/layer_24_width_262k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.24.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.24.self_attn.o_proj.input",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 150,
8
+ "affine_connection": false
9
+ }
attn_out/layer_24_width_262k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.24.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.24.self_attn.o_proj.input",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 60,
8
+ "affine_connection": false
9
+ }
attn_out/layer_24_width_262k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.24.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.24.self_attn.o_proj.input",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 20,
8
+ "affine_connection": false
9
+ }
attn_out/layer_24_width_65k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.24.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.24.self_attn.o_proj.input",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 150,
8
+ "affine_connection": false
9
+ }
attn_out/layer_24_width_65k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.24.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.24.self_attn.o_proj.input",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 60,
8
+ "affine_connection": false
9
+ }
attn_out/layer_24_width_65k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.24.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.24.self_attn.o_proj.input",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 20,
8
+ "affine_connection": false
9
+ }
attn_out/layer_31_width_16k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.31.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.31.self_attn.o_proj.input",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 150,
8
+ "affine_connection": false
9
+ }
attn_out/layer_31_width_16k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.31.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.31.self_attn.o_proj.input",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 60,
8
+ "affine_connection": false
9
+ }
attn_out/layer_31_width_16k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.31.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.31.self_attn.o_proj.input",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 20,
8
+ "affine_connection": false
9
+ }
attn_out/layer_31_width_262k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.31.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.31.self_attn.o_proj.input",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 150,
8
+ "affine_connection": false
9
+ }
attn_out/layer_31_width_262k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.31.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.31.self_attn.o_proj.input",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 60,
8
+ "affine_connection": false
9
+ }
attn_out/layer_31_width_262k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.31.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.31.self_attn.o_proj.input",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 20,
8
+ "affine_connection": false
9
+ }
attn_out/layer_31_width_65k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.31.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.31.self_attn.o_proj.input",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 150,
8
+ "affine_connection": false
9
+ }
attn_out/layer_31_width_65k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.31.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.31.self_attn.o_proj.input",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 60,
8
+ "affine_connection": false
9
+ }
attn_out/layer_31_width_65k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.31.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.31.self_attn.o_proj.input",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 20,
8
+ "affine_connection": false
9
+ }
attn_out/layer_41_width_16k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.41.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.41.self_attn.o_proj.input",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 150,
8
+ "affine_connection": false
9
+ }
attn_out/layer_41_width_16k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.41.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.41.self_attn.o_proj.input",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 60,
8
+ "affine_connection": false
9
+ }
attn_out/layer_41_width_16k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.41.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.41.self_attn.o_proj.input",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 20,
8
+ "affine_connection": false
9
+ }
attn_out/layer_41_width_262k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.41.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.41.self_attn.o_proj.input",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 150,
8
+ "affine_connection": false
9
+ }
attn_out/layer_41_width_262k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.41.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.41.self_attn.o_proj.input",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 60,
8
+ "affine_connection": false
9
+ }
attn_out/layer_41_width_262k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.41.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.41.self_attn.o_proj.input",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 20,
8
+ "affine_connection": false
9
+ }
attn_out/layer_41_width_65k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.41.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.41.self_attn.o_proj.input",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 150,
8
+ "affine_connection": false
9
+ }
attn_out/layer_41_width_65k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.41.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.41.self_attn.o_proj.input",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 60,
8
+ "affine_connection": false
9
+ }
attn_out/layer_41_width_65k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.41.self_attn.o_proj.input",
3
+ "hf_hook_point_out": "model.layers.41.self_attn.o_proj.input",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 20,
8
+ "affine_connection": false
9
+ }
resid_post/layer_12_width_16k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.output",
3
+ "hf_hook_point_out": "model.layers.12.output",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 127,
8
+ "affine_connection": false
9
+ }
resid_post/layer_12_width_16k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.output",
3
+ "hf_hook_point_out": "model.layers.12.output",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 52,
8
+ "affine_connection": false
9
+ }
resid_post/layer_12_width_16k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.output",
3
+ "hf_hook_point_out": "model.layers.12.output",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 17,
8
+ "affine_connection": false
9
+ }
resid_post/layer_12_width_1m_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.output",
3
+ "hf_hook_point_out": "model.layers.12.output",
4
+ "width": 1048576,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 127,
8
+ "affine_connection": false
9
+ }
resid_post/layer_12_width_1m_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.output",
3
+ "hf_hook_point_out": "model.layers.12.output",
4
+ "width": 1048576,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 52,
8
+ "affine_connection": false
9
+ }
resid_post/layer_12_width_1m_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.output",
3
+ "hf_hook_point_out": "model.layers.12.output",
4
+ "width": 1048576,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 17,
8
+ "affine_connection": false
9
+ }
resid_post/layer_12_width_262k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.output",
3
+ "hf_hook_point_out": "model.layers.12.output",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 127,
8
+ "affine_connection": false
9
+ }
resid_post/layer_12_width_262k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.output",
3
+ "hf_hook_point_out": "model.layers.12.output",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 52,
8
+ "affine_connection": false
9
+ }
resid_post/layer_12_width_262k_l0_medium_seed_1/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.output",
3
+ "hf_hook_point_out": "model.layers.12.output",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 52,
8
+ "affine_connection": false
9
+ }
resid_post/layer_12_width_262k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.output",
3
+ "hf_hook_point_out": "model.layers.12.output",
4
+ "width": 262144,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 17,
8
+ "affine_connection": false
9
+ }
resid_post/layer_12_width_65k_l0_medium/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.output",
3
+ "hf_hook_point_out": "model.layers.12.output",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 52,
8
+ "affine_connection": false
9
+ }
resid_post/layer_12_width_65k_l0_small/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.12.output",
3
+ "hf_hook_point_out": "model.layers.12.output",
4
+ "width": 65536,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 17,
8
+ "affine_connection": false
9
+ }
resid_post/layer_24_width_16k_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.24.output",
3
+ "hf_hook_point_out": "model.layers.24.output",
4
+ "width": 16384,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 150,
8
+ "affine_connection": false
9
+ }
resid_post/layer_24_width_1m_l0_big/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "hf_hook_point_in": "model.layers.24.output",
3
+ "hf_hook_point_out": "model.layers.24.output",
4
+ "width": 1048576,
5
+ "model_name": "gemma-v3-12b-pt",
6
+ "architecture": "jump_relu",
7
+ "l0": 150,
8
+ "affine_connection": false
9
+ }