Instructions to use google/gemma-scope-2-27b-pt with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- SAELens
How to use google/gemma-scope-2-27b-pt with SAELens:
# pip install sae-lens from sae_lens import SAE sae, cfg_dict, sparsity = SAE.from_pretrained( release = "RELEASE_ID", # e.g., "gpt2-small-res-jb". See other options in https://github.com/jbloomAus/SAELens/blob/main/sae_lens/pretrained_saes.yaml sae_id = "SAE_ID", # e.g., "blocks.8.hook_resid_pre". Won't always be a hook point ) - Notebooks
- Google Colab
- Kaggle
Add files using upload-large-folder tool
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- crosscoder/layer_16_31_40_53_width_262k_l0_big/config.json +10 -0
- crosscoder/layer_16_31_40_53_width_262k_l0_medium/config.json +10 -0
- crosscoder/layer_16_31_40_53_width_524k_l0_big/config.json +10 -0
- crosscoder/layer_16_31_40_53_width_524k_l0_medium/config.json +10 -0
- crosscoder/layer_16_31_40_53_width_65k_l0_big/config.json +10 -0
- crosscoder/layer_16_31_40_53_width_65k_l0_medium/config.json +10 -0
- mlp_out_all/layer_0_width_16k_l0_small/config.json +10 -0
- mlp_out_all/layer_0_width_262k_l0_small/config.json +10 -0
- mlp_out_all/layer_10_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_11_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_12_width_16k_l0_big/config.json +10 -0
- mlp_out_all/layer_12_width_16k_l0_small/config.json +10 -0
- mlp_out_all/layer_13_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_14_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_14_width_262k_l0_small/config.json +10 -0
- mlp_out_all/layer_17_width_262k_l0_small/config.json +10 -0
- mlp_out_all/layer_18_width_16k_l0_big/config.json +10 -0
- mlp_out_all/layer_18_width_16k_l0_small/config.json +10 -0
- mlp_out_all/layer_18_width_262k_l0_small/config.json +10 -0
- mlp_out_all/layer_1_width_16k_l0_big/config.json +10 -0
- mlp_out_all/layer_1_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_1_width_262k_l0_small/config.json +10 -0
- mlp_out_all/layer_24_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_29_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_29_width_262k_l0_small/config.json +10 -0
- mlp_out_all/layer_30_width_16k_l0_big/config.json +10 -0
- mlp_out_all/layer_30_width_262k_l0_small/config.json +10 -0
- mlp_out_all/layer_31_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_32_width_16k_l0_small/config.json +10 -0
- mlp_out_all/layer_34_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_35_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_37_width_262k_l0_small/config.json +10 -0
- mlp_out_all/layer_3_width_16k_l0_small/config.json +10 -0
- mlp_out_all/layer_40_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_41_width_16k_l0_big/config.json +10 -0
- mlp_out_all/layer_42_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_42_width_262k_l0_small/config.json +10 -0
- mlp_out_all/layer_44_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_4_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_50_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_51_width_16k_l0_small/config.json +10 -0
- mlp_out_all/layer_52_width_16k_l0_big/config.json +10 -0
- mlp_out_all/layer_54_width_16k_l0_small/config.json +10 -0
- mlp_out_all/layer_56_width_16k_l0_big/config.json +10 -0
- mlp_out_all/layer_58_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_59_width_16k_l0_small/config.json +10 -0
- mlp_out_all/layer_60_width_262k_l0_small/config.json +10 -0
- mlp_out_all/layer_61_width_16k_l0_small/config.json +10 -0
- mlp_out_all/layer_6_width_262k_l0_big/config.json +10 -0
- mlp_out_all/layer_6_width_262k_l0_small/config.json +10 -0
crosscoder/layer_16_31_40_53_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.{16,31,40,53}.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.{16,31,40,53}.output",
|
| 4 |
+
"width": 262144,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 150,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "crosscoder"
|
| 10 |
+
}
|
crosscoder/layer_16_31_40_53_width_262k_l0_medium/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.{16,31,40,53}.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.{16,31,40,53}.output",
|
| 4 |
+
"width": 262144,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 50,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "crosscoder"
|
| 10 |
+
}
|
crosscoder/layer_16_31_40_53_width_524k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.{16,31,40,53}.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.{16,31,40,53}.output",
|
| 4 |
+
"width": 524288,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 150,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "crosscoder"
|
| 10 |
+
}
|
crosscoder/layer_16_31_40_53_width_524k_l0_medium/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.{16,31,40,53}.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.{16,31,40,53}.output",
|
| 4 |
+
"width": 524288,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 50,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "crosscoder"
|
| 10 |
+
}
|
crosscoder/layer_16_31_40_53_width_65k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.{16,31,40,53}.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.{16,31,40,53}.output",
|
| 4 |
+
"width": 65536,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 150,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "crosscoder"
|
| 10 |
+
}
|
crosscoder/layer_16_31_40_53_width_65k_l0_medium/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.{16,31,40,53}.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.{16,31,40,53}.output",
|
| 4 |
+
"width": 65536,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 50,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "crosscoder"
|
| 10 |
+
}
|
mlp_out_all/layer_0_width_16k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.0.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.0.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_0_width_262k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.0.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.0.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 262144,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 10,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_10_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.10.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.10.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 262144,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 89,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_11_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.11.post_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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_12_width_16k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 16384,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 94,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_12_width_16k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 16384,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 15,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_13_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.13.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.13.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 262144,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 97,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_14_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.14.post_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": 100,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_14_width_262k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.14.post_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": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_17_width_262k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.17.post_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": 18,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_18_width_16k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.18.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.18.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 16384,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 112,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_18_width_16k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.18.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.18.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 16384,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 18,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_18_width_262k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.18.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.18.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 262144,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 18,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_1_width_16k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.1.post_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": 62,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_1_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.1.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.1.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 262144,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 62,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_1_width_262k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.1.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.1.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 262144,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 10,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_24_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.24.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.24.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_29_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.29.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.29.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_29_width_262k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.29.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.29.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_30_width_16k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.30.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.30.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_30_width_262k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.30.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.30.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_31_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.31.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 262144,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 120,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_32_width_16k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.32.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.32.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_34_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.34.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.34.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_35_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.35.post_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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_37_width_262k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.37.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.37.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_3_width_16k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.3.post_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": 11,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_40_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.40.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.40.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 262144,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 120,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_41_width_16k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.41.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.41.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_42_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.42.post_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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_42_width_262k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.42.post_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": 20,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_44_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.44.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.44.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_4_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.4.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.4.post_feedforward_layernorm.output",
|
| 4 |
+
"width": 262144,
|
| 5 |
+
"model_name": "google/gemma-3-27b-pt",
|
| 6 |
+
"architecture": "jump_relu",
|
| 7 |
+
"l0": 71,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_50_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.50.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.50.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_51_width_16k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.51.post_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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_52_width_16k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.52.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.52.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_54_width_16k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.54.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.54.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_56_width_16k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.56.post_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": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_58_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.58.post_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": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_59_width_16k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.59.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.59.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_60_width_262k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.60.post_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": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_61_width_16k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.61.post_feedforward_layernorm.output",
|
| 3 |
+
"hf_hook_point_out": "model.layers.61.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": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_6_width_262k_l0_big/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.6.post_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": 77,
|
| 8 |
+
"affine_connection": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|
mlp_out_all/layer_6_width_262k_l0_small/config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"hf_hook_point_in": "model.layers.6.post_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": false,
|
| 9 |
+
"type": "sae"
|
| 10 |
+
}
|