diff --git a/mlp_out_all/layer_10_width_262k_l0_small/config.json b/mlp_out_all/layer_10_width_262k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..a0997e395b3018daaeb60a76d937cee8d86a6efb --- /dev/null +++ b/mlp_out_all/layer_10_width_262k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.10.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.10.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 18, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_12_width_16k_l0_big/config.json b/mlp_out_all/layer_12_width_16k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..c56a4263d16570f9a993246f4df1946a0f162841 --- /dev/null +++ b/mlp_out_all/layer_12_width_16k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.12.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.12.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 120, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_13_width_16k_l0_big/config.json b/mlp_out_all/layer_13_width_16k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..72283685b207a779000c1186238f9268d4526b8c --- /dev/null +++ b/mlp_out_all/layer_13_width_16k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.13.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.13.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 120, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_14_width_262k_l0_small/config.json b/mlp_out_all/layer_14_width_262k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..4e97dbae14c06d4cda746e5ea25c148acda78720 --- /dev/null +++ b/mlp_out_all/layer_14_width_262k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.14.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.14.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_16_width_16k_l0_big/config.json b/mlp_out_all/layer_16_width_16k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..5c96303f5e282f6d1cd47e669bdd20b17f1a99da --- /dev/null +++ b/mlp_out_all/layer_16_width_16k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.16.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 120, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_16_width_262k_l0_small/config.json b/mlp_out_all/layer_16_width_262k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..2ec1d63988388b594a99a44939aa7e122391a397 --- /dev/null +++ b/mlp_out_all/layer_16_width_262k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.16.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.16.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_18_width_262k_l0_small/config.json b/mlp_out_all/layer_18_width_262k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..d1657a060c4a024484140ab76f16caffc8a0be23 --- /dev/null +++ b/mlp_out_all/layer_18_width_262k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.18.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.18.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_1_width_16k_l0_small/config.json b/mlp_out_all/layer_1_width_16k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..7fa9698f1d54d97f4ef222b805c62ad662250d9b --- /dev/null +++ b/mlp_out_all/layer_1_width_16k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.1.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.1.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 10, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_1_width_262k_l0_big/config.json b/mlp_out_all/layer_1_width_262k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..2458f8b453416ab5a4ec93755dfbb4e5b396ec41 --- /dev/null +++ b/mlp_out_all/layer_1_width_262k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.1.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.1.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 65, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_25_width_262k_l0_big/config.json b/mlp_out_all/layer_25_width_262k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..1f898a3d5c9781eb0a11bb36a987b01babe6b562 --- /dev/null +++ b/mlp_out_all/layer_25_width_262k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.25.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.25.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 120, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_27_width_262k_l0_big/config.json b/mlp_out_all/layer_27_width_262k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..1aab46b89593b41d810d4dd561ddec2956d98055 --- /dev/null +++ b/mlp_out_all/layer_27_width_262k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.27.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.27.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 120, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_31_width_16k_l0_big/config.json b/mlp_out_all/layer_31_width_16k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..54fd87c62064860d04451264329ed0608c40e09b --- /dev/null +++ b/mlp_out_all/layer_31_width_16k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.31.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 120, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_31_width_16k_l0_small/config.json b/mlp_out_all/layer_31_width_16k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..5ccb6b3ab9949e59dc7fdb14cc49434b5c6434d9 --- /dev/null +++ b/mlp_out_all/layer_31_width_16k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.31.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.31.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_3_width_16k_l0_small/config.json b/mlp_out_all/layer_3_width_16k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..990cde0aae212297e3d1e54d6591c4077a455d1f --- /dev/null +++ b/mlp_out_all/layer_3_width_16k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.3.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.3.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 12, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_5_width_16k_l0_small/config.json b/mlp_out_all/layer_5_width_16k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..3cf85e707b945e0d1fb618bc1ffaa6d241304956 --- /dev/null +++ b/mlp_out_all/layer_5_width_16k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.5.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.5.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 14, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/mlp_out_all/layer_6_width_262k_l0_big/config.json b/mlp_out_all/layer_6_width_262k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..caa111a056f6ebac891e34cde13a820afd1f70b7 --- /dev/null +++ b/mlp_out_all/layer_6_width_262k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.6.post_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.6.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 91, + "affine_connection": false, + "type": "sae" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_16k_l0_big_affine/config.json b/transcoder/layer_17_width_16k_l0_big_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..c7b77a94b7dd5a252b2ed8457b807334bc8d001a --- /dev/null +++ b/transcoder/layer_17_width_16k_l0_big_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 150, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_16k_l0_medium/config.json b/transcoder/layer_17_width_16k_l0_medium/config.json new file mode 100644 index 0000000000000000000000000000000000000000..7c07bf0823a3690ca8aef0cbdb28a84dcee7f761 --- /dev/null +++ b/transcoder/layer_17_width_16k_l0_medium/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_16k_l0_medium_affine/config.json b/transcoder/layer_17_width_16k_l0_medium_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..46fa5a83d637d6ec9442c82774a14341207af761 --- /dev/null +++ b/transcoder/layer_17_width_16k_l0_medium_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_16k_l0_small/config.json b/transcoder/layer_17_width_16k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..07caaf9ec1dcaa8e7239e77922a18fc84bbf4b8a --- /dev/null +++ b/transcoder/layer_17_width_16k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_262k_l0_big/config.json b/transcoder/layer_17_width_262k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..ae39ffd8883de4d4e39e1a9819e40b9f906af2ef --- /dev/null +++ b/transcoder/layer_17_width_262k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 150, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_262k_l0_big_affine/config.json b/transcoder/layer_17_width_262k_l0_big_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..5ca0c690a2ed53d3d651874127083d7d7166c880 --- /dev/null +++ b/transcoder/layer_17_width_262k_l0_big_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 150, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_262k_l0_medium/config.json b/transcoder/layer_17_width_262k_l0_medium/config.json new file mode 100644 index 0000000000000000000000000000000000000000..08d50ab6a3e156307ebc4ccfb0f28e63a7fdfcc1 --- /dev/null +++ b/transcoder/layer_17_width_262k_l0_medium/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_262k_l0_medium_affine/config.json b/transcoder/layer_17_width_262k_l0_medium_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..dfec130065394876bd92ec4554447b835dd73b1a --- /dev/null +++ b/transcoder/layer_17_width_262k_l0_medium_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_262k_l0_small/config.json b/transcoder/layer_17_width_262k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..b6e41fb394c851c52f6b5459e02bec083b2e62d1 --- /dev/null +++ b/transcoder/layer_17_width_262k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_262k_l0_small_affine/config.json b/transcoder/layer_17_width_262k_l0_small_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..aec2da22737a42c86d7cf99b2c74df9b74acfece --- /dev/null +++ b/transcoder/layer_17_width_262k_l0_small_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_65k_l0_big_affine/config.json b/transcoder/layer_17_width_65k_l0_big_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..5e9db2d27f5a3a0e774d0a53b38fa604fac911da --- /dev/null +++ b/transcoder/layer_17_width_65k_l0_big_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 150, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_65k_l0_medium/config.json b/transcoder/layer_17_width_65k_l0_medium/config.json new file mode 100644 index 0000000000000000000000000000000000000000..98cc232cdc8e0035612f0cb85569a2fb0020ac97 --- /dev/null +++ b/transcoder/layer_17_width_65k_l0_medium/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_65k_l0_medium_affine/config.json b/transcoder/layer_17_width_65k_l0_medium_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..fcf9f0004db1a366d0af818add1b712cdfeab71a --- /dev/null +++ b/transcoder/layer_17_width_65k_l0_medium_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_65k_l0_small/config.json b/transcoder/layer_17_width_65k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..201851754da8052d161ea36927cf4f5f8498ab59 --- /dev/null +++ b/transcoder/layer_17_width_65k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_17_width_65k_l0_small_affine/config.json b/transcoder/layer_17_width_65k_l0_small_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..dff0d0701e7437c9411460eab5a4d65ab0408d85 --- /dev/null +++ b/transcoder/layer_17_width_65k_l0_small_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.17.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.17.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_16k_l0_big/config.json b/transcoder/layer_22_width_16k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..561baada3691ee04f585b8f3d461e326a30a96d6 --- /dev/null +++ b/transcoder/layer_22_width_16k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 150, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_16k_l0_medium/config.json b/transcoder/layer_22_width_16k_l0_medium/config.json new file mode 100644 index 0000000000000000000000000000000000000000..435b8852448703b2b21a995c0ef6fc4500163ce8 --- /dev/null +++ b/transcoder/layer_22_width_16k_l0_medium/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_16k_l0_small/config.json b/transcoder/layer_22_width_16k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..5de2bf888e2a459fb0fb50b4602b9e24cccfb3ce --- /dev/null +++ b/transcoder/layer_22_width_16k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_262k_l0_big_affine/config.json b/transcoder/layer_22_width_262k_l0_big_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..268500ced3c725591c14d6ab24248d64067504e0 --- /dev/null +++ b/transcoder/layer_22_width_262k_l0_big_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 150, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_262k_l0_medium/config.json b/transcoder/layer_22_width_262k_l0_medium/config.json new file mode 100644 index 0000000000000000000000000000000000000000..c5e421eeeda3f506b9ed993ff6b00b8d7fecae0e --- /dev/null +++ b/transcoder/layer_22_width_262k_l0_medium/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_262k_l0_medium_affine/config.json b/transcoder/layer_22_width_262k_l0_medium_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..c09df46aecb74f03901d2d4716667fc9aa97078b --- /dev/null +++ b/transcoder/layer_22_width_262k_l0_medium_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_262k_l0_small/config.json b/transcoder/layer_22_width_262k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..ab0239d55834e7ed9b670b868db60579a7f3fc1e --- /dev/null +++ b/transcoder/layer_22_width_262k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_262k_l0_small_affine/config.json b/transcoder/layer_22_width_262k_l0_small_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..54d2d7190f14a39afb4f3179e5c2693e655a7bd8 --- /dev/null +++ b/transcoder/layer_22_width_262k_l0_small_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_65k_l0_big/config.json b/transcoder/layer_22_width_65k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..d0f43815e888684e1656e2ba1ab31542ae713dba --- /dev/null +++ b/transcoder/layer_22_width_65k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 150, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_65k_l0_big_affine/config.json b/transcoder/layer_22_width_65k_l0_big_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..638226c3e498e77695227007dfa9239c49bcfe1c --- /dev/null +++ b/transcoder/layer_22_width_65k_l0_big_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 150, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_65k_l0_medium/config.json b/transcoder/layer_22_width_65k_l0_medium/config.json new file mode 100644 index 0000000000000000000000000000000000000000..c327614a8f17cfb78251ffd413a14e1409d69c33 --- /dev/null +++ b/transcoder/layer_22_width_65k_l0_medium/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_65k_l0_medium_affine/config.json b/transcoder/layer_22_width_65k_l0_medium_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..0b0669a769fa7ee90ab333f12084faec19dfbecf --- /dev/null +++ b/transcoder/layer_22_width_65k_l0_medium_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_65k_l0_small/config.json b/transcoder/layer_22_width_65k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..9c453c171d38162c1aea8e97b55af7d30b81a583 --- /dev/null +++ b/transcoder/layer_22_width_65k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_22_width_65k_l0_small_affine/config.json b/transcoder/layer_22_width_65k_l0_small_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..354fca08d927db98eb84578d16e7a0dd57178821 --- /dev/null +++ b/transcoder/layer_22_width_65k_l0_small_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.22.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.22.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_16k_l0_big/config.json b/transcoder/layer_29_width_16k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..28ae5703fd216bfcc33de260957848058177a68d --- /dev/null +++ b/transcoder/layer_29_width_16k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 150, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_16k_l0_medium/config.json b/transcoder/layer_29_width_16k_l0_medium/config.json new file mode 100644 index 0000000000000000000000000000000000000000..bb6b8b1a8f9e3c0ac9d0253e9d4f00b43b26df70 --- /dev/null +++ b/transcoder/layer_29_width_16k_l0_medium/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_16k_l0_medium_affine/config.json b/transcoder/layer_29_width_16k_l0_medium_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..672af20843f98dfa98702af018e8f2e4fc8a80f2 --- /dev/null +++ b/transcoder/layer_29_width_16k_l0_medium_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_16k_l0_small/config.json b/transcoder/layer_29_width_16k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..f1017f0f5c2aa899c4b03d397a2424cdd8ad27f6 --- /dev/null +++ b/transcoder/layer_29_width_16k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_262k_l0_big_affine/config.json b/transcoder/layer_29_width_262k_l0_big_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..33ece03522e197b431cbaa3dc3ef04d9c8920849 --- /dev/null +++ b/transcoder/layer_29_width_262k_l0_big_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 150, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_262k_l0_medium/config.json b/transcoder/layer_29_width_262k_l0_medium/config.json new file mode 100644 index 0000000000000000000000000000000000000000..e8c5d73fff857b23505b0a3348b81b3e90cba7dd --- /dev/null +++ b/transcoder/layer_29_width_262k_l0_medium/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_262k_l0_medium_affine/config.json b/transcoder/layer_29_width_262k_l0_medium_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..be180a8b3fda07f1a28996c0e68754633e128d65 --- /dev/null +++ b/transcoder/layer_29_width_262k_l0_medium_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_262k_l0_small/config.json b/transcoder/layer_29_width_262k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..0c0743bfbcc209a48c8831ab2b9ffd9028108b4d --- /dev/null +++ b/transcoder/layer_29_width_262k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_65k_l0_big/config.json b/transcoder/layer_29_width_65k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..314b4c0d64c7dd31df433b52cee0bc1fd90fb5bf --- /dev/null +++ b/transcoder/layer_29_width_65k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 150, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_65k_l0_big_affine/config.json b/transcoder/layer_29_width_65k_l0_big_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..fc8bda3888f19c6fa7182c5eb075a0d7835bf1ff --- /dev/null +++ b/transcoder/layer_29_width_65k_l0_big_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 150, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_65k_l0_medium/config.json b/transcoder/layer_29_width_65k_l0_medium/config.json new file mode 100644 index 0000000000000000000000000000000000000000..ea12c70eaf9cf003ed4c27b9e14ce82b1d1446d3 --- /dev/null +++ b/transcoder/layer_29_width_65k_l0_medium/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_65k_l0_medium_affine/config.json b/transcoder/layer_29_width_65k_l0_medium_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..b8a45e87d3ee01829475404fb34cdbc019616626 --- /dev/null +++ b/transcoder/layer_29_width_65k_l0_medium_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 60, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_29_width_65k_l0_small/config.json b/transcoder/layer_29_width_65k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..c8a68eaed45af037439f7b79110e388493d3c2d7 --- /dev/null +++ b/transcoder/layer_29_width_65k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.29.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.29.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 20, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_16k_l0_big/config.json b/transcoder/layer_9_width_16k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..95017bc7bd20d18cf4b9967b832dfcd3c4d8cd3e --- /dev/null +++ b/transcoder/layer_9_width_16k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 131, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_16k_l0_big_affine/config.json b/transcoder/layer_9_width_16k_l0_big_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..9733a1811f979f4108a4dfc5f829820db090dbaf --- /dev/null +++ b/transcoder/layer_9_width_16k_l0_big_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 131, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_16k_l0_medium/config.json b/transcoder/layer_9_width_16k_l0_medium/config.json new file mode 100644 index 0000000000000000000000000000000000000000..1b18d95e8132dfa54873a6db95414e2fd6b2229b --- /dev/null +++ b/transcoder/layer_9_width_16k_l0_medium/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 53, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_16k_l0_medium_affine/config.json b/transcoder/layer_9_width_16k_l0_medium_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..6a17d01356034d40c5a1c060ec3e1e0b5aeb82b8 --- /dev/null +++ b/transcoder/layer_9_width_16k_l0_medium_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 53, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_16k_l0_small/config.json b/transcoder/layer_9_width_16k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..24aff2fd0b6e371b1744645afb0427a730fd0584 --- /dev/null +++ b/transcoder/layer_9_width_16k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 17, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_16k_l0_small_affine/config.json b/transcoder/layer_9_width_16k_l0_small_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..ab2cbb143d0052dfe5b50d0496fd2fb504ad9236 --- /dev/null +++ b/transcoder/layer_9_width_16k_l0_small_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 16384, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 17, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_262k_l0_big/config.json b/transcoder/layer_9_width_262k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..432e10118acf2e2579f970294de73bdf0b30e767 --- /dev/null +++ b/transcoder/layer_9_width_262k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 131, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_262k_l0_big_affine/config.json b/transcoder/layer_9_width_262k_l0_big_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..a1e41992d50a11adb968f295bc8146734c8f092d --- /dev/null +++ b/transcoder/layer_9_width_262k_l0_big_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 131, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_262k_l0_medium_affine/config.json b/transcoder/layer_9_width_262k_l0_medium_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..c8fed89bded243c774d0e5b25acd20a19224aba2 --- /dev/null +++ b/transcoder/layer_9_width_262k_l0_medium_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 53, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_262k_l0_small/config.json b/transcoder/layer_9_width_262k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..ecdeedcdc52ef812cd4e1a8f3a408489e0fcb726 --- /dev/null +++ b/transcoder/layer_9_width_262k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 17, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_262k_l0_small_affine/config.json b/transcoder/layer_9_width_262k_l0_small_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..e5bf92adb47f403f610f986a6094749872224f3f --- /dev/null +++ b/transcoder/layer_9_width_262k_l0_small_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 262144, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 17, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_65k_l0_big/config.json b/transcoder/layer_9_width_65k_l0_big/config.json new file mode 100644 index 0000000000000000000000000000000000000000..ffe36c12f144b09f29df8cbc6973f75fdeaade01 --- /dev/null +++ b/transcoder/layer_9_width_65k_l0_big/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 131, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_65k_l0_big_affine/config.json b/transcoder/layer_9_width_65k_l0_big_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..d0c82d2cbea46991f1853b74aa4176eaf7559a87 --- /dev/null +++ b/transcoder/layer_9_width_65k_l0_big_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 131, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_65k_l0_medium/config.json b/transcoder/layer_9_width_65k_l0_medium/config.json new file mode 100644 index 0000000000000000000000000000000000000000..177a034a7b17845a4b01e9e23f45df76fcbf561a --- /dev/null +++ b/transcoder/layer_9_width_65k_l0_medium/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 53, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_65k_l0_medium_affine/config.json b/transcoder/layer_9_width_65k_l0_medium_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..0ca21a016bf6724a6dc22931afd0e09a624dbcf9 --- /dev/null +++ b/transcoder/layer_9_width_65k_l0_medium_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 53, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_65k_l0_small/config.json b/transcoder/layer_9_width_65k_l0_small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..671b79afde5ed038fc2e6f14b22478cdc0fb65e8 --- /dev/null +++ b/transcoder/layer_9_width_65k_l0_small/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 17, + "affine_connection": false, + "type": "transcoder" +} \ No newline at end of file diff --git a/transcoder/layer_9_width_65k_l0_small_affine/config.json b/transcoder/layer_9_width_65k_l0_small_affine/config.json new file mode 100644 index 0000000000000000000000000000000000000000..177bc299247292c058aee712eef26566992b61c8 --- /dev/null +++ b/transcoder/layer_9_width_65k_l0_small_affine/config.json @@ -0,0 +1,10 @@ +{ + "hf_hook_point_in": "model.layers.9.pre_feedforward_layernorm.output", + "hf_hook_point_out": "model.layers.9.post_feedforward_layernorm.output", + "width": 65536, + "model_name": "google/gemma-3-4b-it", + "architecture": "jump_relu", + "l0": 17, + "affine_connection": true, + "type": "transcoder" +} \ No newline at end of file