diff --git a/.gitattributes b/.gitattributes
index a6344aac8c09253b3b630fb776ae94478aa0275b..e993dca7b2554ac0d56ad8267b153248b4253315 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
*.zip filter=lfs diff=lfs merge=lfs -text
*.zst filter=lfs diff=lfs merge=lfs -text
*tfevents* filter=lfs diff=lfs merge=lfs -text
+benchmark_stats.png filter=lfs diff=lfs merge=lfs -text
diff --git a/benchmark_stats.csv b/benchmark_stats.csv
new file mode 100644
index 0000000000000000000000000000000000000000..d7a7d35fdaee59372e15cfc3711efb67b8ac69a1
--- /dev/null
+++ b/benchmark_stats.csv
@@ -0,0 +1,13 @@
+,version,d_sae,layer,kl_div_with_sae,kl_div_with_ablation,ce_loss_with_sae,ce_loss_without_sae,ce_loss_with_ablation,kl_div_score,ce_loss_score,l2_norm_in,l2_norm_out,l2_ratio,l0,l1,explained_variance,mse,total_tokens_evaluated,filepath
+OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_0/metrics.json,5,32,0,0.0048454091884195805,3.094083309173584,3.6054646968841553,3.599064588546753,6.694648742675781,0.9984339758486613,0.9979325038445924,29.933448791503906,29.601543426513672,0.9893707036972046,32.0,71.21115112304688,0.9667970538139343,21.729291915893555,6144.0,OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_0/metrics.json
+OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_1/metrics.json,5,32,1,0.006601419299840927,0.05105271190404892,3.605595588684082,3.599064588546753,3.6525371074676514,0.8706940522131719,0.8778624933119316,18.973735809326172,17.91716766357422,0.910649299621582,32.0,86.5653305053711,0.8854424357414246,25.637441635131836,6144.0,OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_1/metrics.json
+OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_2/metrics.json,5,32,2,0.00936876516789198,0.05874736234545708,3.601879119873047,3.599064588546753,3.6459126472473145,0.8405244968650671,0.9399221354232932,49.106536865234375,47.64448165893555,0.8887979388237,31.875,85.81163024902344,0.9745470285415649,37.83729553222656,6144.0,OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_2/metrics.json
+OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_3/metrics.json,5,32,3,0.010681239888072014,0.07059153914451599,3.6096014976501465,3.599064588546753,3.658677577972412,0.8486895169376428,0.823244745735597,16.98731803894043,15.157210350036621,0.8746687173843384,31.91145896911621,85.93821716308594,0.7805342078208923,50.548057556152344,6144.0,OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_3/metrics.json
+OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_4/metrics.json,5,32,4,0.012658017687499523,0.06332532316446304,3.611159086227417,3.599064588546753,3.6600804328918457,0.8001112816333331,0.8017810322797447,17.251985549926758,15.012179374694824,0.8525444865226746,31.95556640625,82.4767074584961,0.7294961810112,63.70451354980469,6144.0,OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_4/metrics.json
+OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_5/metrics.json,5,32,5,0.014466611668467522,0.06850520521402359,3.613976240158081,3.599064588546753,3.6693859100341797,0.7888246356861349,0.7879497811486054,18.888967514038086,16.209918975830078,0.8484395742416382,32.0,81.43401336669922,0.7174215316772461,87.28172302246094,6144.0,OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_5/metrics.json
+OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_6/metrics.json,5,32,6,0.016599537804722786,0.07569437474012375,3.6187987327575684,3.599064588546753,3.67651629447937,0.7807031518298047,0.7452071071490137,21.466564178466797,18.40247344970703,0.852634608745575,32.0,78.82976531982422,0.7063077688217163,117.07249450683594,6144.0,OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_6/metrics.json
+OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_7/metrics.json,5,32,7,0.017009764909744263,0.08048636466264725,3.614975929260254,3.599064588546753,3.6727118492126465,0.7886627756013151,0.7839520361024154,25.444438934326172,22.004989624023438,0.8624889254570007,32.0,76.41993713378906,0.7180025577545166,157.79141235351562,6144.0,OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_7/metrics.json
+OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_8/metrics.json,5,32,8,0.01810324750840664,0.08732372522354126,3.6162450313568115,3.599064588546753,3.6803367137908936,0.792688098657451,0.7886059610420089,30.250225067138672,26.306936264038086,0.8676368594169617,32.0,76.72819519042969,0.7239155769348145,219.98291015625,6144.0,OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_8/metrics.json
+OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_9/metrics.json,5,32,9,0.01999707892537117,0.09758877754211426,3.6174559593200684,3.599064588546753,3.6962451934814453,0.7950883346526042,0.8107506041388108,40.192413330078125,35.94580841064453,0.8897998332977295,32.0,72.42656707763672,0.7423521876335144,318.1434326171875,6144.0,OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_9/metrics.json
+OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_10/metrics.json,5,32,10,0.02311515063047409,0.12674781680107117,3.6171722412109375,3.599064588546753,3.708984136581421,0.8176288064452188,0.8352644912761693,81.75682830810547,78.3930892944336,0.9553598165512085,32.0,50.45811462402344,0.7926573753356934,514.5535888671875,6144.0,OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_10/metrics.json
+OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_11/metrics.json,5,32,11,0.028952505439519882,0.17384076118469238,3.62371826171875,3.599064588546753,3.783318281173706,0.8334538732906256,0.8661971284238421,92.90629577636719,87.66377258300781,0.9233807325363159,32.0,73.98703002929688,0.840599000453949,742.95751953125,6144.0,OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_11/metrics.json
diff --git a/benchmark_stats.html b/benchmark_stats.html
new file mode 100644
index 0000000000000000000000000000000000000000..2d16c17e7e3ee01da8e247b83b841ee97f090ebd
--- /dev/null
+++ b/benchmark_stats.html
@@ -0,0 +1,646 @@
+
+
+
+
+ |
+ version |
+ d_sae |
+ layer |
+ kl_div_with_sae |
+ kl_div_with_ablation |
+ ce_loss_with_sae |
+ ce_loss_without_sae |
+ ce_loss_with_ablation |
+ kl_div_score |
+ ce_loss_score |
+ l2_norm_in |
+ l2_norm_out |
+ l2_ratio |
+ l0 |
+ l1 |
+ explained_variance |
+ mse |
+ total_tokens_evaluated |
+ filepath |
+
+
+
+
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_0/metrics.json |
+ 5 |
+ 32 |
+ 0 |
+ 0.004845 |
+ 3.094083 |
+ 3.605465 |
+ 3.599065 |
+ 6.694649 |
+ 0.998434 |
+ 0.997933 |
+ 29.933449 |
+ 29.601543 |
+ 0.989371 |
+ 32.000000 |
+ 71.211151 |
+ 0.966797 |
+ 21.729292 |
+ 6144.000000 |
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_0/metrics.json |
+
+
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_1/metrics.json |
+ 5 |
+ 32 |
+ 1 |
+ 0.006601 |
+ 0.051053 |
+ 3.605596 |
+ 3.599065 |
+ 3.652537 |
+ 0.870694 |
+ 0.877862 |
+ 18.973736 |
+ 17.917168 |
+ 0.910649 |
+ 32.000000 |
+ 86.565331 |
+ 0.885442 |
+ 25.637442 |
+ 6144.000000 |
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_1/metrics.json |
+
+
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_2/metrics.json |
+ 5 |
+ 32 |
+ 2 |
+ 0.009369 |
+ 0.058747 |
+ 3.601879 |
+ 3.599065 |
+ 3.645913 |
+ 0.840524 |
+ 0.939922 |
+ 49.106537 |
+ 47.644482 |
+ 0.888798 |
+ 31.875000 |
+ 85.811630 |
+ 0.974547 |
+ 37.837296 |
+ 6144.000000 |
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_2/metrics.json |
+
+
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_3/metrics.json |
+ 5 |
+ 32 |
+ 3 |
+ 0.010681 |
+ 0.070592 |
+ 3.609601 |
+ 3.599065 |
+ 3.658678 |
+ 0.848690 |
+ 0.823245 |
+ 16.987318 |
+ 15.157210 |
+ 0.874669 |
+ 31.911459 |
+ 85.938217 |
+ 0.780534 |
+ 50.548058 |
+ 6144.000000 |
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_3/metrics.json |
+
+
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_4/metrics.json |
+ 5 |
+ 32 |
+ 4 |
+ 0.012658 |
+ 0.063325 |
+ 3.611159 |
+ 3.599065 |
+ 3.660080 |
+ 0.800111 |
+ 0.801781 |
+ 17.251986 |
+ 15.012179 |
+ 0.852544 |
+ 31.955566 |
+ 82.476707 |
+ 0.729496 |
+ 63.704514 |
+ 6144.000000 |
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_4/metrics.json |
+
+
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_5/metrics.json |
+ 5 |
+ 32 |
+ 5 |
+ 0.014467 |
+ 0.068505 |
+ 3.613976 |
+ 3.599065 |
+ 3.669386 |
+ 0.788825 |
+ 0.787950 |
+ 18.888968 |
+ 16.209919 |
+ 0.848440 |
+ 32.000000 |
+ 81.434013 |
+ 0.717422 |
+ 87.281723 |
+ 6144.000000 |
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_5/metrics.json |
+
+
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_6/metrics.json |
+ 5 |
+ 32 |
+ 6 |
+ 0.016600 |
+ 0.075694 |
+ 3.618799 |
+ 3.599065 |
+ 3.676516 |
+ 0.780703 |
+ 0.745207 |
+ 21.466564 |
+ 18.402473 |
+ 0.852635 |
+ 32.000000 |
+ 78.829765 |
+ 0.706308 |
+ 117.072495 |
+ 6144.000000 |
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_6/metrics.json |
+
+
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_7/metrics.json |
+ 5 |
+ 32 |
+ 7 |
+ 0.017010 |
+ 0.080486 |
+ 3.614976 |
+ 3.599065 |
+ 3.672712 |
+ 0.788663 |
+ 0.783952 |
+ 25.444439 |
+ 22.004990 |
+ 0.862489 |
+ 32.000000 |
+ 76.419937 |
+ 0.718003 |
+ 157.791412 |
+ 6144.000000 |
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_7/metrics.json |
+
+
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_8/metrics.json |
+ 5 |
+ 32 |
+ 8 |
+ 0.018103 |
+ 0.087324 |
+ 3.616245 |
+ 3.599065 |
+ 3.680337 |
+ 0.792688 |
+ 0.788606 |
+ 30.250225 |
+ 26.306936 |
+ 0.867637 |
+ 32.000000 |
+ 76.728195 |
+ 0.723916 |
+ 219.982910 |
+ 6144.000000 |
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_8/metrics.json |
+
+
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_9/metrics.json |
+ 5 |
+ 32 |
+ 9 |
+ 0.019997 |
+ 0.097589 |
+ 3.617456 |
+ 3.599065 |
+ 3.696245 |
+ 0.795088 |
+ 0.810751 |
+ 40.192413 |
+ 35.945808 |
+ 0.889800 |
+ 32.000000 |
+ 72.426567 |
+ 0.742352 |
+ 318.143433 |
+ 6144.000000 |
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_9/metrics.json |
+
+
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_10/metrics.json |
+ 5 |
+ 32 |
+ 10 |
+ 0.023115 |
+ 0.126748 |
+ 3.617172 |
+ 3.599065 |
+ 3.708984 |
+ 0.817629 |
+ 0.835264 |
+ 81.756828 |
+ 78.393089 |
+ 0.955360 |
+ 32.000000 |
+ 50.458115 |
+ 0.792657 |
+ 514.553589 |
+ 6144.000000 |
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_10/metrics.json |
+
+
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_11/metrics.json |
+ 5 |
+ 32 |
+ 11 |
+ 0.028953 |
+ 0.173841 |
+ 3.623718 |
+ 3.599065 |
+ 3.783318 |
+ 0.833454 |
+ 0.866197 |
+ 92.906296 |
+ 87.663773 |
+ 0.923381 |
+ 32.000000 |
+ 73.987030 |
+ 0.840599 |
+ 742.957520 |
+ 6144.000000 |
+ OAI_GPT2Small_v5_32k_resid_delta_mlp/v5_32k_layer_11/metrics.json |
+
+
+
diff --git a/benchmark_stats.png b/benchmark_stats.png
new file mode 100644
index 0000000000000000000000000000000000000000..73a131c1df718fd5296b8542291d1638b5d1d087
--- /dev/null
+++ b/benchmark_stats.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c06d084693334119a81889b712cccee3f900115be574d2420e6304a50e9b2dc
+size 4548552
diff --git a/v5_32k_layer_0/cfg.json b/v5_32k_layer_0/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..381b394dd00b185f70e688a0302fb005e2a701fe
--- /dev/null
+++ b/v5_32k_layer_0/cfg.json
@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.0.hook_mlp_out", "hook_layer": 0, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"}
\ No newline at end of file
diff --git a/v5_32k_layer_0/metrics.json b/v5_32k_layer_0/metrics.json
new file mode 100644
index 0000000000000000000000000000000000000000..92f6694bfa8d9640887fabb0ac315b320ea2759d
--- /dev/null
+++ b/v5_32k_layer_0/metrics.json
@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.0048454091884195805, "metrics/kl_div_with_ablation": 3.094083309173584, "metrics/ce_loss_with_sae": 3.6054646968841553, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 6.694648742675781, "metrics/kl_div_score": 0.9984339758486613, "metrics/ce_loss_score": 0.9979325038445924, "metrics/l2_norm_in": 29.933448791503906, "metrics/l2_norm_out": 29.601543426513672, "metrics/l2_ratio": 0.9893707036972046, "metrics/l0": 32.0, "metrics/l1": 71.21115112304688, "metrics/explained_variance": 0.9667970538139343, "metrics/mse": 21.729291915893555, "metrics/total_tokens_evaluated": 6144}
\ No newline at end of file
diff --git a/v5_32k_layer_0/sae_weights.safetensors b/v5_32k_layer_0/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..79d31456524867db78ade6149d5c9745b7621478
--- /dev/null
+++ b/v5_32k_layer_0/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab7f7cbb78776f97da06167e5297aa3d045a8fc83eda19e91076e4b98ce2934f
+size 201461056
diff --git a/v5_32k_layer_0/sparsity.safetensors b/v5_32k_layer_0/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..39584b16c76f37d41202a86426da961dce018f58
--- /dev/null
+++ b/v5_32k_layer_0/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ead023e8744163c7b1fd85e383df10bf880155bea44ad396ac7182c5a2890d81
+size 131152
diff --git a/v5_32k_layer_1/cfg.json b/v5_32k_layer_1/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..7c8336853af871822a6f78383992846821298141
--- /dev/null
+++ b/v5_32k_layer_1/cfg.json
@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.1.hook_mlp_out", "hook_layer": 1, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"}
\ No newline at end of file
diff --git a/v5_32k_layer_1/metrics.json b/v5_32k_layer_1/metrics.json
new file mode 100644
index 0000000000000000000000000000000000000000..4f5df8ef4260f4ac927ab6219c1bb92c3feb8e8c
--- /dev/null
+++ b/v5_32k_layer_1/metrics.json
@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.006601419299840927, "metrics/kl_div_with_ablation": 0.05105271190404892, "metrics/ce_loss_with_sae": 3.605595588684082, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6525371074676514, "metrics/kl_div_score": 0.8706940522131719, "metrics/ce_loss_score": 0.8778624933119316, "metrics/l2_norm_in": 18.973735809326172, "metrics/l2_norm_out": 17.91716766357422, "metrics/l2_ratio": 0.910649299621582, "metrics/l0": 32.0, "metrics/l1": 86.5653305053711, "metrics/explained_variance": 0.8854424357414246, "metrics/mse": 25.637441635131836, "metrics/total_tokens_evaluated": 6144}
\ No newline at end of file
diff --git a/v5_32k_layer_1/sae_weights.safetensors b/v5_32k_layer_1/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8d800d7d9cfc29d47d26d3e9b18689e6f4a4da1b
--- /dev/null
+++ b/v5_32k_layer_1/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9fe30891436eb0a90f46ebc5850c61ccba4022a2acaca044335f7c740f746dba
+size 201461056
diff --git a/v5_32k_layer_1/sparsity.safetensors b/v5_32k_layer_1/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e4c346e25c019d18d3b4f221245e47505f283adf
--- /dev/null
+++ b/v5_32k_layer_1/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:302f9ced1609a7507d856f4d3e8c31bae60f1db043c13595c527427f6c7e7f02
+size 131152
diff --git a/v5_32k_layer_10/cfg.json b/v5_32k_layer_10/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..8f9006c0717a8b4bdba41f245920ce9a2a6a37c4
--- /dev/null
+++ b/v5_32k_layer_10/cfg.json
@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.10.hook_mlp_out", "hook_layer": 10, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"}
\ No newline at end of file
diff --git a/v5_32k_layer_10/metrics.json b/v5_32k_layer_10/metrics.json
new file mode 100644
index 0000000000000000000000000000000000000000..046e21a4a82fcde83b051121ec69ed15db693a0a
--- /dev/null
+++ b/v5_32k_layer_10/metrics.json
@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.02311515063047409, "metrics/kl_div_with_ablation": 0.12674781680107117, "metrics/ce_loss_with_sae": 3.6171722412109375, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.708984136581421, "metrics/kl_div_score": 0.8176288064452188, "metrics/ce_loss_score": 0.8352644912761693, "metrics/l2_norm_in": 81.75682830810547, "metrics/l2_norm_out": 78.3930892944336, "metrics/l2_ratio": 0.9553598165512085, "metrics/l0": 32.0, "metrics/l1": 50.45811462402344, "metrics/explained_variance": 0.7926573753356934, "metrics/mse": 514.5535888671875, "metrics/total_tokens_evaluated": 6144}
\ No newline at end of file
diff --git a/v5_32k_layer_10/sae_weights.safetensors b/v5_32k_layer_10/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f57b424b0811aae6bae86769432b1baa3e727906
--- /dev/null
+++ b/v5_32k_layer_10/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c676f5187854407e5c6f79d7db14fafe6bf3a662a9883e7fa442effa194d6d04
+size 201461056
diff --git a/v5_32k_layer_10/sparsity.safetensors b/v5_32k_layer_10/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7ff10f5d6fd70f6b8872ef625583f1b775be86b3
--- /dev/null
+++ b/v5_32k_layer_10/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7164ca2c04e1c69c8662b4b0a105ee61196163ffbb1ab4eea3e1875644c25895
+size 131152
diff --git a/v5_32k_layer_11/cfg.json b/v5_32k_layer_11/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..78d7d7f4004f43b2ed9c7ae638645d96747f526b
--- /dev/null
+++ b/v5_32k_layer_11/cfg.json
@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.11.hook_mlp_out", "hook_layer": 11, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"}
\ No newline at end of file
diff --git a/v5_32k_layer_11/metrics.json b/v5_32k_layer_11/metrics.json
new file mode 100644
index 0000000000000000000000000000000000000000..26085daec2e9cf053950140a4abc22c79045e8fe
--- /dev/null
+++ b/v5_32k_layer_11/metrics.json
@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.028952505439519882, "metrics/kl_div_with_ablation": 0.17384076118469238, "metrics/ce_loss_with_sae": 3.62371826171875, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.783318281173706, "metrics/kl_div_score": 0.8334538732906256, "metrics/ce_loss_score": 0.8661971284238421, "metrics/l2_norm_in": 92.90629577636719, "metrics/l2_norm_out": 87.66377258300781, "metrics/l2_ratio": 0.9233807325363159, "metrics/l0": 32.0, "metrics/l1": 73.98703002929688, "metrics/explained_variance": 0.840599000453949, "metrics/mse": 742.95751953125, "metrics/total_tokens_evaluated": 6144}
\ No newline at end of file
diff --git a/v5_32k_layer_11/sae_weights.safetensors b/v5_32k_layer_11/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ff2c31787bd092e92a5d9c5f65abbc5b9ad23768
--- /dev/null
+++ b/v5_32k_layer_11/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72876f75e313f3785999392dddef0afe7b2dc44f6a0671b007dd1dcd176ee8c5
+size 201461056
diff --git a/v5_32k_layer_11/sparsity.safetensors b/v5_32k_layer_11/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8393b81eb029636d966a82fe9ed8fa1a46735019
--- /dev/null
+++ b/v5_32k_layer_11/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26db3acaed898f49efb23762df7d70847bc97c5da288926681567dc89f8107a9
+size 131152
diff --git a/v5_32k_layer_2/cfg.json b/v5_32k_layer_2/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..f86334a54c38510b269e530fdc62935d556e8635
--- /dev/null
+++ b/v5_32k_layer_2/cfg.json
@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.2.hook_mlp_out", "hook_layer": 2, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"}
\ No newline at end of file
diff --git a/v5_32k_layer_2/metrics.json b/v5_32k_layer_2/metrics.json
new file mode 100644
index 0000000000000000000000000000000000000000..c8d5210ec8d29dd1b8a559e4de83b4277224e71e
--- /dev/null
+++ b/v5_32k_layer_2/metrics.json
@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.00936876516789198, "metrics/kl_div_with_ablation": 0.05874736234545708, "metrics/ce_loss_with_sae": 3.601879119873047, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6459126472473145, "metrics/kl_div_score": 0.8405244968650671, "metrics/ce_loss_score": 0.9399221354232932, "metrics/l2_norm_in": 49.106536865234375, "metrics/l2_norm_out": 47.64448165893555, "metrics/l2_ratio": 0.8887979388237, "metrics/l0": 31.875, "metrics/l1": 85.81163024902344, "metrics/explained_variance": 0.9745470285415649, "metrics/mse": 37.83729553222656, "metrics/total_tokens_evaluated": 6144}
\ No newline at end of file
diff --git a/v5_32k_layer_2/sae_weights.safetensors b/v5_32k_layer_2/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..25d0aa4a65ed1710bf0f7447b40d00c18d3c7f4a
--- /dev/null
+++ b/v5_32k_layer_2/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80a546f38807810c98365c13fdf6ba7c883b2f177fec3d79963dd869992f3d9f
+size 201461056
diff --git a/v5_32k_layer_2/sparsity.safetensors b/v5_32k_layer_2/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0222d7530ce7aeb7ec88b3b40c396b4a7fd1e6ec
--- /dev/null
+++ b/v5_32k_layer_2/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f0e1dcea111995b08893a3af42c3de1444f9ac0b12dc28c7df6d88125e51ee8
+size 131152
diff --git a/v5_32k_layer_3/cfg.json b/v5_32k_layer_3/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..40125db1b3e4cb725fb1adf6d71b420babc9899d
--- /dev/null
+++ b/v5_32k_layer_3/cfg.json
@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.3.hook_mlp_out", "hook_layer": 3, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"}
\ No newline at end of file
diff --git a/v5_32k_layer_3/metrics.json b/v5_32k_layer_3/metrics.json
new file mode 100644
index 0000000000000000000000000000000000000000..8466ca82093a5f945bbf3f2456077bec008fe54d
--- /dev/null
+++ b/v5_32k_layer_3/metrics.json
@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.010681239888072014, "metrics/kl_div_with_ablation": 0.07059153914451599, "metrics/ce_loss_with_sae": 3.6096014976501465, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.658677577972412, "metrics/kl_div_score": 0.8486895169376428, "metrics/ce_loss_score": 0.823244745735597, "metrics/l2_norm_in": 16.98731803894043, "metrics/l2_norm_out": 15.157210350036621, "metrics/l2_ratio": 0.8746687173843384, "metrics/l0": 31.91145896911621, "metrics/l1": 85.93821716308594, "metrics/explained_variance": 0.7805342078208923, "metrics/mse": 50.548057556152344, "metrics/total_tokens_evaluated": 6144}
\ No newline at end of file
diff --git a/v5_32k_layer_3/sae_weights.safetensors b/v5_32k_layer_3/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..45cbfe8bfb2785aee2c462d322164eb2627348a9
--- /dev/null
+++ b/v5_32k_layer_3/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb781b7a737037ffe78ab1bdca78ba1af47bb297c189ebd9902c60a3fd5f333f
+size 201461056
diff --git a/v5_32k_layer_3/sparsity.safetensors b/v5_32k_layer_3/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..dd296034ac5b9f7d58e427b9e592c0747228c1a3
--- /dev/null
+++ b/v5_32k_layer_3/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26543a937a617417c20fe430ff78e4e031b1b941d59cc80b4db96b4ef7c784a5
+size 131152
diff --git a/v5_32k_layer_4/cfg.json b/v5_32k_layer_4/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..95cd0ff595aaef2a1d2c882b2b99bd7cd2734063
--- /dev/null
+++ b/v5_32k_layer_4/cfg.json
@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.4.hook_mlp_out", "hook_layer": 4, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"}
\ No newline at end of file
diff --git a/v5_32k_layer_4/metrics.json b/v5_32k_layer_4/metrics.json
new file mode 100644
index 0000000000000000000000000000000000000000..b1ed7359b108fd6fa557f632d1bffd79296ca58f
--- /dev/null
+++ b/v5_32k_layer_4/metrics.json
@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.012658017687499523, "metrics/kl_div_with_ablation": 0.06332532316446304, "metrics/ce_loss_with_sae": 3.611159086227417, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6600804328918457, "metrics/kl_div_score": 0.8001112816333331, "metrics/ce_loss_score": 0.8017810322797447, "metrics/l2_norm_in": 17.251985549926758, "metrics/l2_norm_out": 15.012179374694824, "metrics/l2_ratio": 0.8525444865226746, "metrics/l0": 31.95556640625, "metrics/l1": 82.4767074584961, "metrics/explained_variance": 0.7294961810112, "metrics/mse": 63.70451354980469, "metrics/total_tokens_evaluated": 6144}
\ No newline at end of file
diff --git a/v5_32k_layer_4/sae_weights.safetensors b/v5_32k_layer_4/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..347c977e325f0dce2228fc224417a662a80a0234
--- /dev/null
+++ b/v5_32k_layer_4/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:17f173d1919c9c9f53f3a956c96aebe863b0cc90f964c74b13d0056553a09ab3
+size 201461056
diff --git a/v5_32k_layer_4/sparsity.safetensors b/v5_32k_layer_4/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..37a19d1a5f035d957e89ed08fec67cf09417717a
--- /dev/null
+++ b/v5_32k_layer_4/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc4ef7f40430139c30bc32dbce3fbb64304191924edf791b404eb896583dec26
+size 131152
diff --git a/v5_32k_layer_5/cfg.json b/v5_32k_layer_5/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..f173c114874f719174f0e5f87c5a31ed20ed2739
--- /dev/null
+++ b/v5_32k_layer_5/cfg.json
@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.5.hook_mlp_out", "hook_layer": 5, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"}
\ No newline at end of file
diff --git a/v5_32k_layer_5/metrics.json b/v5_32k_layer_5/metrics.json
new file mode 100644
index 0000000000000000000000000000000000000000..02569cc62dfaec794c48669951a6251ecea9d689
--- /dev/null
+++ b/v5_32k_layer_5/metrics.json
@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.014466611668467522, "metrics/kl_div_with_ablation": 0.06850520521402359, "metrics/ce_loss_with_sae": 3.613976240158081, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6693859100341797, "metrics/kl_div_score": 0.7888246356861349, "metrics/ce_loss_score": 0.7879497811486054, "metrics/l2_norm_in": 18.888967514038086, "metrics/l2_norm_out": 16.209918975830078, "metrics/l2_ratio": 0.8484395742416382, "metrics/l0": 32.0, "metrics/l1": 81.43401336669922, "metrics/explained_variance": 0.7174215316772461, "metrics/mse": 87.28172302246094, "metrics/total_tokens_evaluated": 6144}
\ No newline at end of file
diff --git a/v5_32k_layer_5/sae_weights.safetensors b/v5_32k_layer_5/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5d237d99f80b052399c2435a485ce330b3a8f8bd
--- /dev/null
+++ b/v5_32k_layer_5/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a5dbd05b4cb8acc8150fa92853ab5c42373121b1bc35bc108ffac1c6cc73ad4
+size 201461056
diff --git a/v5_32k_layer_5/sparsity.safetensors b/v5_32k_layer_5/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b62869cdd95d24da5c9a73efad0201e661f1bbb7
--- /dev/null
+++ b/v5_32k_layer_5/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d092024c813ffab3ab07a3d24630235d7d4348f5b4d68b3d737ad926b4d50022
+size 131152
diff --git a/v5_32k_layer_6/cfg.json b/v5_32k_layer_6/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..dfba00bfdc745fb90290ec4f5ec515675e0ad5f8
--- /dev/null
+++ b/v5_32k_layer_6/cfg.json
@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.6.hook_mlp_out", "hook_layer": 6, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"}
\ No newline at end of file
diff --git a/v5_32k_layer_6/metrics.json b/v5_32k_layer_6/metrics.json
new file mode 100644
index 0000000000000000000000000000000000000000..b837e4501dc6c46692293f8d449a0d4ad89b6e3f
--- /dev/null
+++ b/v5_32k_layer_6/metrics.json
@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.016599537804722786, "metrics/kl_div_with_ablation": 0.07569437474012375, "metrics/ce_loss_with_sae": 3.6187987327575684, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.67651629447937, "metrics/kl_div_score": 0.7807031518298047, "metrics/ce_loss_score": 0.7452071071490137, "metrics/l2_norm_in": 21.466564178466797, "metrics/l2_norm_out": 18.40247344970703, "metrics/l2_ratio": 0.852634608745575, "metrics/l0": 32.0, "metrics/l1": 78.82976531982422, "metrics/explained_variance": 0.7063077688217163, "metrics/mse": 117.07249450683594, "metrics/total_tokens_evaluated": 6144}
\ No newline at end of file
diff --git a/v5_32k_layer_6/sae_weights.safetensors b/v5_32k_layer_6/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9a3345273f4d8201f132f74b14d738263f8d2ee0
--- /dev/null
+++ b/v5_32k_layer_6/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3105f1072fa405f49c442269f43c553fbdc3e6526b6bc0445eaa6c5cd683838b
+size 201461056
diff --git a/v5_32k_layer_6/sparsity.safetensors b/v5_32k_layer_6/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..34a246b5100b22b1698b7b7b5da085c9d001d1d7
--- /dev/null
+++ b/v5_32k_layer_6/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71eaa2c4d06116f1c4001fc6c162b791a70bcc72f5abccfe6abb5aa870dfe8d9
+size 131152
diff --git a/v5_32k_layer_7/cfg.json b/v5_32k_layer_7/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..404ffa80285d93eef7a758456c02797464775091
--- /dev/null
+++ b/v5_32k_layer_7/cfg.json
@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.7.hook_mlp_out", "hook_layer": 7, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"}
\ No newline at end of file
diff --git a/v5_32k_layer_7/metrics.json b/v5_32k_layer_7/metrics.json
new file mode 100644
index 0000000000000000000000000000000000000000..45abcf6f579c73f580e331b5efac217c3fce7718
--- /dev/null
+++ b/v5_32k_layer_7/metrics.json
@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.017009764909744263, "metrics/kl_div_with_ablation": 0.08048636466264725, "metrics/ce_loss_with_sae": 3.614975929260254, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6727118492126465, "metrics/kl_div_score": 0.7886627756013151, "metrics/ce_loss_score": 0.7839520361024154, "metrics/l2_norm_in": 25.444438934326172, "metrics/l2_norm_out": 22.004989624023438, "metrics/l2_ratio": 0.8624889254570007, "metrics/l0": 32.0, "metrics/l1": 76.41993713378906, "metrics/explained_variance": 0.7180025577545166, "metrics/mse": 157.79141235351562, "metrics/total_tokens_evaluated": 6144}
\ No newline at end of file
diff --git a/v5_32k_layer_7/sae_weights.safetensors b/v5_32k_layer_7/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f2748c77df1111458872d306d93e64aad006dad9
--- /dev/null
+++ b/v5_32k_layer_7/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:062c7f24dd951d7780e3000d052e82f091564685418cd3c07c61717f23616ee3
+size 201461056
diff --git a/v5_32k_layer_7/sparsity.safetensors b/v5_32k_layer_7/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0e2c93b14b99a9a67f311a4f8de39dddd0faf6ba
--- /dev/null
+++ b/v5_32k_layer_7/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64ac63219db967150e3ba33adb0fdbae4328f4c0b3259fc34f6c8617fba230d9
+size 131152
diff --git a/v5_32k_layer_8/cfg.json b/v5_32k_layer_8/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..195acde32ef0a52f6994166bb226452a434f691d
--- /dev/null
+++ b/v5_32k_layer_8/cfg.json
@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.8.hook_mlp_out", "hook_layer": 8, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"}
\ No newline at end of file
diff --git a/v5_32k_layer_8/metrics.json b/v5_32k_layer_8/metrics.json
new file mode 100644
index 0000000000000000000000000000000000000000..26cec739223c991ad772f439fcf89e76a8009c0e
--- /dev/null
+++ b/v5_32k_layer_8/metrics.json
@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.01810324750840664, "metrics/kl_div_with_ablation": 0.08732372522354126, "metrics/ce_loss_with_sae": 3.6162450313568115, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6803367137908936, "metrics/kl_div_score": 0.792688098657451, "metrics/ce_loss_score": 0.7886059610420089, "metrics/l2_norm_in": 30.250225067138672, "metrics/l2_norm_out": 26.306936264038086, "metrics/l2_ratio": 0.8676368594169617, "metrics/l0": 32.0, "metrics/l1": 76.72819519042969, "metrics/explained_variance": 0.7239155769348145, "metrics/mse": 219.98291015625, "metrics/total_tokens_evaluated": 6144}
\ No newline at end of file
diff --git a/v5_32k_layer_8/sae_weights.safetensors b/v5_32k_layer_8/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c5ef7a3743e48e088c61d36766863cbcfd9fba01
--- /dev/null
+++ b/v5_32k_layer_8/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:33f7d0d26aabda775fb2fe1fb8ae9e416e0a4cc92e12e491750d595294d320f1
+size 201461056
diff --git a/v5_32k_layer_8/sparsity.safetensors b/v5_32k_layer_8/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..79669e4ae2b8ef8cb0222f21228993c316dd18cf
--- /dev/null
+++ b/v5_32k_layer_8/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cdd2f0c4c36e44ec05d021122bb7d7435ef8cfefe737018b58f5150987c84fea
+size 131152
diff --git a/v5_32k_layer_9/cfg.json b/v5_32k_layer_9/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..bafeddd0c44e9b4432e2d8d4939ccc9726e279a9
--- /dev/null
+++ b/v5_32k_layer_9/cfg.json
@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.9.hook_mlp_out", "hook_layer": 9, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"}
\ No newline at end of file
diff --git a/v5_32k_layer_9/metrics.json b/v5_32k_layer_9/metrics.json
new file mode 100644
index 0000000000000000000000000000000000000000..dd800f011003ca1e5c049dfc1955b54ef5e51367
--- /dev/null
+++ b/v5_32k_layer_9/metrics.json
@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.01999707892537117, "metrics/kl_div_with_ablation": 0.09758877754211426, "metrics/ce_loss_with_sae": 3.6174559593200684, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6962451934814453, "metrics/kl_div_score": 0.7950883346526042, "metrics/ce_loss_score": 0.8107506041388108, "metrics/l2_norm_in": 40.192413330078125, "metrics/l2_norm_out": 35.94580841064453, "metrics/l2_ratio": 0.8897998332977295, "metrics/l0": 32.0, "metrics/l1": 72.42656707763672, "metrics/explained_variance": 0.7423521876335144, "metrics/mse": 318.1434326171875, "metrics/total_tokens_evaluated": 6144}
\ No newline at end of file
diff --git a/v5_32k_layer_9/sae_weights.safetensors b/v5_32k_layer_9/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0fa46e33280ca301c9d76279220f41719a366318
--- /dev/null
+++ b/v5_32k_layer_9/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e471a86fd93d6c293aa4c9b10cd43abac13cc50ee33ed636628bfc7e36124ab
+size 201461056
diff --git a/v5_32k_layer_9/sparsity.safetensors b/v5_32k_layer_9/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b72edf4e1f057c0ec2296ebfb7316987452662c7
--- /dev/null
+++ b/v5_32k_layer_9/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f257554da48ef7ede841b79335fe5147d9fefd79359a1dca7a5212750f1a56a6
+size 131152